WO2019229983A1 - Step change machine-learned model switching system, edge device, step change machine-learned model switching method, and program - Google Patents

Step change machine-learned model switching system, edge device, step change machine-learned model switching method, and program Download PDF

Info

Publication number
WO2019229983A1
WO2019229983A1 PCT/JP2018/021215 JP2018021215W WO2019229983A1 WO 2019229983 A1 WO2019229983 A1 WO 2019229983A1 JP 2018021215 W JP2018021215 W JP 2018021215W WO 2019229983 A1 WO2019229983 A1 WO 2019229983A1
Authority
WO
WIPO (PCT)
Prior art keywords
machine
learned model
process change
edge device
cloud
Prior art date
Application number
PCT/JP2018/021215
Other languages
French (fr)
Japanese (ja)
Inventor
将仁 谷口
Original Assignee
株式会社ウフル
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 株式会社ウフル filed Critical 株式会社ウフル
Priority to PCT/JP2018/021215 priority Critical patent/WO2019229983A1/en
Publication of WO2019229983A1 publication Critical patent/WO2019229983A1/en

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N99/00—Subject matter not provided for in other groups of this subclass

Definitions

  • the present invention relates to a technology for switching from a cloud to an edge device installed in a production line to a machine-learned model corresponding to the process change in order to cope with a process change of the production line, and IoT (Internet of Things) Used in the field of
  • Patent Document 1 a technique for updating a model of a machine learning system is provided (Patent Document 1).
  • the server connected via the WAN and gateway with the edge device receives the accuracy of the machine learning model of the edge device from the edge device, determines whether the machine learning model should be updated,
  • Patent Document 2 A technique for providing a model to an edge device and updating the model based on an external command is provided (Patent Document 2).
  • Patent Document 1 In the technique described in Patent Document 1, a machine learning system model is trained, but a machine-learned model is switched from the cloud to an edge device installed in the field in order to respond to a change in use. There is a problem that cannot be done.
  • the technology described in Patent Document 2 updates the machine learning model in order to optimize the accuracy of the machine learning model. However, it corresponds to a change in use for an edge device installed in the field from the cloud. Therefore, there is a problem that the machine-learned model cannot be switched. For example, when an edge device installed in the field performs processing using a machine-learned model corresponding to a change in use, it is necessary to change the machine-learned model corresponding to a change in use of the edge device. . However, it is difficult to store a plurality of machine-learned models having different applications in an edge device with a small storage capacity. Therefore, it cannot respond to the process change of the manufacturing line which must respond to the change of use.
  • the present invention aims to switch from a cloud to a machine-learned model corresponding to a process change for an edge device installed in the production line in order to cope with a process change of the production line.
  • the present invention provides the following solutions.
  • the invention according to the first feature is a process change machine learning that switches from a cloud to a machine learned model corresponding to a process change for an edge device installed in the production line from the cloud in order to cope with a process change of the production line.
  • a grasping means for grasping process change information related to a process change, and causing the edge device to acquire a machine-learned model associated with the grasped process change information from the cloud.
  • the first control means for controlling the machine and the edge device so as to switch from the machine-learned model of another process already applied to the machine-learned model associated with the acquired process change information.
  • a second control means for controlling the process change machine-learned model switching system.
  • the invention according to the first aspect is an edge device that is installed in a production line and is used by switching to a machine-learned model corresponding to the process change in order to cope with the process change of the production line.
  • the machine-learned model associated with the process change information is acquired from the cloud, and the machine-learned model of another process that has already been used, An edge device that switches to a machine-learned model associated with the acquired process change information is provided.
  • the invention according to the first feature is a process change machine learning that switches from a cloud to a machine learned model corresponding to a process change for an edge device installed in the production line from the cloud in order to cope with a process change of the production line.
  • a process change machine-learned model switching method comprising: a second control step for controlling.
  • the invention according to the first feature is a process for switching to a machine-learned model corresponding to a process change from the cloud to an edge device installed in the production line in order to cope with a process change of the production line.
  • a second control step for performing control as described above.
  • the outline figure of a process change machine learning completed model change system A list of machine-learned models associated with process change information. An example of switching to a machine-learned model corresponding to a process change. An example of a process change machine learning completed model switching process.
  • the process change machine-learned model switching system of the present invention switches from a cloud to a machine-learned model corresponding to a process change for an edge device installed on the production line in order to cope with a process change of the production line.
  • System. Used in the field of IoT (Internet of Things).
  • An edge device is a camera, various sensors, a GPU (Graphics Processing Unit), a gateway, or the like that can be connected to a network.
  • FIG. 1 is a schematic diagram of a process change machine-learned model switching system according to a preferred embodiment of the present invention.
  • the process change machine-learned model switching system includes grasping means, first control means, and second control means that are realized by reading a predetermined program. Moreover, although not shown in figure, you may provide an update means and a cooperation means similarly.
  • Each means described above may be realized by a single computer, or may be realized by two or more computers (for example, a server and an edge device).
  • the grasp acquisition means grasps the process change information related to the process change.
  • the process change information can be grasped by linking with the process schedule.
  • the process change information related to the process change may be grasped by analyzing sensor data output from a sensor included in the edge device.
  • the edge device is a camera
  • the image data For example, in the case of a camera, it is possible to grasp a change in a process by capturing an image of an object flowing on a production line and analyzing the image data.
  • it is also possible to grasp a process by imaging a specific object or character indicating the process and analyzing the image data. For example, a paper with a character string for defect detection processing or a box indicating the defect detection processing is poured on the production line, the image is captured with a camera, and the image data is analyzed to change to the defect detection processing step. You can understand that.
  • the first control means controls the edge device to acquire the machine-learned model associated with the grasped process change information from the cloud.
  • the process change information is a defect detection process for an edge device, as shown in FIG. 2, a machine-learned model for defect detection associated with the defect detection process is acquired from the cloud.
  • the first control means may be installed on the edge device as a software agent. Via this software agent, the edge device is controlled so that the machine-learned model associated with the grasped process change information is acquired from the cloud.
  • the first control unit may perform control so that the edge device acquires a machine-learned model using an encrypted communication method in consideration of security. Since machine-learned models are a lump of know-how, security is important.
  • the second control means controls the edge device to switch from the machine-learned model already applied in another process to the machine-learned model associated with the acquired process change information.
  • FIG. 3 is an example of switching to a machine-learned model corresponding to a process change.
  • a machine-learned model for foreign matter detection is applied to the camera.
  • it is necessary to switch from a machine learning model for foreign object detection already applied to the camera to a machine learned model for defect detection. Otherwise, the defect detection process cannot be performed.
  • a machine-learned model for foreign object detection that has already been applied may be deleted because it is overwritten with the machine-learned model for defect detection when it is switched.
  • the second control means may be installed on the edge device as a software agent. Via this software agent, control is performed so that the machine-learned model already applied in another process is switched to the machine-learned model associated with the acquired process change information.
  • the update means updates the machine-learned model to the latest version by machine learning using the learning device as sensor data acquired from the edge device.
  • the image data for foreign object detection is acquired from the camera, and machine learning is performed as learning data using a learning device, thereby further progress of learning.
  • the first control means controls the edge device to acquire the updated latest version of the machine-learned model from the cloud.
  • the second control unit controls the edge device to switch from the machine-learned model already applied in another process to the updated latest version of the machine-learned model that has been acquired.
  • the latest version of the learned model is used during the first foreign object detection process.
  • the machine in another process that has already been applied is controlled. Control is performed to switch to the updated latest version of the machine-learned model obtained from the learned model.
  • the linkage means API linkage with an external AI platform provided by AI vendor.
  • AI vendors often have their own AI platforms. Therefore, if the AI platform publishes API (Application Programming Interface), the linkage means can link API with those AI platforms. That is, connection is possible.
  • the first control unit controls the edge device to acquire the machine-learned model associated with the process change information from the AI platform, and the second control unit controls the edge device. Then, control is performed so as to switch from a machine-learned model that has already been applied to a machine-learned model acquired from an external AI platform.
  • the AI platform is linked with the AI platform by the linkage means, and The edge device is controlled so that a machine-learned model for defect detection is acquired from the AI platform by one control means, and the machine-learned model of another process already applied to the edge device by the second control means To switch to the machine-learned model for defect detection obtained from this AI platform.
  • An edge device is an edge device that is installed in a production line and is used by switching to a machine-learned model corresponding to the process change in order to cope with a process change of the production line.
  • a machine-learned model associated with the process change information is acquired from the cloud, and is associated with the acquired process change information from a machine-learned model of another process that has already been used.
  • the machine-learned model may be acquired using an encrypted communication method.
  • This encrypted communication method refers to a communication method in which data is encrypted for communication.
  • a known encrypted communication method such as SSL (Secure Sockets Layer) or TLS (Transport Layer Security) may be used.
  • SSL Secure Sockets Layer
  • TLS Transport Layer Security
  • a machine-learned model for foreign object detection that has already been applied may be deleted because it is overwritten with the machine-learned model for defect detection when switched. Since the machine-learned model required at that moment is always acquired from the cloud and switched, it is effective even for edge devices with a small storage capacity.
  • an edge device is a camera on a production line, and performs predetermined processing by inputting image data indicating a captured image into a machine-learned model. For example, different processing is performed depending on the machine-learned model associated with the process change information.
  • the edge device is a computer including a processor, a memory, a storage, a communication unit, and an imaging unit, for example. These devices are connected via a bus. As shown in FIG. 4, the storage does not have a storage capacity for storing, for example, four or more types of machine-learned models.
  • the imaging unit captures an image.
  • the imaging unit captures an image by forming an image on the imaging element using, for example, an optical system. This image may be a still image or a moving image.
  • the function of the edge device is realized by the processor performing calculations or controlling communication by the communication unit in cooperation with the client program stored in the memory or storage and the processor that executes the program.
  • the process change machine-learned model switching method of the present invention switches from a cloud to a machine-learned model corresponding to a process change for an edge device installed on the production line in order to cope with a process change of the production line. Is the method.
  • the process change machine-learned model switching method includes a grasping step, a first control step, and a second control step. Moreover, although not shown in figure, you may provide an update step and a cooperation step similarly.
  • Each of the above steps may be realized by a single computer, or may be realized by two or more computers (for example, a server and an edge device).
  • the grasp acquisition step grasps process change information related to the process change.
  • the process change information can be grasped by linking with the process schedule.
  • the process change information related to the process change may be grasped by analyzing sensor data output from a sensor included in the edge device.
  • the edge device is a camera
  • the image data For example, in the case of a camera, it is possible to grasp a change in a process by capturing an image of an object flowing on a production line and analyzing the image data.
  • it is also possible to grasp a process by imaging a specific object or character indicating the process and analyzing the image data. For example, a paper with a character string for defect detection processing or a box indicating the defect detection processing is poured on the production line, the image is captured with a camera, and the image data is analyzed to change to the defect detection processing step. You can understand that.
  • the first control step controls the edge device to acquire the machine-learned model associated with the grasped process change information from the cloud.
  • the process change information is a defect detection process for an edge device, as shown in FIG. 2, a machine-learned model for defect detection associated with the defect detection process is acquired from the cloud.
  • the first control step may be installed on the edge device as a software agent. Via this software agent, the edge device is controlled so that the machine-learned model associated with the grasped process change information is acquired from the cloud.
  • the edge device may be controlled to acquire a machine-learned model using an encrypted communication method in consideration of security. Since machine-learned models are a lump of know-how, security is important.
  • FIG. 3 is an example of switching to a machine-learned model corresponding to a process change.
  • a machine-learned model for foreign matter detection is applied to the camera.
  • it is necessary to switch from a machine learning model for foreign object detection already applied to the camera to a machine learned model for defect detection. Otherwise, the defect detection process cannot be performed.
  • a machine-learned model for foreign object detection that has already been applied may be deleted because it is overwritten with the machine-learned model for defect detection when it is switched.
  • the second control step may be installed on the edge device as a software agent. Via this software agent, control is performed so that the machine-learned model already applied in another process is switched to the machine-learned model associated with the acquired process change information.
  • the machine learning model is updated to the latest version by machine learning using the learning device as sensor data obtained from the edge device.
  • the image data for foreign object detection is acquired from the camera, and machine learning is performed as learning data using a learning device, thereby further progress of learning.
  • the edge device is controlled to acquire the updated latest version of the machine-learned model from the cloud.
  • the edge device is controlled to switch from the machine-learned model of another process that has already been applied to the acquired updated latest version of the machine-learned model.
  • the latest version of the learned model is used during the first foreign object detection process.
  • the machine in another process that has already been applied is controlled. Control is performed to switch to the updated latest version of the machine-learned model obtained from the learned model.
  • the linkage step is API linkage with an external AI platform provided by an AI vendor.
  • AI vendors often have their own AI platforms. Therefore, if the AI platform publishes API (Application Programming ⁇ ⁇ Interface), the linkage step can be API-linked with those AI platforms. That is, connection is possible.
  • the first control step controls the edge device to acquire the machine-learned model associated with the process change information from the AI platform, and the second control step controls the edge device. Then, control is performed so as to switch from a machine-learned model that has already been applied to a machine-learned model acquired from an external AI platform.
  • the AI platform is linked with the AI platform through the linkage step, Control is performed so that a machine-learned model for defect detection is acquired from this AI platform by one control step, and this AI is obtained from a machine-learned model of another process that has already been applied to the edge device by the second control step. Control to switch to the machine-learned model for defect detection acquired from the platform.
  • the edge device in order to cope with the process change of the production line, it is possible to switch from the cloud to the machine learned model corresponding to the process change for the edge device. Further, since the edge device does not have to have a storage capacity for storing a plurality of machine-learned models, hardware resources required for the edge device can be reduced. In addition, high processing capability is required to update a machine-learned model. However, in the above-described embodiment, since an AI vendor updates a new machine-learned model, the edge device has high processing capability. It does not have to be.
  • the machine-learned model is grasped using the encrypted communication method, it is possible to prevent the machine-learned model from being grasped by a malicious third party.
  • the edge device is not limited to a camera.
  • the edge device may be any device as long as the device performs processing using the machine-learned model.
  • the edge device may include a sensor, and sensor data output from the sensor may be used as learning data.
  • the number of machine-learned models that the edge device switches is not limited to two.
  • the edge device may switch between three or more machine-learned models.
  • the machine learning algorithm generates a model based on the result of machine learning from given data, for example, on a computer, and inputs new input data to the model.
  • a so-called supervised learning algorithm for outputting an event predicted from the input data may be used.
  • the machine learning algorithm is not limited to a so-called supervised learning algorithm, and may be an algorithm for machine learning such as unsupervised learning, semi-supervised learning, reinforcement learning, and expression learning.
  • the algorithm for machine learning may include other learning algorithms such as data mining and deep learning. Note that these learning algorithms use various techniques or techniques such as decision tree learning, correlation rule learning, neural network, genetic programming, inductive logic programming, support vector machine, clustering, and Bayesian network. included. In short, any machine learning algorithm may be used as long as it is processed together with some data provided by the data provider and outputs information desired by the user as a result of the processing.
  • process steps performed in the process change machine-learned model switching method are not limited to the example described in the above embodiment.
  • the steps of this process may be interchanged as long as there is no contradiction.
  • the present invention may be provided as a process change machine learned model switching method performed in a process change machine learned model switching system.
  • the present invention may be provided as a program executed in a cloud or an edge device.
  • This program may be downloaded via a communication line such as the Internet, or a computer such as a magnetic recording medium (magnetic tape, magnetic disk, etc.), an optical recording medium (optical disk, etc.), a magneto-optical recording medium, or a semiconductor memory. May be provided in a state of being recorded on a readable recording medium.
  • the means and functions described above are realized by a computer (including a CPU, an information processing apparatus, and various terminals) reading and executing a predetermined program.
  • the program may be, for example, an application installed on a computer, or may be in the form of SaaS (software as a service) provided from a computer via a network, for example, a flexible disk, a CD It may be provided in a form recorded on a computer-readable recording medium such as a CD-ROM (DVD-ROM, DVD-RAM, etc.).
  • the computer reads the program from the recording medium, transfers it to the internal storage device or the external storage device, stores it, and executes it.
  • the program may be recorded in advance in a storage device (recording medium) such as a magnetic disk, an optical disk, or a magneto-optical disk, and provided from the storage device to a computer via a communication line.
  • a nearest neighbor method a naive Bayes method, a decision tree, a support vector machine, reinforcement learning, or the like may be used.
  • deep learning may be used in which a characteristic amount for learning is generated by using a neural network.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • General Factory Administration (AREA)

Abstract

[Problem] To switch, for an edge device, to a machine-learned model that supports a step change from the cloud in order to respond to a step change in a manufacturing line. [Solution] A step change machine-learned model switching system that, in order to respond to a step change in a manufacturing line, performs switching from the cloud on an edge device installed in the manufacturing line to a machine-learned model that corresponds to the step change, wherein the switching system is provided with: an ascertaining means for ascertaining step change information pertaining to a step change; a first control means for performing a control on the edge device so as to acquire, from the cloud, a machine-learned model associated with the ascertained step change information; and a second control means for performing a control on the edge device so as to switch from an already adapted machine-learned model of another step to a machine-learned model associated with the acquired step change information.

Description

工程変更機械学習済みモデル切り替えシステム、エッジデバイス、工程変更機械学習済みモデル切り替え方法、及びプログラムProcess change machine learned model switching system, edge device, process change machine learned model switching method, and program
　本発明は、製造ラインの工程変更に対応するために、クラウドから当該製造ラインに設置されたエッジデバイスに対して、工程変更に対応する機械学習済みモデルに切り替える技術に関し、ＩｏＴ（Internet of Things）の分野で利用される。 The present invention relates to a technology for switching from a cloud to an edge device installed in a production line to a machine-learned model corresponding to the process change in order to cope with a process change of the production line, and IoT (Internet of Things) Used in the field of
　近年、ＡＩ（Artificial Intelligence）技術が進歩しているが、エッジデバイスから取得したセンサデータをクラウドに垂れ流してクラウドで分析処理をさせると、通信コストやストレージコストの増大になる問題がある。そこで、エッジデバイスで分析処理をさせるために、エッジデバイスに対する学習済みモデルの更新技術が注目されている。例えば、機械学習システムのモデルを更新するための技術が提供されている（特許文献１）。また、エッジデバイスとＷＡＮおよびゲートウェイを介して接続されたサーバが、エッジデバイスの機械学習モデルの精度をエッジデバイスから受信し、機械学習モデルを更新すべきかどうかを決定し、更新された機械学習済みモデルをエッジデバイスに提供し、外部命令に基づいて更新する技術が提供されている（特許文献２）。 In recent years, AI (Artificial Intelligence) technology has progressed, but if sensor data acquired from an edge device is dropped into the cloud and analyzed in the cloud, there is a problem that communication costs and storage costs increase. Therefore, in order to perform analysis processing by the edge device, attention is paid to a technique for updating a learned model for the edge device. For example, a technique for updating a model of a machine learning system is provided (Patent Document 1). In addition, the server connected via the WAN and gateway with the edge device receives the accuracy of the machine learning model of the edge device from the edge device, determines whether the machine learning model should be updated, A technique for providing a model to an edge device and updating the model based on an external command is provided (Patent Document 2).
特開２０１７－１２０６４７JP 2017-120647 A ＷＯ２０１６／１１８８１５WO2016 / 118815
　特許文献１に記載の技術には、機械学習システムのモデルをトレーニングしているが、クラウドから現場に設置されたエッジデバイスに対して、用途の変更に対応するために、機械学習済みモデルを切り替えることが出来ない問題がある。特許文献２に記載の技術には、機械学習モデルの精度を最適化するために機械学習モデルを更新しているが、クラウドから現場に設置されたエッジデバイスに対して、用途の変更に対応するために、機械学習済みモデルを切り替えることが出来ない問題がある。例えば、現場に設置されたエッジデバイスが、用途の変更に対応する機械学習済みモデルを用いて処理を行う場合には、エッジデバイスの用途の変更に対応する機械学習済みモデルを変更する必要がある。しかし、記憶容量の少ないエッジデバイスでは、用途の異なる機械学習済みモデルを複数記憶しておくのが難しい。そのため、用途の変更に対応しなければならない製造ラインの工程変更には対応できない。 In the technique described in Patent Document 1, a machine learning system model is trained, but a machine-learned model is switched from the cloud to an edge device installed in the field in order to respond to a change in use. There is a problem that cannot be done. The technology described in Patent Document 2 updates the machine learning model in order to optimize the accuracy of the machine learning model. However, it corresponds to a change in use for an edge device installed in the field from the cloud. Therefore, there is a problem that the machine-learned model cannot be switched. For example, when an edge device installed in the field performs processing using a machine-learned model corresponding to a change in use, it is necessary to change the machine-learned model corresponding to a change in use of the edge device. . However, it is difficult to store a plurality of machine-learned models having different applications in an edge device with a small storage capacity. Therefore, it cannot respond to the process change of the manufacturing line which must respond to the change of use.
　本発明は、製造ラインの工程変更に対応するために、クラウドから当該製造ラインに設置されたエッジデバイスに対して、工程変更に対応する機械学習済みモデルに切り替えることを目的とする。 The present invention aims to switch from a cloud to a machine-learned model corresponding to a process change for an edge device installed in the production line in order to cope with a process change of the production line.
　本発明では、以下のような解決手段を提供する。 The present invention provides the following solutions.
　第１の特徴に係る発明は、製造ラインの工程変更に対応するために、クラウドから当該製造ラインに設置されたエッジデバイスに対して、工程変更に対応する機械学習済みモデルに切り替える工程変更機械学習済みモデル切り替えシステムであって、工程変更に関する工程変更情報を把握する把握手段と、前記エッジデバイスに対して、把握された前記工程変更情報に対応付けられた機械学習済みモデルをクラウドから取得させるように制御する第１制御手段と、前記エッジデバイスに対して、既に適用されている別工程の機械学習済みモデルから、取得された前記工程変更情報に対応付けられた機械学習済みモデルに切り替えるように制御する第２制御手段と、を備える工程変更機械学習済みモデル切り替えシステムを提供する。 The invention according to the first feature is a process change machine learning that switches from a cloud to a machine learned model corresponding to a process change for an edge device installed in the production line from the cloud in order to cope with a process change of the production line. And a grasping means for grasping process change information related to a process change, and causing the edge device to acquire a machine-learned model associated with the grasped process change information from the cloud. The first control means for controlling the machine and the edge device so as to switch from the machine-learned model of another process already applied to the machine-learned model associated with the acquired process change information. And a second control means for controlling the process change machine-learned model switching system.
　第１の特徴に係る発明は、製造ラインに設置されており、当該製造ラインの工程変更に対応するために、工程変更に対応する機械学習済みモデルに切り替えて使用するエッジデバイスであって、請求項１に記載の工程変更機械学習済みモデル切り替えシステムの制御に従って、クラウドから前記工程変更情報に対応付けられた機械学習済みモデルを取得し、既に使用している別工程の機械学習済みモデルから、前記取得された工程変更情報に対応付けられた機械学習済みモデルに切り替えるエッジデバイスを提供する。 The invention according to the first aspect is an edge device that is installed in a production line and is used by switching to a machine-learned model corresponding to the process change in order to cope with the process change of the production line. In accordance with the control of the process change machine-learned model switching system according to item 1, the machine-learned model associated with the process change information is acquired from the cloud, and the machine-learned model of another process that has already been used, An edge device that switches to a machine-learned model associated with the acquired process change information is provided.
　第１の特徴に係る発明は、製造ラインの工程変更に対応するために、クラウドから当該製造ラインに設置されたエッジデバイスに対して、工程変更に対応する機械学習済みモデルに切り替える工程変更機械学習済みモデル切り替え方法であって、工程変更に関する工程変更情報を把握する把握ステップと、前記エッジデバイスに対して、把握された前記工程変更情報に対応付けられた機械学習済みモデルをクラウドから取得させるように制御する第１制御ステップと、前記エッジデバイスに対して、既に適用されている別工程の機械学習済みモデルから、取得された前記工程変更情報に対応付けられた機械学習済みモデルに切り替えるように制御する第２制御ステップと、を備える工程変更機械学習済みモデル切り替え方法を提供する。 The invention according to the first feature is a process change machine learning that switches from a cloud to a machine learned model corresponding to a process change for an edge device installed in the production line from the cloud in order to cope with a process change of the production line. A grasping step for grasping process change information related to a process change, and causing the edge device to acquire a machine-learned model associated with the grasped process change information from the cloud. A first control step of controlling the edge device to switch from a machine-learned model of another process already applied to the edge device to a machine-learned model associated with the acquired process change information A process change machine-learned model switching method comprising: a second control step for controlling.
　第１の特徴に係る発明は、コンピュータに、製造ラインの工程変更に対応するために、クラウドから当該製造ラインに設置されたエッジデバイスに対して、工程変更に対応する機械学習済みモデルに切り替える処理を実行させるためのプログラムであって、工程変更に関する工程変更情報を把握する把握ステップと、前記エッジデバイスに対して、把握された前記工程変更情報に対応付けられた機械学習済みモデルをクラウドから取得させるように制御する第１制御ステップと、前記エッジデバイスに対して、既に適用されている別工程の機械学習済みモデルから、取得された前記工程変更情報に対応付けられた機械学習済みモデルに切り替えるように制御する第２制御ステップと、を実行させるためのプログラムを提供する。 The invention according to the first feature is a process for switching to a machine-learned model corresponding to a process change from the cloud to an edge device installed in the production line in order to cope with a process change of the production line. A step for grasping process change information relating to a process change, and obtaining a machine-learned model associated with the grasped process change information from the cloud for the edge device. A first control step for controlling to perform, and switching from a machine-learned model of another process already applied to the edge device to a machine-learned model associated with the acquired process change information And a second control step for performing control as described above.
　製造ラインの工程変更に対応するために、クラウドから当該製造ラインに設置されたエッジデバイスに対して、工程変更に対応する機械学習済みモデルに切り替えることができる。 In order to cope with a process change of a production line, it is possible to switch from a cloud to a machine-learned model corresponding to a process change for an edge device installed on the production line.
工程変更機械学習済みモデル切り替えシステムの概要図。The outline figure of a process change machine learning completed model change system. 工程変更情報に対応付けられた機械学習済みモデルの一覧。A list of machine-learned models associated with process change information. 工程変更に対応する機械学習済みモデルに切り替える一例。An example of switching to a machine-learned model corresponding to a process change. 工程変更機械学習済みモデル切り替え処理の一例。An example of a process change machine learning completed model switching process.
　以下、本発明を実施するための最良の形態について説明する。なお、これはあくまでも一例であって、本発明の技術的範囲はこれに限られるものではない。 Hereinafter, the best mode for carrying out the present invention will be described. This is merely an example, and the technical scope of the present invention is not limited to this.
[システム構成]
　本発明の工程変更機械学習済みモデル切り替えシステムは、製造ラインの工程変更に対応するために、クラウドから当該製造ラインに設置されたエッジデバイスに対して、工程変更に対応する機械学習済みモデルに切り替えるシステムである。ＩｏＴ（Internet of Things）の分野で利用される。
エッジデバイスとは、ネットワークに接続可能なカメラ、各種センサ、GPU（Graphics Processing Unit）およびゲートウェイなどである。 [System configuration]
The process change machine-learned model switching system of the present invention switches from a cloud to a machine-learned model corresponding to a process change for an edge device installed on the production line in order to cope with a process change of the production line. System. Used in the field of IoT (Internet of Things).
An edge device is a camera, various sensors, a GPU (Graphics Processing Unit), a gateway, or the like that can be connected to a network.
本発明の好適な実施形態の概要について、図１に基づいて説明する。図１は、本発明の好適な実施形態である工程変更機械学習済みモデル切り替えシステムの概要図である。 An outline of a preferred embodiment of the present invention will be described with reference to FIG. FIG. 1 is a schematic diagram of a process change machine-learned model switching system according to a preferred embodiment of the present invention.
　図１にあるように、工程変更機械学習済みモデル切り替えシステムは、所定のプログラムを読み込むことで実現される、把握手段、第１制御手段、第２制御手段、を備える。また図示しないが、同様に、更新手段、連携手段、を備えてもよい。上述の各手段が、単独のコンピュータで実現されてもよいし、２台以上のコンピュータ（例えば、サーバとエッジデバイスのような場合）で実現されてもよい。 As shown in FIG. 1, the process change machine-learned model switching system includes grasping means, first control means, and second control means that are realized by reading a predetermined program. Moreover, although not shown in figure, you may provide an update means and a cooperation means similarly. Each means described above may be realized by a single computer, or may be realized by two or more computers (for example, a server and an edge device).
　把握取得手段は、工程変更に関する工程変更情報を把握する。例えば、工程スケジュールと連携することで工程変更情報を把握することができる。また、エッジデバイスが有するセンサから出力されるセンサデータを解析することで、工程変更に関する工程変更情報を把握してもよい。例えば、エッジデバイスがカメラである場合、画像データを解析することで、異物検知処理の工程から欠損検知処理の工程に変更になったことを把握することができる。例えば、カメラであれば、製造ラインに流れてきている物体を撮像して、画像データを解析することで工程の変更を把握することができる。また、カメラであれば、工程を示す特定の物や文字などを撮像して、画像データを解析することで工程を把握することもできる。例えば、製造ラインに、欠損検知処理の文字列が記載された紙や、欠損検知処理を示す箱を流し、それをカメラで撮像して、画像データを解析することで欠損検知処理の工程に変更になったことを把握できる。 The grasp acquisition means grasps the process change information related to the process change. For example, the process change information can be grasped by linking with the process schedule. Further, the process change information related to the process change may be grasped by analyzing sensor data output from a sensor included in the edge device. For example, when the edge device is a camera, it is possible to grasp that the foreign object detection process has been changed to the defect detection process by analyzing the image data. For example, in the case of a camera, it is possible to grasp a change in a process by capturing an image of an object flowing on a production line and analyzing the image data. In the case of a camera, it is also possible to grasp a process by imaging a specific object or character indicating the process and analyzing the image data. For example, a paper with a character string for defect detection processing or a box indicating the defect detection processing is poured on the production line, the image is captured with a camera, and the image data is analyzed to change to the defect detection processing step. You can understand that.
　第１制御手段は、エッジデバイスに対して、把握された工程変更情報に対応付けられた機械学習済みモデルをクラウドから取得させるように制御する。例えば、エッジデバイスに対して、工程変更情報が欠損検知処理であった場合には、図２にあるように、欠損検知処理に対応付けられた欠損検知用の機械学習済みモデルをクラウドから取得させるように制御する。図２にあるように、欠損検知処理以外にも対応付けられた様々な機械学習済みモデルがある。特に第１制御手段は、ソフトウェアエージェントとしてエッジデバイスにインストールされることがある。このソフトウェアエージェントを介して、把握された工程変更情報に対応付けられた機械学習済みモデルをクラウドから取得させるようにエッジデバイスを制御する。 The first control means controls the edge device to acquire the machine-learned model associated with the grasped process change information from the cloud. For example, when the process change information is a defect detection process for an edge device, as shown in FIG. 2, a machine-learned model for defect detection associated with the defect detection process is acquired from the cloud. To control. As shown in FIG. 2, there are various machine-learned models associated with the defect detection process. In particular, the first control means may be installed on the edge device as a software agent. Via this software agent, the edge device is controlled so that the machine-learned model associated with the grasped process change information is acquired from the cloud.
また、第１制御手段は、セキュリティを考慮して暗号化通信方式を用いて、エッジデバイスに対して機械学習済みモデルを取得させるように制御してもよい。機械学習済みモデルはノウハウの塊であるので、セキュリティは重要となる。 Further, the first control unit may perform control so that the edge device acquires a machine-learned model using an encrypted communication method in consideration of security. Since machine-learned models are a lump of know-how, security is important.
　第２制御手段は、エッジデバイスに対して、既に適用されている別工程の機械学習済みモデルから、取得された工程変更情報に対応付けられた機械学習済みモデルに切り替えるように制御する。図３は、工程変更に対応する機械学習済みモデルに切り替える一例である。図３にあるように、既に異物検知処理の工程が行われている場合はカメラに対して異物検知用の機械学習済みモデルが適用されている。これを欠損検知処理の工程に工程変更するためには、カメラに対して、既に適用されている異物検知用の機械学習済みモデルから、欠損検知用の機械学習済みモデルに切り替えなければならない。そうしなければ、欠損検知処理を行えない。例えば、既に適用されている異物検知用の機械学習済みモデルは、切り替えられる際に、欠損検知用の機械学習済みモデルに上書きされるため、消去されてもよい。常にクラウドから、その瞬間に必要な機械学習済みモデルを取得して切り替えるため、記憶容量の少ないカメラに対しても効果的である。特に第２制御手段は、ソフトウェアエージェントとしてエッジデバイスにインストールされることがある。このソフトウェアエージェントを介して、既に適用されている別工程の機械学習済みモデルから、取得された工程変更情報に対応付けられた機械学習済みモデルに切り替えるように制御する。 The second control means controls the edge device to switch from the machine-learned model already applied in another process to the machine-learned model associated with the acquired process change information. FIG. 3 is an example of switching to a machine-learned model corresponding to a process change. As shown in FIG. 3, when a foreign matter detection process is already performed, a machine-learned model for foreign matter detection is applied to the camera. In order to change the process to the defect detection process, it is necessary to switch from a machine learning model for foreign object detection already applied to the camera to a machine learned model for defect detection. Otherwise, the defect detection process cannot be performed. For example, a machine-learned model for foreign object detection that has already been applied may be deleted because it is overwritten with the machine-learned model for defect detection when it is switched. Since the machine-learned model necessary for the moment is always acquired from the cloud and switched, it is effective for a camera with a small storage capacity. In particular, the second control means may be installed on the edge device as a software agent. Via this software agent, control is performed so that the machine-learned model already applied in another process is switched to the machine-learned model associated with the acquired process change information.
更新手段は、学習器を用いて、エッジデバイスから取得されたセンサデータを学習データとして機械学習することにより、機械学習済みモデルを最新バージョンに更新する。図３のように、既に異物検知処理の工程が行われている場合は、カメラから異物検知用の画像データを取得して学習データとして学習器を用いて機械学習することで、更に学習の進んだ最新の異物検知用の機械学習モデルを生成して更新することが可能となる。そして、第１制御手段は、エッジデバイスに対して、更新された最新バージョンの機械学習済みモデルをクラウドから取得させるように制御する。そして、第２制御手段は、エッジデバイスに対して、既に適用されている別工程の機械学習済みモデルから、取得された前記更新された最新バージョンの機械学習済みモデルに切り替えるように制御する。例えば、異物検知処置→欠損検知処理→パッケージ不良検知処理→サイズ不良検知処理→異物検知処理の順番で工程変更になる場合、1回目の異物検知処理の工程の際に最新バージョンの学習済みモデルになるように更新しておき、工程変更を経て2回目の異物検知処理に変更されるタイミングで、最新バージョンの機械学習済みモデルを取得するように制御して、既に適用されている別工程の機械学習済みモデルから取得された前記更新された最新バージョンの機械学習済みモデルに切り替えるように制御する。 The update means updates the machine-learned model to the latest version by machine learning using the learning device as sensor data acquired from the edge device. As shown in FIG. 3, when the foreign object detection process has already been performed, the image data for foreign object detection is acquired from the camera, and machine learning is performed as learning data using a learning device, thereby further progress of learning. However, it is possible to generate and update the latest machine learning model for foreign object detection. Then, the first control means controls the edge device to acquire the updated latest version of the machine-learned model from the cloud. Then, the second control unit controls the edge device to switch from the machine-learned model already applied in another process to the updated latest version of the machine-learned model that has been acquired. For example, if the process changes in the order of foreign object detection treatment → defect detection process → package defect detection process → size defect detection process → foreign object detection process, the latest version of the learned model is used during the first foreign object detection process. In order to obtain the latest version of the machine-learned model at the timing when the process is changed and changed to the second foreign object detection process, the machine in another process that has already been applied is controlled. Control is performed to switch to the updated latest version of the machine-learned model obtained from the learned model.
連携手段は、AIベンダーが提供している外部のAIプラットフォームとAPI連携する。例えば、AIベンダーはそれぞれ自社のAIプラットフォームを持っていることが多い。そこで、AIプラットフォームがAPI（Application Programming Interface）を公開していれば、連携手段はそれらのAIプラットフォームとAPI連携することができる。つまり、接続可能となる。そして、第１制御手段は、エッジデバイスに対して、工程変更情報に対応付けられた機械学習済みモデルを、このAIプラットフォームから取得させるように制御し、第２制御手段は、エッジデバイスに対して、既に適用されている別工程の機械学習済みモデルから、外部のAIプラットフォームから取得された機械学習済みモデルに切り替えるように制御する。つまり、AIベンダーが自社のAIプラットフォームに欠損検知用の機械学習済みモデルを登録していれば、工程変更情報は欠損検知処理である場合に、連携手段によってそのAIプラットフォームとAPI連携して、第１制御手段によってこのAIプラットフォームから欠損検知用の機械学習済みモデルを取得するようにエッジデバイスを制御して、第２制御手段によってエッジデバイスに対して既に適用されている別工程の機械学習済みモデルからこのAIプラットフォームから取得された欠損検知用の機械学習済みモデルに切り替えるように制御する。 The linkage means API linkage with an external AI platform provided by AI vendor. For example, AI vendors often have their own AI platforms. Therefore, if the AI platform publishes API (Application Programming Interface), the linkage means can link API with those AI platforms. That is, connection is possible. Then, the first control unit controls the edge device to acquire the machine-learned model associated with the process change information from the AI platform, and the second control unit controls the edge device. Then, control is performed so as to switch from a machine-learned model that has already been applied to a machine-learned model acquired from an external AI platform. In other words, if the AI vendor has registered a machine-learned model for defect detection on their AI platform, and the process change information is a defect detection process, the AI platform is linked with the AI platform by the linkage means, and The edge device is controlled so that a machine-learned model for defect detection is acquired from the AI platform by one control means, and the machine-learned model of another process already applied to the edge device by the second control means To switch to the machine-learned model for defect detection obtained from this AI platform.
[エッジデバイスの説明]
エッジデバイスは、製造ラインに設置されており、当該製造ラインの工程変更に対応するために、工程変更に対応する機械学習済みモデルに切り替えて使用するエッジデバイスであって、工程変更機械学習済みモデル切り替えシステムの制御に従って、クラウドから前記工程変更情報に対応付けられた機械学習済みモデルを取得し、既に使用している別工程の機械学習済みモデルから、前記取得された工程変更情報に対応付けられた機械学習済みモデルに切り替える。このとき、機械学習済みモデルは、暗号化通信方式を用いて取得されてもよい。この暗号化通信方式とは、データを暗号化して通信する通信方式をいう。暗号化通信方式には、ＳＳＬ（Secure Sockets Layer）やＴＬＳ（Transport Layer Security）等の周知の暗号化通信方式が用いられてもよい。例えば、既に適用されている異物検知用の機械学習済みモデルは、切り替えられる際に、欠損検知用の機械学習済みモデルに上書きされるため、消去されてもよい。常にクラウドから、その瞬間に必要な機械学習済みモデルを取得して切り替えるため、記憶容量の少ないエッジデバイスに対しても効果的である。 [Description of edge device]
An edge device is an edge device that is installed in a production line and is used by switching to a machine-learned model corresponding to the process change in order to cope with a process change of the production line. According to the control of the switching system, a machine-learned model associated with the process change information is acquired from the cloud, and is associated with the acquired process change information from a machine-learned model of another process that has already been used. Switch to a machine-learned model. At this time, the machine-learned model may be acquired using an encrypted communication method. This encrypted communication method refers to a communication method in which data is encrypted for communication. As the encrypted communication method, a known encrypted communication method such as SSL (Secure Sockets Layer) or TLS (Transport Layer Security) may be used. For example, a machine-learned model for foreign object detection that has already been applied may be deleted because it is overwritten with the machine-learned model for defect detection when switched. Since the machine-learned model required at that moment is always acquired from the cloud and switched, it is effective even for edge devices with a small storage capacity.
例えば、エッジデバイスとは、製造ラインのカメラであり、撮影された画像を示す画像データを機械学習済みモデルに入力することにより、所定の処理を行う。例えば、工程変更情報に対応付けられた機械学習済みモデルによって異なる処理を行う。エッジデバイスは、例えばプロセッサ、メモリ、ストレージ、通信部、撮像部を備えるコンピュータである。これらの装置は、バスを介して接続されている。ストレージは、図４にあるように、例えば４種類以上の機械学習済みモデルを記憶するだけの記憶容量を有していない。撮像部は、画像を撮影する。撮像部は、例えば光学系を用いて撮像素子上に像を結ばせて、画像を撮影する。この画像は、静止画であってもよいし、動画であってもよい。エッジデバイスの機能は、メモリ又はストレージに記憶されたクライアントプログラムと、このプログラムを実行するプロセッサとの協働により、プロセッサが演算を行い又は通信部による通信を制御することにより実現される。 For example, an edge device is a camera on a production line, and performs predetermined processing by inputting image data indicating a captured image into a machine-learned model. For example, different processing is performed depending on the machine-learned model associated with the process change information. The edge device is a computer including a processor, a memory, a storage, a communication unit, and an imaging unit, for example. These devices are connected via a bus. As shown in FIG. 4, the storage does not have a storage capacity for storing, for example, four or more types of machine-learned models. The imaging unit captures an image. The imaging unit captures an image by forming an image on the imaging element using, for example, an optical system. This image may be a still image or a moving image. The function of the edge device is realized by the processor performing calculations or controlling communication by the communication unit in cooperation with the client program stored in the memory or storage and the processor that executes the program.
[動作の説明]
次に、図４に基づいて、工程変更機械学習済みモデル切り替え方法について説明する。本発明の工程変更機械学習済みモデル切り替え方法は、製造ラインの工程変更に対応するために、クラウドから当該製造ラインに設置されたエッジデバイスに対して、工程変更に対応する機械学習済みモデルに切り替える方法である。 [Description of operation]
Next, a process change machine learned model switching method will be described with reference to FIG. The process change machine-learned model switching method of the present invention switches from a cloud to a machine-learned model corresponding to a process change for an edge device installed on the production line in order to cope with a process change of the production line. Is the method.
　工程変更機械学習済みモデル切り替え方法は、把握ステップ、第１制御ステップ、第２制御ステップ、を備える。また図示しないが、同様に、更新ステップ、連携ステップ、を備えてもよい。上述の各ステップが、単独のコンピュータで実現されてもよいし、２台以上のコンピュータ（例えば、サーバとエッジデバイスのような場合）で実現されてもよい。 The process change machine-learned model switching method includes a grasping step, a first control step, and a second control step. Moreover, although not shown in figure, you may provide an update step and a cooperation step similarly. Each of the above steps may be realized by a single computer, or may be realized by two or more computers (for example, a server and an edge device).
　把握取得ステップは、工程変更に関する工程変更情報を把握する。例えば、工程スケジュールと連携することで工程変更情報を把握することができる。また、エッジデバイスが有するセンサから出力されるセンサデータを解析することで、工程変更に関する工程変更情報を把握してもよい。例えば、エッジデバイスがカメラである場合、画像データを解析することで、異物検知処理の工程から欠損検知処理の工程に変更になったことを把握することができる。例えば、カメラであれば、製造ラインに流れてきている物体を撮像して、画像データを解析することで工程の変更を把握することができる。また、カメラであれば、工程を示す特定の物や文字などを撮像して、画像データを解析することで工程を把握することもできる。例えば、製造ラインに、欠損検知処理の文字列が記載された紙や、欠損検知処理を示す箱を流し、それをカメラで撮像して、画像データを解析することで欠損検知処理の工程に変更になったことを把握できる。 The grasp acquisition step grasps process change information related to the process change. For example, the process change information can be grasped by linking with the process schedule. Further, the process change information related to the process change may be grasped by analyzing sensor data output from a sensor included in the edge device. For example, when the edge device is a camera, it is possible to grasp that the foreign object detection process has been changed to the defect detection process by analyzing the image data. For example, in the case of a camera, it is possible to grasp a change in a process by capturing an image of an object flowing on a production line and analyzing the image data. In the case of a camera, it is also possible to grasp a process by imaging a specific object or character indicating the process and analyzing the image data. For example, a paper with a character string for defect detection processing or a box indicating the defect detection processing is poured on the production line, the image is captured with a camera, and the image data is analyzed to change to the defect detection processing step. You can understand that.
　第１制御ステップは、エッジデバイスに対して、把握された工程変更情報に対応付けられた機械学習済みモデルをクラウドから取得させるように制御する。例えば、エッジデバイスに対して、工程変更情報が欠損検知処理であった場合には、図２にあるように、欠損検知処理に対応付けられた欠損検知用の機械学習済みモデルをクラウドから取得させるように制御する。図２にあるように、欠損検知処理以外にも対応付けられた様々な機械学習済みモデルがある。特に第１制御ステップは、ソフトウェアエージェントとしてエッジデバイスにインストールされることがある。このソフトウェアエージェントを介して、把握された工程変更情報に対応付けられた機械学習済みモデルをクラウドから取得させるようにエッジデバイスを制御する。 The first control step controls the edge device to acquire the machine-learned model associated with the grasped process change information from the cloud. For example, when the process change information is a defect detection process for an edge device, as shown in FIG. 2, a machine-learned model for defect detection associated with the defect detection process is acquired from the cloud. To control. As shown in FIG. 2, there are various machine-learned models associated with the defect detection process. In particular, the first control step may be installed on the edge device as a software agent. Via this software agent, the edge device is controlled so that the machine-learned model associated with the grasped process change information is acquired from the cloud.
また、第１制御ステップは、セキュリティを考慮して暗号化通信方式を用いて、エッジデバイスに対して機械学習済みモデルを取得させるように制御してもよい。機械学習済みモデルはノウハウの塊であるので、セキュリティは重要となる。 In the first control step, the edge device may be controlled to acquire a machine-learned model using an encrypted communication method in consideration of security. Since machine-learned models are a lump of know-how, security is important.
　第２制御ステップは、エッジデバイスに対して、既に適用されている別工程の機械学習済みモデルから、取得された工程変更情報に対応付けられた機械学習済みモデルに切り替えるように制御する。図３は、工程変更に対応する機械学習済みモデルに切り替える一例である。図３にあるように、既に異物検知処理の工程が行われている場合はカメラに対して異物検知用の機械学習済みモデルが適用されている。これを欠損検知処理の工程に工程変更するためには、カメラに対して、既に適用されている異物検知用の機械学習済みモデルから、欠損検知用の機械学習済みモデルに切り替えなければならない。そうしなければ、欠損検知処理を行えない。例えば、既に適用されている異物検知用の機械学習済みモデルは、切り替えられる際に、欠損検知用の機械学習済みモデルに上書きされるため、消去されてもよい。常にクラウドから、その瞬間に必要な機械学習済みモデルを取得して切り替えるため、記憶容量の少ないカメラに対しても効果的である。特に第２制御ステップは、ソフトウェアエージェントとしてエッジデバイスにインストールされることがある。このソフトウェアエージェントを介して、既に適用されている別工程の機械学習済みモデルから、取得された工程変更情報に対応付けられた機械学習済みモデルに切り替えるように制御する。 In the second control step, the edge device is controlled to switch from the machine learned model of another process already applied to the machine learned model associated with the acquired process change information. FIG. 3 is an example of switching to a machine-learned model corresponding to a process change. As shown in FIG. 3, when a foreign matter detection process is already performed, a machine-learned model for foreign matter detection is applied to the camera. In order to change the process to the defect detection process, it is necessary to switch from a machine learning model for foreign object detection already applied to the camera to a machine learned model for defect detection. Otherwise, the defect detection process cannot be performed. For example, a machine-learned model for foreign object detection that has already been applied may be deleted because it is overwritten with the machine-learned model for defect detection when it is switched. Since the machine-learned model necessary for the moment is always acquired from the cloud and switched, it is effective for a camera with a small storage capacity. In particular, the second control step may be installed on the edge device as a software agent. Via this software agent, control is performed so that the machine-learned model already applied in another process is switched to the machine-learned model associated with the acquired process change information.
更新ステップは、学習器を用いて、エッジデバイスから取得されたセンサデータを学習データとして機械学習することにより、機械学習済みモデルを最新バージョンに更新する。図３のように、既に異物検知処理の工程が行われている場合は、カメラから異物検知用の画像データを取得して学習データとして学習器を用いて機械学習することで、更に学習の進んだ最新の異物検知用の機械学習モデルを生成して更新することが可能となる。そして、第１制御ステップは、エッジデバイスに対して、更新された最新バージョンの機械学習済みモデルをクラウドから取得させるように制御する。そして、第２制御ステップは、エッジデバイスに対して、既に適用されている別工程の機械学習済みモデルから、取得された前記更新された最新バージョンの機械学習済みモデルに切り替えるように制御する。例えば、異物検知処置→欠損検知処理→パッケージ不良検知処理→サイズ不良検知処理→異物検知処理の順番で工程変更になる場合、1回目の異物検知処理の工程の際に最新バージョンの学習済みモデルになるように更新しておき、工程変更を経て2回目の異物検知処理に変更されるタイミングで、最新バージョンの機械学習済みモデルを取得するように制御して、既に適用されている別工程の機械学習済みモデルから取得された前記更新された最新バージョンの機械学習済みモデルに切り替えるように制御する。 In the updating step, the machine learning model is updated to the latest version by machine learning using the learning device as sensor data obtained from the edge device. As shown in FIG. 3, when the foreign object detection process has already been performed, the image data for foreign object detection is acquired from the camera, and machine learning is performed as learning data using a learning device, thereby further progress of learning. However, it is possible to generate and update the latest machine learning model for foreign object detection. In the first control step, the edge device is controlled to acquire the updated latest version of the machine-learned model from the cloud. In the second control step, the edge device is controlled to switch from the machine-learned model of another process that has already been applied to the acquired updated latest version of the machine-learned model. For example, if the process changes in the order of foreign object detection treatment → defect detection process → package defect detection process → size defect detection process → foreign object detection process, the latest version of the learned model is used during the first foreign object detection process. In order to obtain the latest version of the machine-learned model at the timing when the process is changed and changed to the second foreign object detection process, the machine in another process that has already been applied is controlled. Control is performed to switch to the updated latest version of the machine-learned model obtained from the learned model.
連携ステップは、AIベンダーが提供している外部のAIプラットフォームとAPI連携する。例えば、AIベンダーはそれぞれ自社のAIプラットフォームを持っていることが多い。そこで、AIプラットフォームがAPI（Application Programming Interface）を公開していれば、連携ステップはそれらのAIプラットフォームとAPI連携することができる。つまり、接続可能となる。そして、第１制御ステップは、エッジデバイスに対して、工程変更情報に対応付けられた機械学習済みモデルを、このAIプラットフォームから取得させるように制御し、第２制御ステップは、エッジデバイスに対して、既に適用されている別工程の機械学習済みモデルから、外部のAIプラットフォームから取得された機械学習済みモデルに切り替えるように制御する。つまり、AIベンダーが自社のAIプラットフォームに欠損検知用の機械学習済みモデルを登録していれば、工程変更情報は欠損検知処理である場合に、連携ステップによってそのAIプラットフォームとAPI連携して、第１制御ステップによってこのAIプラットフォームから欠損検知用の機械学習済みモデルを取得するように制御して、第２制御ステップによってエッジデバイスに対して既に適用されている別工程の機械学習済みモデルからこのAIプラットフォームから取得された欠損検知用の機械学習済みモデルに切り替えるように制御する。 The linkage step is API linkage with an external AI platform provided by an AI vendor. For example, AI vendors often have their own AI platforms. Therefore, if the AI platform publishes API (Application Programming 連 携 Interface), the linkage step can be API-linked with those AI platforms. That is, connection is possible. The first control step controls the edge device to acquire the machine-learned model associated with the process change information from the AI platform, and the second control step controls the edge device. Then, control is performed so as to switch from a machine-learned model that has already been applied to a machine-learned model acquired from an external AI platform. In other words, if the AI vendor has registered a machine-learned model for defect detection on their AI platform, and the process change information is defect detection processing, the AI platform is linked with the AI platform through the linkage step, Control is performed so that a machine-learned model for defect detection is acquired from this AI platform by one control step, and this AI is obtained from a machine-learned model of another process that has already been applied to the edge device by the second control step. Control to switch to the machine-learned model for defect detection acquired from the platform.
　以上説明した実施形態によれば、製造ラインの工程変更に対応するために、クラウドからエッジデバイスに対して、工程変更に対応する機械学習済みモデルに切り替えることができる。また、エッジデバイスは、複数の機械学習済みモデルを記憶するだけの記憶容量を有していなくてもよいため、エッジデバイスが必要なハードウェアリソースを減らすことができる。また、機械学習済みモデルの更新には高い処理能力が必要となるが、上述した実施形態では、AIベンダーが新機械学習済みモデルを更新しているため、エッジデバイスが高い処理能力を有していなくてもよい。 According to the embodiment described above, in order to cope with the process change of the production line, it is possible to switch from the cloud to the machine learned model corresponding to the process change for the edge device. Further, since the edge device does not have to have a storage capacity for storing a plurality of machine-learned models, hardware resources required for the edge device can be reduced. In addition, high processing capability is required to update a machine-learned model. However, in the above-described embodiment, since an AI vendor updates a new machine-learned model, the edge device has high processing capability. It does not have to be.
　さらに、上述した実施形態では、エッジデバイスに対して、機械学習済みモデルをクラウドから取得させるため、情報セキュリティが高くなる。つまり、悪意ある別工程の装置から送信された情報を受け取ることがないため、これによる被害がないため情報セキュリティが高くなる。 Furthermore, in the above-described embodiment, since the machine-learned model is acquired from the cloud by the edge device, information security is increased. In other words, since information transmitted from a malicious device in another process is not received, there is no damage caused by this, and information security is increased.
　さらに、上述した実施形態によれば、暗号化通信方式を用いて機械学習済みモデルが把握されるため、悪意ある第三者により機械学習済みモデルが把握されるのを防ぐことができる。 Furthermore, according to the above-described embodiment, since the machine-learned model is grasped using the encrypted communication method, it is possible to prevent the machine-learned model from being grasped by a malicious third party.
　本発明は上述した実施形態に限定されない。上述した実施形態に対し、種々の変形がなされてもよい。また、以下の変形例が組み合わせて実施されてもよい。 The present invention is not limited to the embodiment described above. Various modifications may be made to the above-described embodiment. Moreover, the following modifications may be implemented in combination.
　上述した実施形態において、エッジデバイスは、カメラに限定されない。エッジデバイスは、機械学習済みモデルを用いて処理を行う装置であれば、どのような装置であってもよい。また、エッジデバイスは、センサを備え、このセンサから出力されたセンサデータが、学習データとして用いられてもよい。 In the embodiment described above, the edge device is not limited to a camera. The edge device may be any device as long as the device performs processing using the machine-learned model. The edge device may include a sensor, and sensor data output from the sensor may be used as learning data.
　上述した実施形態において、エッジデバイスが切り替える機械学習済みモデルの数は、２つに限定されない。エッジデバイスは、３つ以上の機械学習済みモデルを切り替えて使用してもよい。 In the embodiment described above, the number of machine-learned models that the edge device switches is not limited to two. The edge device may switch between three or more machine-learned models.
　上述した実施形態において、機械学習のアルゴリズムは、例えばコンピュータ上において、与えられたデータから機械学習した結果に基づいてモデルを生成し、そのモデルに対してさらに新たな入力データを入力することで、その入力データから予測される事象を出力するための、いわゆる教師あり学習のアルゴリズムであってもよい。ただし、機械学習用のアルゴリズムは、いわゆる教師あり学習のアルゴリズムに限定されず、教師なし学習、半教師あり学習、強化学習、表現学習等の機械学習用のアルゴリズムであってもよい。また、機械学習用のアルゴリズムは、データマイニングやディープラーニング等の、その他の学習用のアルゴリズムを含んでもよい。なお、これらの学習用のアルゴリズムは、例えば決定木学習、相関ルール学習、ニューラルネットワーク、遺伝的プログラミング、帰納論理プログラミング、サポートベクターマシン、クラスタリング、ベイジアンネットワーク等の各種の技法乃至技術を用いたものが含まれる。要するに、機械学習用のアルゴリズムは、データ提供者により提供される何らかのデータとともに処理されて、その処理の結果、ユーザが得たい情報を出力するものであればよい。 In the embodiment described above, the machine learning algorithm generates a model based on the result of machine learning from given data, for example, on a computer, and inputs new input data to the model. A so-called supervised learning algorithm for outputting an event predicted from the input data may be used. However, the machine learning algorithm is not limited to a so-called supervised learning algorithm, and may be an algorithm for machine learning such as unsupervised learning, semi-supervised learning, reinforcement learning, and expression learning. Further, the algorithm for machine learning may include other learning algorithms such as data mining and deep learning. Note that these learning algorithms use various techniques or techniques such as decision tree learning, correlation rule learning, neural network, genetic programming, inductive logic programming, support vector machine, clustering, and Bayesian network. included. In short, any machine learning algorithm may be used as long as it is processed together with some data provided by the data provider and outputs information desired by the user as a result of the processing.
　工程変更機械学習済みモデル切り替え方法において行われる処理のステップは、上述した実施形態で説明した例に限定されない。この処理のステップは、矛盾のない限り、入れ替えられてもよい。また、本発明は、工程変更機械学習済みモデル切り替えシステムにおいて行われる工程変更機械学習済みモデル切り替え方法として提供されてもよい。 The process steps performed in the process change machine-learned model switching method are not limited to the example described in the above embodiment. The steps of this process may be interchanged as long as there is no contradiction. Further, the present invention may be provided as a process change machine learned model switching method performed in a process change machine learned model switching system.
　本発明は、クラウドまたはエッジデバイスにおいて実行されるプログラムとして提供されてもよい。このプログラムは、インターネットなどの通信回線を介してダウンロードされてもよいし、磁気記録媒体（磁気テープ、磁気ディスクなど）、光記録媒体（光ディスクなど）、光磁気記録媒体、半導体メモリなどの、コンピュータが読取可能な記録媒体に記録した状態で提供されてもよい。 The present invention may be provided as a program executed in a cloud or an edge device. This program may be downloaded via a communication line such as the Internet, or a computer such as a magnetic recording medium (magnetic tape, magnetic disk, etc.), an optical recording medium (optical disk, etc.), a magneto-optical recording medium, or a semiconductor memory. May be provided in a state of being recorded on a readable recording medium.
　上述した手段、機能は、コンピュータ（ＣＰＵ、情報処理装置、各種端末を含む）が、所定のプログラムを読み込んで、実行することによって実現される。プログラムは、例えば、コンピュータにインストールされるアプリケーションであってもよいし、コンピュータからネットワーク経由で提供されるＳａａＳ（ソフトウェア・アズ・ア・サービス）形態であってもよいし、例えば、フレキシブルディスク、ＣＤ（ＣＤ－ＲＯＭなど）、ＤＶＤ（ＤＶＤ－ＲＯＭ、ＤＶＤ－ＲＡＭなど）等のコンピュータ読取可能な記録媒体に記録された形態で提供されてもよい。この場合、コンピュータはその記録媒体からプログラムを読み取って内部記憶装置または外部記憶装置に転送し記憶して実行する。また、そのプログラムを、例えば、磁気ディスク、光ディスク、光磁気ディスク等の記憶装置（記録媒体）に予め記録しておき、その記憶装置から通信回線を介してコンピュータに提供するようにしてもよい。 The means and functions described above are realized by a computer (including a CPU, an information processing apparatus, and various terminals) reading and executing a predetermined program. The program may be, for example, an application installed on a computer, or may be in the form of SaaS (software as a service) provided from a computer via a network, for example, a flexible disk, a CD It may be provided in a form recorded on a computer-readable recording medium such as a CD-ROM (DVD-ROM, DVD-RAM, etc.). In this case, the computer reads the program from the recording medium, transfers it to the internal storage device or the external storage device, stores it, and executes it. The program may be recorded in advance in a storage device (recording medium) such as a magnetic disk, an optical disk, or a magneto-optical disk, and provided from the storage device to a computer via a communication line.
上述した機械学習の具体的なアルゴリズムとしては、最近傍法、ナイーブベイズ法、決定木、サポートベクターマシン、強化学習などを利用してよい。また、ニューラルネットワークを利用して、学習するための特徴量を自ら生成する深層学習（ディープラーニング）であってもよい。 As a specific algorithm of the machine learning described above, a nearest neighbor method, a naive Bayes method, a decision tree, a support vector machine, reinforcement learning, or the like may be used. Further, deep learning may be used in which a characteristic amount for learning is generated by using a neural network.
　以上、本発明の実施形態について説明したが、本発明は上述したこれらの実施形態に限るものではない。また、本発明の実施形態に記載された効果は、本発明から生じる最も好適な効果を列挙したに過ぎず、本発明による効果は、本発明の実施形態に記載されたものに限定されるものではない。

  As mentioned above, although embodiment of this invention was described, this invention is not limited to these embodiment mentioned above. The effects described in the embodiments of the present invention are only the most preferable effects resulting from the present invention, and the effects of the present invention are limited to those described in the embodiments of the present invention. is not.

Claims (9)

  1. 　製造ラインの工程変更に対応するために、クラウドから当該製造ラインに設置されたエッジデバイスに対して、工程変更に対応する機械学習済みモデルに切り替える工程変更機械学習済みモデル切り替えシステムであって、
    　工程変更に関する工程変更情報を把握する把握手段と、
    　前記エッジデバイスに対して、把握された前記工程変更情報に対応付けられた機械学習済みモデルをクラウドから取得させるように制御する第１制御手段と、
    　前記エッジデバイスに対して、既に適用されている別工程の機械学習済みモデルから、取得された前記工程変更情報に対応付けられた機械学習済みモデルに切り替えるように制御する第２制御手段と、
    　を備える工程変更機械学習済みモデル切り替えシステム。     A process change machine learning model switching system for switching to a machine learned model corresponding to a process change for an edge device installed in the production line from the cloud in order to cope with a process change of the production line,
    A grasping means for grasping process change information related to a process change;
    First control means for controlling the edge device to acquire a machine-learned model associated with the grasped process change information from the cloud;
    Second control means for controlling the edge device to switch from a machine-learned model already applied in another process to a machine-learned model associated with the acquired process change information;
    Process change machine learning completed model switching system.
  2. 　前記第１制御手段は、暗号化通信方式を用いて、前記エッジデバイスに対して、前記機械学習済みモデルを取得させるように制御する
    　請求項１に記載の工程変更機械学習済みモデル切り替えシステム。     The process change machine-learned model switching system according to claim 1, wherein the first control unit controls the edge device to acquire the machine-learned model using an encrypted communication method.
  3. 　学習器を用いて、前記エッジデバイスから取得されたセンサデータを学習データとして機械学習することにより、前記機械学習済みモデルを最新バージョンに更新する更新手段を備え、
    　前記第１制御手段は、前記エッジデバイスに対して、前記更新された最新バージョンの機械学習済みモデルをクラウドから取得させるように制御し、
    前記第２制御手段は、前記エッジデバイスに対して、既に適用されている別工程の機械学習済みモデルから、取得された前記更新された最新バージョンの機械学習済みモデルに切り替えるように制御する
    　請求項１に記載の工程変更機械学習済みモデル切り替えシステム。     By using a learning device, by performing machine learning sensor data acquired from the edge device as learning data, an update means for updating the machine-learned model to the latest version,
    The first control means controls the edge device to acquire the updated latest version of the machine-learned model from the cloud,
    The second control means controls the edge device to switch from a machine learning model already applied in another process to the updated latest version of the machine learning model that has been applied. The process change machine learning completed model switching system according to 1.
  4. AIベンダーが提供している外部のAIプラットフォームとAPI連携する連携手段を備え、
    　前記第１制御手段は、前記エッジデバイスに対して、前記工程変更情報に対応付けられた機械学習済みモデルを前記外部のAIプラットフォームから取得させるように制御し、
    前記第２制御手段は、前記エッジデバイスに対して、既に適用されている別工程の機械学習済みモデルから、前記外部のAIプラットフォームから取得された機械学習済みモデルに切り替えるように制御する
    請求項１に記載の工程変更機械学習済みモデル切り替えシステム。     Provide a way to link APIs with external AI platforms provided by AI vendors,
    The first control means controls the edge device to acquire a machine-learned model associated with the process change information from the external AI platform,
    The second control means controls the edge device to switch from a machine-learned model already applied in another process to a machine-learned model acquired from the external AI platform. The process change machine learning model switching system described in 1.
  5. 　前記把握手段は、前記エッジデバイスが有するセンサから出力されるセンサデータを解析することで、工程変更に関する工程変更情報を把握する
    請求項１に記載の工程変更機械学習済みモデル切り替えシステム。     The process change machine-learned model switching system according to claim 1, wherein the grasping unit grasps process change information related to a process change by analyzing sensor data output from a sensor included in the edge device.
  6. 　前記把握手段は、前記工程変更のスケジュール帳と連携することで工程変更情報を把握する
    請求項１に記載の工程変更機械学習済みモデル切り替えシステム。     The process change machine-learned model switching system according to claim 1, wherein the grasping means grasps process change information in cooperation with the process change schedule book.
  7. 　製造ラインに設置されており、当該製造ラインの工程変更に対応するために、工程変更に対応する機械学習済みモデルに切り替えて使用するエッジデバイスであって、
    　請求項１に記載の工程変更機械学習済みモデル切り替えシステムの制御に従って、クラウドから前記工程変更情報に対応付けられた機械学習済みモデルを取得し、既に使用している別工程の機械学習済みモデルから、前記取得された工程変更情報に対応付けられた機械学習済みモデルに切り替えるエッジデバイス。     An edge device that is installed in a production line and is used by switching to a machine-learned model corresponding to the process change in order to cope with the process change of the production line,
    According to the control of the process-change machine-learned model switching system according to claim 1, a machine-learned model associated with the process-change information is acquired from the cloud, and the machine-learned model of another process that is already used An edge device that switches to a machine-learned model associated with the acquired process change information.
  8. 　製造ラインの工程変更に対応するために、クラウドから当該製造ラインに設置されたエッジデバイスに対して、工程変更に対応する機械学習済みモデルに切り替える工程変更機械学習済みモデル切り替え方法であって、
    　工程変更に関する工程変更情報を把握する把握ステップと、
    　前記エッジデバイスに対して、把握された前記工程変更情報に対応付けられた機械学習済みモデルをクラウドから取得させるように制御する第１制御ステップと、
    　前記エッジデバイスに対して、既に適用されている別工程の機械学習済みモデルから、取得された前記工程変更情報に対応付けられた機械学習済みモデルに切り替えるように制御する第２制御ステップと、
    　を備える工程変更機械学習済みモデル切り替え方法。     In order to cope with the process change of the production line, for the edge device installed in the production line from the cloud, a process change machine learning completed model switching method for switching to a machine learned model corresponding to the process change,
    A grasping step for grasping process change information related to a process change;
    A first control step of controlling the edge device to acquire a machine-learned model associated with the grasped process change information from the cloud;
    A second control step for controlling the edge device to switch from a machine-learned model of another process that has already been applied to a machine-learned model associated with the acquired process change information;
    A process change machine-learned model switching method comprising:
  9. 　コンピュータに、製造ラインの工程変更に対応するために、クラウドから当該製造ラインに設置されたエッジデバイスに対して、工程変更に対応する機械学習済みモデルに切り替える処理を実行させるためのプログラムであって、
    　工程変更に関する工程変更情報を把握する把握ステップと、
    　前記エッジデバイスに対して、把握された前記工程変更情報に対応付けられた機械学習済みモデルをクラウドから取得させるように制御する第１制御ステップと、
    　前記エッジデバイスに対して、既に適用されている別工程の機械学習済みモデルから、取得された前記工程変更情報に対応付けられた機械学習済みモデルに切り替えるように制御する第２制御ステップと、
    　を実行させるためのプログラム。
          A program for causing a computer to execute a process of switching from a cloud to a machine-learned model corresponding to a process change for an edge device installed on the production line in order to respond to a process change of the production line. ,
    A grasping step for grasping process change information related to a process change;
    A first control step of controlling the edge device to acquire a machine-learned model associated with the grasped process change information from the cloud;
    A second control step for controlling the edge device to switch from a machine-learned model of another process that has already been applied to a machine-learned model associated with the acquired process change information;
    A program for running
PCT/JP2018/021215 2018-06-01 2018-06-01 Step change machine-learned model switching system, edge device, step change machine-learned model switching method, and program WO2019229983A1 (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/JP2018/021215 WO2019229983A1 (en) 2018-06-01 2018-06-01 Step change machine-learned model switching system, edge device, step change machine-learned model switching method, and program

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/JP2018/021215 WO2019229983A1 (en) 2018-06-01 2018-06-01 Step change machine-learned model switching system, edge device, step change machine-learned model switching method, and program

Publications (1)

Publication Number Publication Date
WO2019229983A1 true WO2019229983A1 (en) 2019-12-05

Family

ID=68697956

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP2018/021215 WO2019229983A1 (en) 2018-06-01 2018-06-01 Step change machine-learned model switching system, edge device, step change machine-learned model switching method, and program

Country Status (1)

Country Link
WO (1) WO2019229983A1 (en)

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2017211713A (en) * 2016-05-23 2017-11-30 ルネサスエレクトロニクス株式会社 Production system

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2017211713A (en) * 2016-05-23 2017-11-30 ルネサスエレクトロニクス株式会社 Production system

Similar Documents

Publication Publication Date Title
KR102414096B1 (en) Create and deploy packages for machine learning on end devices
US11113585B1 (en) Artificially intelligent systems, devices, and methods for learning and/or using visual surrounding for autonomous object operation
KR102225822B1 (en) Apparatus and method for generating learning data for artificial intelligence performance
US20210158147A1 (en) Training approach determination for large deep learning models
US20190244287A1 (en) Utilizing a machine learning model and blockchain technology to manage collateral
US20190130272A1 (en) Generating compressed representation neural networks having high degree of accuracy
US9852070B2 (en) Cache memory system using a tag comparator to determine update candidates and operating method thereof
KR20220054398A (en) Custom route processes for application groups
Arnold An active-set evolution strategy for optimization with known constraints
JP2023544186A (en) Adversarial interpolation backdoor detection
WO2019229983A1 (en) Step change machine-learned model switching system, edge device, step change machine-learned model switching method, and program
CN113641525A (en) Variable exception recovery method, apparatus, medium, and computer program product
CN111414343B (en) Log writing method, device, electronic equipment and medium
JP7310673B2 (en) Data management system, data management method, and data management program
CN114175069A (en) Distributed machine learning with privacy protection
JP6759472B2 (en) Trained model update system, trained model update method, and program
CN111727108B (en) Method, device and system for controlling robot and storage medium
US20190332992A1 (en) Cross domain integration in product lifecycle management
JP2020035000A (en) Machine learning system and calculation method of boltzmann machine
US20190385091A1 (en) Reinforcement learning exploration by exploiting past experiences for critical events
US11803999B2 (en) Job scheduling using reinforcement learning
US11586176B2 (en) High performance UI for customer edge IIoT applications
KR102315386B1 (en) Object detecting system and operating method thereof
WO2020110272A1 (en) Machine learning device, machine learning method, and computer-readable recording medium
JP2021179702A (en) Data processing device, method, computer program, and recording medium

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18920441

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 18920441

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: JP