Management device, program and processing system

Abstract

To stably perform data acquisition and analysis processing.SOLUTION: In a processing system 1, when an AI edge device 11A in the operating state detects an abnormality, under the control of a management server 31, the AI edge device 11A is switched to the standby state, and an AI edge device 11B in the standby state is selected as the switching destination, and a container 46A is caused to be deployed and then switched to the operating state. Therefore, in the processing system 1, even when AI processing cannot be continued in the switching source AI edge device 11A, the switching processing to the switching destination AI edge device 11B can be completed in a relatively short time, and the AI processing can be started immediately to resume the generation and delivery of analytical data.SELECTED DRAWING: Figure 7

Description

[Technical field]

[0001] The present invention Management device and program The present invention relates to a processing system and, for example, is suitable for application to a processing system in which a management device manages multiple AI edge devices that perform AI (Artificial Intelligence) processing. [Background technology]

[0002] In recent years, the technology known as IoT (Internet of Things) has been expected to connect various sensors and home appliances to a network to collect a wide variety of data, and to use this data to gain new knowledge and lead to further technological development. For example, the data collected in this way can be used for advanced analysis and interpretation by performing deep learning using AI processing.

[0003] However, the amount of data collected in this way is extremely large, and when it includes high-resolution video and audio, each piece of data also becomes extremely large. For this reason, if this data were collected directly on a cloud or other network, it would generate a huge amount of traffic during communication and require a huge amount of memory space when stored. In addition, it would be necessary to prepare computer resources on the cloud side to perform a huge amount of AI processing.

[0004] Therefore, it is conceivable to use a technology called edge computing to build a processing system in which an AI edge device is placed between sensors and a network. In this processing system, the AI ​​edge device performs AI processing on data acquired from the sensor to generate analysis data, and transmits this analysis data to the cloud side via the network. As a result, this processing system can significantly reduce the communication capacity on the network, the amount of computational processing on the cloud side, and memory capacity, etc., compared to a case in which all data is directly transmitted to the cloud side (for example, see Non-Patent Document 1). [Prior art documents] [Non-patent literature]

[0005] [Non-Patent Document 1] Oki Electric Industry Co., Ltd., "Opening up the future with AI edge computing", [online], October 3, 2019, Oki Electric Industry Co., Ltd., [searched February 9, 2021], Internet<URL:https: / / www.oki.com / jp / AIedge / img / AIedgeComputing_WhitePaper.pdf> Summary of the Invention [Problem to be solved by the invention]

[0006] In a processing system having such a configuration, when multiple AI edge devices are connected to a network, it is possible to connect a dedicated management server to the network and manage each AI edge device through the management server. In this case, the management server can, for example, monitor the operating status of each AI edge device, and if it detects an abnormality in the AI ​​edge device, it can instruct the AI ​​edge device to rerun various programs or restart the AI ​​edge device.

[0007] However, in this processing system, if the abnormality in the AI ​​edge device is not resolved by rebooting, etc., a maintenance worker or the like must perform maintenance work such as repairing or replacing the AI ​​edge device. In such cases, since the processing system requires time to arrange for a maintenance worker and perform the maintenance work, there is a possibility that problems will occur, such as the inability to transmit analysis data to the cloud side or missing data from the sensor until the maintenance work is completed.

[0008] The present invention has been made in consideration of the above points, and is capable of stably acquiring data and carrying out analysis processing. Management device and program The present invention proposes a processing system for detecting the presence of a magnetic field. [Means for solving the problem]

[0009] In order to solve such problems, the present invention management device In this system, multiple AI edge devices execute AI (Artificial Intelligence) processing based on data supplied from sensor devices. of Management device to manage And, Acquire information showing the status of each AI edge device Department and AI edge device anomaly detection Department and selecting a switching destination AI edge device that is another AI edge device that should execute the target AI processing that is the AI ​​processing that was being executed in the switching source AI edge device that is the AI ​​edge device in which the abnormality was detected. Department And the AI ​​edge device to be switched to In response to this, the AI ​​edge device Target AI processing A transmitter that sends instructions to execute And The detection unit detects an abnormality in the switching source AI edge device by receiving a switching request from the switching source AI edge device to request that the target AI processing be switched to another AI edge device. I did so. In addition, the management device of the present invention manages a plurality of AI edge devices that execute AI processing based on data supplied from a sensor device, and is provided with an acquisition unit that acquires information representing the state of each AI edge device, a management unit that manages containers that virtualize applications executed by the AI ​​edge devices, a detection unit that detects abnormalities in the AI ​​edge devices, a selection unit that selects a destination AI edge device that is another AI edge device that should execute a target AI processing, which is the AI ​​processing that was being executed in the source AI edge device, which is the AI ​​edge device in which the abnormality was detected, and a transmission unit that transmits an instruction to the destination AI edge device to execute the target AI processing in the destination AI edge device, and if the destination AI edge device does not have a container for executing the target AI processing, the transmission unit transmits the container to the destination AI edge device so that the container that virtualizes the application is deployed and executed in the destination AI edge device. Furthermore, the management device of the present invention is a management device that manages a plurality of AI edge devices that execute AI processing based on data supplied from a sensor device, and is provided with an acquisition unit that acquires information representing the state of each AI edge device, a detection unit that detects abnormalities in the AI ​​edge devices, a selection unit that selects a destination AI edge device, which is another AI edge device that should execute a target AI processing, which is the AI ​​processing that was being executed in the source AI edge device, which is the AI ​​edge device in which the abnormality was detected, and a transmission unit that transmits an instruction to the destination AI edge device to execute the target AI processing in the destination AI edge device, and the selection unit is an AI edge device that is pre-associated in a set of two or more devices, and one of the remaining AI edge devices pre-associated with the source AI edge device is selected as the destination AI edge device. Furthermore, the management device of the present invention is a management device that manages a plurality of AI edge devices that execute AI processing based on data supplied from a sensor device, and is provided with an acquisition unit that acquires information representing the state of each AI edge device, a detection unit that detects abnormalities in the AI ​​edge devices, a selection unit that selects a destination AI edge device, which is another AI edge device that should execute a target AI processing, which is the AI ​​processing that was being executed in the source AI edge device, which is the AI ​​edge device in which the abnormality was detected, and a transmission unit that transmits an instruction to the destination AI edge device to execute the target AI processing in the destination AI edge device, and the selection unit selects another AI edge device that is capable of acquiring data necessary for the target AI processing as the destination AI edge device. Furthermore, the management device of the present invention manages a plurality of AI edge devices that execute AI processing based on data supplied from a sensor device, and is provided with an acquisition unit that acquires information representing the status of each AI edge device, a memory unit that stores the storage capacity required to execute the AI ​​processing in the AI ​​edge device in correspondence with the AI ​​processing, a detection unit that detects abnormalities in the AI ​​edge device, a selection unit that selects a destination AI edge device that is another AI edge device that should execute a target AI processing, which is the AI ​​processing that was being executed in the source AI edge device where the abnormality was detected, and a transmission unit that transmits an instruction to the destination AI edge device to execute the target AI processing in the destination AI edge device, and the acquisition unit acquires information representing the available storage capacity in each of the AI ​​edge devices in addition to information on the AI ​​processing being executed in each AI edge device, and the selection unit selects the other AI edge device whose available storage capacity is greater than the storage capacity required to execute the target AI processing as the destination AI edge device. Furthermore, the management device of the present invention is a management device that manages a plurality of AI edge devices that execute AI processing based on data supplied from a sensor device, and is provided with an acquisition unit that acquires information representing the status of each AI edge device, a detection unit that detects an abnormality in the AI ​​edge device, a selection unit that selects a destination AI edge device that is another AI edge device that should execute a target AI processing, which is an AI processing that was being executed in the source AI edge device, which is the AI ​​edge device in which the abnormality was detected, and a transmission unit that transmits an instruction to the destination AI edge device to execute the target AI processing in the destination AI edge device, and the acquisition unit acquires information representing the usage status of resources in each AI edge device, and the detection unit detects the resource usage status of the AI ​​edge device in addition to the abnormality in the AI ​​edge device. When a resource shortage is detected, the selection unit determines whether the target AI processing should be executed in a distributed manner by the allocating AI edge device, which is the AI ​​edge device that detected the resource shortage, and the assigned AI edge devices, which are one or more other AI edge devices to which the target AI processing should be assigned, or whether the target AI processing should be executed collectively only by the assigned AI edge device, and selects a number of assigned AI edge devices based on the result of the determination.When it is determined that the target AI processing should be executed in a distributed manner, the transmission unit transmits an instruction to execute the target AI processing in a distributed manner by the allocating AI edge device and the assigned AI edge devices, and when it is determined that the target AI processing should be executed collectively, it transmits an instruction to execute the target AI processing collectively by one assigned AI edge device.

[0010] moreover The present invention program In this case, A program for operating an information processing device as any one of the above-mentioned management devices, the program causing the information processing device to function as each unit included in the management device. I did so.

[0011] Furthermore, in the processing system of the present invention, a plurality of AI edge devices that execute AI processing based on data supplied from a sensor device and a management device that manages the AI ​​edge devices are connected to the processing system, The management device is any one of the management devices described above. I did so. Furthermore, the processing system of the present invention is a processing system in which a plurality of AI edge devices that perform AI processing based on data supplied from sensor devices are connected to a management device that manages the AI ​​edge devices, and the management device is an information processing device on which the above-mentioned program runs and functions as each of the parts provided in the management device.

[0012] The present invention relates to an AI edge device that executes AI processing. Receive a switching request from the AI ​​edge device If an abnormality is detected in Management deviceAnother AI edge device can be selected to execute the AI ​​processing, i.e., the AI ​​edge device can be switched to continue the AI ​​processing. This allows the present invention to continue AI processing of data obtained from the sensor device, and as a result, the generation of analysis data can be continued. Effect of the Invention

[0013] According to the present invention, data acquisition and analysis can be stably performed. Management device and program and a processing system. [Brief description of the drawings]

[0014] [Figure 1] 1 is a schematic block diagram showing a configuration of a processing system according to a first embodiment. [Diagram 2] FIG. 1 is a schematic block diagram showing the configuration of an AI edge device. [Diagram 3] FIG. 1 is a schematic diagram showing a hierarchical configuration of software in an AI edge device. [Figure 4] 2 is a schematic block diagram showing a configuration of a management server according to the first embodiment. FIG. [Diagram 5] 4 is a schematic diagram illustrating a configuration of a state management table according to the first embodiment. FIG. [Figure 6] 13 is a sequence chart showing a table update process. [Figure 7] 4 is a sequence chart showing a state switching process according to the first embodiment. [Figure 8] 4 is a schematic diagram illustrating a configuration of a state management table after updating according to the first embodiment. FIG. [Figure 9] FIG. 11 is a schematic block diagram showing a configuration of a processing system according to a second embodiment. [Figure 10] FIG. 11 is a schematic block diagram showing a configuration of a management server according to a second embodiment. [Figure 11] 13 is a schematic diagram illustrating a configuration of a state management table according to the second embodiment. [Figure 12]13 is a sequence chart showing an allocation change process according to the second embodiment. [Figure 13] FIG. 13 is a schematic block diagram showing a configuration of a processing system according to a third embodiment. [Figure 14] FIG. 13 is a schematic block diagram showing a configuration of a management server according to a third embodiment. [Figure 15] 13 is a schematic diagram illustrating a configuration of a state management table according to a third embodiment. [Figure 16] 13 is a flowchart showing a switching allocation request processing procedure according to the third embodiment. [Figure 17] 13 is a flowchart showing a switching assignment management process according to the third embodiment. [Figure 18] FIG. 13 is a schematic diagram illustrating a configuration of a state management table according to a fourth embodiment. [Figure 19] 13 is a sequence chart showing a batch allocation process according to the fourth embodiment. [Figure 20] 13 is a flowchart showing a switching allocation request processing procedure according to the fourth embodiment. [Figure 21] 13 is a flowchart showing a switching assignment management process according to the fourth embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0015] Hereinafter, modes for carrying out the invention (hereinafter referred to as embodiments) will be described with reference to the drawings.

[0016] 1. First embodiment [1-1. Processing system configuration] As shown in Figure 1, in the processing system 1 according to the first embodiment, a plurality of information processing devices such as a plurality of AI edge devices 11 and a management server 31 are installed at an installation site 2 and a management center 3, and each information processing device is connected to each other via a wide area network 4.

[0017] The installation site 2 is provided with a network 10, multiple AI edge devices 11 (11A, 11B, 11C, and 11D), multiple sensor devices 12 (12A and 12B), and a router 13. The network 10 connects the AI ​​edge devices 11 and the router 13 via a wired LAN (Local Area Network) that complies with standards such as IEEE (Institute of Electrical and Electronics Engineers) 802.3 (IEEE802.3u / ab / an / ae).

[0018] At the installation site 2, two AI edge devices 11A and 11B form a pair and are connected to the same sensor device 12A. At the installation site 2, two AI edge devices 11C and 11C also form a pair and are also connected to the same sensor device 12B.

[0019] 2, the AI ​​edge device 11 has a configuration in which a control unit 21, a storage unit 22, a communication unit 23, and a sensor connection unit 24 are connected to a bus 20. The control unit 21 has a central processing unit (CPU), a visual processing unit (VPU), a read only memory (ROM), a random access memory (RAM), etc., which are not shown. The control unit 21 can perform various processes by using the RAM as a work area and executing programs read from the ROM, the storage unit 22, etc., with the CPU and the VPU.

[0020] The control unit 21 also has a self-management circuit (not shown) that is capable of detecting an abnormality in the AI ​​edge device 11 based on information such as the operating state, connection state, or temperature of each unit. For example, the control unit 21 monitors the temperatures of the CPU, VPU, etc., and can detect the occurrence of an abnormality when these temperatures exceed a predetermined threshold value.

[0021] The storage unit 22 has a non-volatile storage medium such as an SSD (Solid State Drive) or a hard disk drive, and is also called storage. The storage unit 22 stores various programs, setting information, and the like, as well as various data acquired from the sensor device 12 and analysis data generated based on the data. The communication unit 23 is an interface such as a wired LAN conforming to a standard such as IEEE802.3. The communication unit 23 transmits and receives various information to and from the network 10 in accordance with a predetermined protocol. The sensor connection unit 24 is an interface conforming to a wired or wireless communication standard such as USB (Universal Serial Bus) or Bluetooth (registered trademark).

[0022] Moreover, the various hardware and software in the AI ​​edge device 11 form a hierarchical structure as shown in Fig. 3. That is, in the AI ​​edge device 11, a host OS (Operating System) 42 runs on the lowest-layer hardware 41. On this host OS, a container platform 43, an IoT Edge runtime 44, and a setting / operation WebUI / SDK (Web User Interface / Software Development Kit) 45 run, and further a container 46 runs on them.

[0023] The container 46 is a mechanism for constructing a virtual machine and then executing various applications, and executes a guest OS 51, on which a library 52 and a trained model 53 are placed, and an application 54 is executed. Incidentally, the AI ​​edge device 11 is capable of executing various containers 46, and is also capable of executing multiple containers 46 in parallel.

[0024] This AI edge device 11 executes a container 46 suited to the type and content of data supplied from the sensor device 12, thereby performing arithmetic processing such as appropriate analysis of the data as so-called AI processing, and as a result, can generate analytical data. After that, the AI ​​edge device 11 transmits the analytical data thus generated to a predetermined data server device (not shown) connected to the wide area network 4 for storage.

[0025] The AI ​​edge device 11 can switch between an operating state and a standby state as an operating state. The operating state is a state in which the container 46 is executed to perform analysis processing and the like on data supplied from the sensor device 12. The standby state is a state in which the container 46 can be executed immediately but in which analysis processing is not actually performed.

[0026] The sensor device 12 (FIG. 1) has one or more types of sensors (not shown), such as a temperature sensor, a humidity sensor, an illuminance sensor, an ultrasonic sensor, a tilt sensor, a pressure sensor, a human presence sensor, a smoke sensor, a static electricity sensor, etc., and generates data based on the results detected by these various sensors.

[0027] The sensor device 12 also has an interface that complies with a wired or wireless communication standard, such as USB or Bluetooth (registered trademark), and is connected to the AI ​​edge device 11. Therefore, the sensor device 12 can transmit generated data to the connected AI edge device 11 by communication processing that complies with these communication standards.

[0028] The router 13 performs address conversion processing, packet conversion processing, etc. between the wide area network 4 and the network 10 within the installation site 2, thereby allowing information to be sent and received between each device connected to the wide area network 4 and each device connected to the network 10.

[0029] The management center 3 is provided with a network 30, a management server 31, a management terminal device 32, and a router 33. Similar to the network 10 at the installation site 2, the network 30 connects the management server 31, the management terminal device 32, and the router 33 via a wired LAN conforming to a standard such as IEEE802.3.

[0030] 4, the management server 31 as a management device has a configuration in which a control unit 61, a storage unit 62, and a communication unit 63 are connected to a bus 60. The control unit 61 has a CPU, a ROM, a RAM, etc., which are not shown. The control unit 61 can perform various processes by using the CPU to execute programs read from the ROM, the storage unit 62, etc., while using the RAM as a work area.

[0031] Moreover, the control unit 61 reads out and executes a predetermined program from the storage unit 62, thereby forming therein a number of functional blocks, such as a state management unit 71 and a switching management unit 72. The state management unit 71 manages the state of each AI edge device 11 installed at the installation site 2 (described in detail below). The switching management unit 72 manages the switching of the operating state of each AI edge device 11 installed at the installation site 2 (described in detail below).

[0032] The storage unit 62 has a non-volatile storage medium such as an SSD or a hard disk drive, and stores various programs, setting information, etc. The storage unit 62 is provided with a status management table 81 and a container storage unit 82. As shown in Fig. 5, the status management table 81 stores information regarding the status of each AI edge device 11 for each item, such as IP address, operating state, sensor (connected sensor device 12), and container (running container 46). The container storage unit 82 stores the container 46 run by each AI edge device 11.

[0033] The management terminal device 32 (FIG. 1), like a general computer device, has a control unit, a memory unit, a communication unit, a display unit, and an operation unit (none of which are shown), etc. This management terminal device 32 performs processing such as displaying the operation status of each AI edge device 11 on the display unit based on information in the status management table 81, or switching the operation status of each AI edge device 11 according to operation instructions from staff of the management center 3, etc.

[0034] The router 33 is configured in the same manner as the router 13 at the installation site 2, and performs address conversion processing, packet conversion processing, etc. between the wide area network 4 and the network 30 within the management center 3, thereby allowing information to be sent and received between each device connected to the wide area network 4 and each device connected to the network 30.

[0035] The wide area network 4 is a network that is deployed over a wide area, such as the Internet, and is connected to the installation site 2, the management center 3, and various server devices (not shown). The wide area network 4 is configured by appropriately combining various communication methods, such as a wired LAN conforming to a standard such as IEEE802.3, a wireless LAN conforming to a standard such as IEEE802.11 (IEEE802.11a / b / g / n / ac / ax), or a fifth generation mobile communication system (5G: 5th Generation).

[0036] [1-2. Table update process] In the processing system 1, when the status or the like of each AI edge device 11 changes, this is updated to reflect the status management table 81 (FIG. 5), and the latest information is stored in the status management table 81. Below, the table update process in the processing system 1 will be described with reference to the sequence chart in FIG. 6, taking as an example a case where the status or the like of the AI ​​edge device 11 has changed.

[0037] In the AI ​​edge device 11, the occurrence of some change, such as when a new container 46 starts to run or when the connected sensor device 12 is changed, is set as an information transmission condition and stored in the storage unit 22. When this information transmission condition is satisfied, the control unit 21 (FIG. 2) of the AI ​​edge device 11A reads and executes a predetermined information transmission program from the storage unit 22, thereby starting an information transmission processing procedure RT11 and proceeding to the first step SP11.

[0038] In step SP11, the control unit 21 transmits status information to the management server 31. This status information includes information on the operating state of the AI ​​edge device 11A, the container 46 being executed, and the sensor device 12 being connected.

[0039] Meanwhile, the control unit 61 of the management server 31 has already started the table update processing procedure RT12 by reading and executing a predetermined table update program from the memory unit 62, and when it receives status information from the AI ​​edge device 11A, it proceeds to the first step SP21.

[0040] In step SP21, the control unit 61 updates the status management table 81 (FIG. 5) in the storage unit 62 using the status management unit 71 (FIG. 4) based on the acquired status information, and proceeds to the next step SP22. In step SP22, the control unit 61 causes the status management unit 71 to transmit a status information confirmation notification to the AI ​​edge device 11A notifying that the status information has been received and the status management table 81 has been updated. Thereafter, the control unit 61 returns to the initial step SP21 and repeats the series of processes.

[0041] When the control unit 21 of the AI ​​edge device 11A receives the status information confirmation notification from the management server 31, the control unit 21 proceeds to the next step SP12 and ends the information transmission processing procedure RT11. Incidentally, the control unit 21 executes the information transmission processing procedure RT11 at predetermined time intervals or whenever the status of the AI ​​edge device 11A changes.

[0042] [1-3. State switching process] In the processing system 1, as shown in the status management table 81 (FIG. 5), some of the multiple AI edge devices 11 installed at the installation site 2 are in an operating state, and the rest are in a standby state. In this processing system 1, if an abnormality occurs in an AI edge device 11 that is in an operating state and executing a specific container 46, under the management of the management server 31, the other AI edge device 11 that belongs to the same set and is in a standby state is switched to an operating state and takes over the AI ​​processing that was being executed (hereinafter also referred to as the target AI processing).

[0043] In the following, the operation switching process in the processing system 1 will be described with reference to the sequence chart in Fig. 7, taking as an example a case in which an abnormality occurs in the AI ​​edge device 11A in operation because the temperature of the VPU exceeds a threshold, and the AI ​​edge device 11B in standby mode takes over the processing. In the following, the AI ​​edge device 11A will also be referred to as the switching source, and the AI ​​edge device 11B will also be referred to as the switching destination.

[0044] When the control unit 21 (FIG. 2, hereinafter also referred to as the switching source control unit 21A) of the switching source AI edge device 11A detects the occurrence of an abnormality, it reads out a predetermined switching source program from the storage unit 22 and executes it to start the switching source processing procedure RT13 (FIG. 7) and proceeds to the first step SP31. In step SP31, the switching source control unit 21A transmits a switching request to the management server 31.

[0045] Meanwhile, the control unit 61 (Figure 4) of the management server 31 has already started the switching management processing procedure RT14 (Figure 7) by reading and executing a predetermined switching management program from the memory unit 62, and when it receives a switching request from the AI ​​edge device 11A, it proceeds to the first step SP41.

[0046] In step SP41, the control unit 61 refers to the status management table 81 (FIG. 5) of the memory unit 62 via the switching management unit 72 (FIG. 4), selects the AI ​​edge device 11 to be the switching destination, and proceeds to the next step SP42. Specifically, the switching management unit 72 selects, as the switching destination AI edge device 11, the AI ​​edge device 11B that is in a standby state in the status management table 81 and is connected to the same sensor device 12 (i.e., sensor device 12A) as the switching source AI edge device 11A.

[0047] In step SP42, the control unit 61 transmits a state switching instruction to the AI ​​edge device 11A via the switching management unit 72. This state switching instruction notifies that an appropriate AI edge device 11 to be the switching destination has been selected, and includes an instruction to switch the operating state from an active state to a standby state, etc.

[0048] When the switching source control unit 21A of the AI ​​edge device 11A receives this state switching instruction, it proceeds to the next step SP32, switches the operating state from an active state to a standby state, and proceeds to the next step SP33. In step SP33, the switching source control unit 21A transmits a state switching completion notification to the management server 31 to notify that switching of the operating state has been completed, and then proceeds to the next step SP34, where it ends the switching source processing procedure RT13.

[0049] When the control unit 61 of the management server 31 receives a state switching completion notification from the AI ​​edge device 11A, it proceeds to the next step SP43, and the switching management unit 72 (Figure 4) reads out from the container memory unit 82 of the memory unit 62 the same container 46A that was being executed in the original AI edge device 11A, and sends it to the destination AI edge device 11B.

[0050] The control unit 21 of the AI ​​edge device 11B, which is the switching destination (hereinafter also referred to as the switching destination control unit 21B), has previously started the switching destination processing procedure RT15 by reading and executing the switching destination program from the memory unit 22, and when it receives the container 46A from the management server 31, it proceeds to step SP51.

[0051] In step SP51, the switching destination control unit 21B makes the received container 46A executable, that is, performs so-called deployment, and proceeds to the next step SP52. In step SP52, the switching destination control unit 21B transmits a deployment completion notification to the management server 31 to notify that the deployment has been completed.

[0052] When the control unit 61 of the management server 31 receives a deployment completion notification from the AI ​​edge device 11B, it proceeds to the next step SP44, and sends a state switching instruction to the AI ​​edge device 11B via the switching management unit 72 (Figure 4) to switch the operating state from a standby state to an operating state.

[0053] When the switching destination control unit 21B of the AI ​​edge device 11B receives the state switching instruction from the management server 31, it proceeds to the next step SP53, switches the operating state from the standby state to the operating state, and proceeds to the next step SP54. As a result, the AI ​​edge device 11B starts execution of AI processing by the container 46A based on the data supplied from the sensor device 12A.

[0054] In step SP54, the switching destination control unit 21B transmits a state switching completion notification to the management server 31 to notify that switching of the operating state has been completed, and then proceeds to the next step SP55, where the switching destination processing procedure RT15 is terminated. Thereafter, the AI ​​edge device 11B continues to execute AI processing by the container 46A based on the data supplied from the sensor device 12A.

[0055] When the control unit 61 of the management server 31 receives the state switching completion notification from the AI ​​edge device 11B, the control unit 61 proceeds to the next step SP45, and updates the state management table 81 (FIG. 5) using the state management unit 71 (FIG. 4). Specifically, as shown in FIG. 8, the state management unit 71 switches the operating state of the AI ​​edge device 11A to a standby state in the state management table 81, and switches the operating state of the AI ​​edge device 11B to an operating state, and updates the container item to container 46A. After that, the control unit 61 proceeds to the next step SP46, and ends the switching management processing procedure RT14.

[0056] [1-4. Effects, etc.] In the above configuration, in the processing system 1 according to the first embodiment, when the AI ​​edge device 11A in an operating state detects an abnormality, the management server 31 receives a switching request from the AI ​​edge device 11A, selects the AI ​​edge device 11B in a standby state as the switching destination, and takes over and executes the AI ​​processing.

[0057] That is, the management server 31 switches the source AI edge device 11A to a standby state, transmits the container 46A that was being executed in the AI ​​edge device 11A to the destination AI edge device 11B, and switches the destination AI edge device 11B to an operating state when the deployment is complete. As a result, in the processing system 1, if an abnormality occurs in the AI ​​edge device 11A, the container 46A can be immediately deployed and executed by the AI ​​edge device 11B selected as the destination, thereby making it possible to smoothly continue AI processing of data generated by the sensor device 12A.

[0058] Generally, if an abnormality occurs in an AI edge device and the abnormality cannot be resolved even by restarting the device, maintenance work such as repair or replacement is required. In this regard, in the processing system 1 according to the present embodiment, in such a case, the AI ​​edge device 11B to be switched to execute the container 46A can be switched to immediately resume AI processing of the data generated by the sensor device 12A. Therefore, in the processing system 1, there is almost no risk of missing out on data generated by the sensor device 12A until the maintenance work on the AI ​​edge device 11A is completed, and it is also possible to set a schedule with ample time for the maintenance work on the AI ​​edge device 11A.

[0059] Furthermore, the management server 31 uses a status management table 81 (FIG. 5) to manage information related to the operating status of each AI edge device 11, the connected sensor devices 12, the operating containers 46, etc. Therefore, when the processing system 1 receives a switching request from the AI ​​edge device 11A in which an abnormality has occurred, it can immediately select the AI ​​edge device 11B that is the most appropriate switching destination in an extremely short time by referring to this status management table 81, and can start AI processing by the AI ​​edge device 11B early. In other words, the processing system 1 can effectively avoid missing data supplied from the sensor devices 12, and can supply analysis data obtained by AI processing with as little delay as possible.

[0060] Furthermore, in the processing system 1, after receiving a switching request, the container 46 is transmitted to the switching destination AI edge device 11B to be deployed. Therefore, in the processing system 1, when the container 46 is frequently updated, it is sufficient to store the latest container 46 in the container storage unit 82 (FIG. 4) of the storage unit 62 in the AI ​​edge device 11 that executes the container 46 and in the management server 31. In other words, in the processing system 1, it is not necessary to perform wasteful processing such as updating the AI ​​edge device 11 that does not execute the container 46 every time the container 46 is updated.

[0061] According to the above configuration, in the processing system 1 according to the first embodiment, when the AI ​​edge device 11A in an operating state detects an abnormality, the AI ​​edge device 11A is switched to a standby state under the management of the management server 31, the standby AI edge device 11B is selected as the switching destination, the container 46A is deployed, and the AI ​​edge device 11B is switched to the operating state. Therefore, in the processing system 1, even if the AI ​​processing cannot be continued in the switching source AI edge device 11A, the switching process to the switching destination AI edge device 11B can be completed in a relatively short time, and the AI ​​processing can be immediately started to resume the generation and supply of analysis data.

[0062] 2. Second embodiment [2-1. Processing system configuration] As shown in FIG. 9 corresponding to FIG. 1, a processing system 201 according to the second embodiment is composed of an installation site 202 and a management center 203 instead of the installation site 2 and the management center 3, and a wide area network 4.

[0063] The installation site 202 differs from the installation site 2 (FIG. 1) according to the first embodiment in that it is provided with AI edge devices 211 (211A, 211B, and 211C) and sensor devices 212 (212A, 212B, and 212C) instead of the AI ​​edge device 11 and the sensor device 12. Furthermore, at the installation site 202, each sensor device 212 is connected to one AI edge device 211.

[0064] The AI ​​edge device 211 (FIG. 2) differs from the AI ​​edge device 11 according to the first embodiment in that it has a control unit 221 instead of the control unit 21, but is otherwise configured in a similar manner. Like the control unit 21 according to the first embodiment, the control unit 221 has a CPU, VPU, ROM, RAM, etc. (not shown), and can perform various processes by executing various programs using the CPU and VPU.

[0065] The sensor device 212, like the sensor device 12 according to the first embodiment, has various sensors (not shown), generates data based on the results detected by the various sensors, and supplies the data to the AI ​​edge device 211.

[0066] The management center 203 differs from the management center 3 (FIG. 1) according to the first embodiment in that it has a management server 231 instead of the management server 31. As shown in FIG. 10 corresponding to FIG. 4, the management server 231 differs in that it has a control unit 261 and a storage unit 262 instead of the control unit 61 and the storage unit 62, but is otherwise configured in the same way.

[0067] The control unit 261, like the control unit 61 according to the first embodiment, has a CPU, a ROM, a RAM, etc. (not shown). This control unit 261 can perform various processes by using the CPU to execute programs read from the ROM, the storage unit 262, etc. while using the RAM as a work area.

[0068] Moreover, the control unit 261 reads out and executes a predetermined program from the storage unit 262, thereby forming multiple functional blocks therein, such as a status management unit 71 and an allocation management unit 273. As in the first embodiment, the status management unit 71 manages the status of each AI edge device 211 installed at the installation site 202. The allocation management unit 273 manages the allocation of AI processing to each AI edge device 311 (described in detail later).

[0069] The storage unit 262, like the storage unit 62 according to the first embodiment, has a non-volatile storage medium such as an SSD or a hard disk drive, and stores various programs, setting information, etc. The storage unit 262 is provided with a state management table 281 and a container storage unit 82.

[0070] 5, the status management table 281 stores information on the status of each AI edge device 211 for each item, such as IP address, utilization rate of various resources (CPU, VPU, memory, and storage), running container 1, and running container 2 (running container 46). The container storage unit 82 stores the containers 46 executed by each AI edge device 211, as in the first embodiment.

[0071] In addition, in the processing system 201, as in the first embodiment, a table update process (Figure 6) is performed to sequentially update the status management table 281 (Figure 11) based on the status information transmitted from each AI edge device 211.

[0072] [2-2. Allocation change processing] However, in the processing system 201, unlike the first embodiment, when the utilization rate of various resources increases in any of the AI ​​edge devices 211, the allocation is changed so that the AI ​​processing being executed in that AI edge device 211 is distributed and executed together with other AI edge devices 211.

[0073] In the AI ​​edge device 211, a resource condition for determining whether or not the resource usage rate is high is set in advance and stored in the storage unit 22 (FIG. 2). For example, this resource condition is set to satisfy any one of "the average VPU usage rate is 95% or more," "the average CPU usage rate is 95% or more," or "the memory usage rate is 90% or more." In other words, determining that this resource condition is satisfied means that a so-called resource shortage is detected. In addition, the AI ​​edge device 211 constantly monitors the usage rate of its own resources and determines whether or not the resource condition is satisfied.

[0074] The following describes the allocation change process in the processing system 201 with reference to the sequence chart in Fig. 12, taking as an example a case in which the resource usage rate of the AI ​​edge device 211A executing AI processing increases, satisfying the resource conditions, and part of the AI ​​processing is allocated to the AI ​​edge device 211B. In the following, the AI ​​edge device 211A is also referred to as the allocation source, and the AI ​​edge device 211B is also referred to as the allocation destination.

[0075] When the control unit 221 (FIG. 2, hereinafter also referred to as the allocation source control unit 221A) of the allocation source AI edge device 211A determines that the resource condition is satisfied, it reads out and executes a predetermined allocation source program from the storage unit 22, thereby starting the allocation source processing procedure RT23 and proceeding to the first step SP231. In step SP231, the allocation source control unit 221A transmits an allocation request to the management server 231.

[0076] Meanwhile, the control unit 261 of the management server 231 has already started the allocation management processing procedure RT24 by reading and executing a predetermined allocation management program from the memory unit 262, and when it receives an allocation request from the AI ​​edge device 211A, it proceeds to the first step SP241.

[0077] In step SP241, the control unit 261 causes the allocation management unit 273 (FIG. 10) to refer to the status management table 281 (FIG. 11) in the storage unit 262 to select an AI edge device 211 to be allocated to, and proceeds to the next step SP242. Specifically, the allocation management unit 273 selects the AI ​​edge device 211B in the status management table 281 as the AI ​​edge device 211 with the lowest VPU usage rate, for example.

[0078] In step SP242, the control unit 261 transmits a distributed allocation notification to the AI ​​edge device 211A via the allocation management unit 273, and proceeds to the next step SP243. This distributed allocation notification notifies that the AI ​​edge device 211B has been selected as the allocation destination, and includes an instruction to prepare allocation data to be transmitted to the allocation destination, etc.

[0079] When the allocation control unit 221A of the AI ​​edge device 211A receives this allocation information, the process proceeds to the next step SP232, where the allocation control unit 221A prepares allocation data to be transmitted to the allocation destination. This allocation data is a portion of the data to be processed by the AI ​​edge device 211A that is to be shared by the allocation destination AI edge device 211B.

[0080] In step SP243, the control unit 261, via the allocation management unit 273 (Figure 10), reads out the same container 46A as that being executed in the allocation source AI edge device 211A from the container memory unit 82 of the memory unit 262, and transmits it to the switching destination AI edge device 211B.

[0081] The control unit 221 of the AI ​​edge device 211B to be assigned (hereinafter also referred to as the assignment control unit 221B) has previously started the assignment processing procedure RT25 by reading and executing the assignment program from the memory unit 222, and when it receives the container 46A from the management server 231, it proceeds to step SP251.

[0082] In step SP251, the allocation control unit 221B makes the received container 46A executable, that is, performs so-called deployment, and proceeds to the next step SP252. In step SP252, the allocation control unit 221B transmits a deployment completion notification to the management server 231 to notify that the deployment has been completed.

[0083] When the control unit 261 of the management server 231 receives a deployment completion notification from the AI ​​edge device 211B, it proceeds to the next step SP244, and the allocation management unit 273 (Figure 10) sends a distributed allocation completion notification to the AI ​​edge device 211A that originated the allocation to start distributed processing, and then proceeds to the next step SP245.

[0084] In step SP245, the control section 261 updates the status management table 281 (FIG. 11) by the status management section 71 (FIG. 10), and then proceeds to the next step SP246 to terminate the allocation management processing procedure RT24.

[0085] When the allocation source control unit 221A of the AI ​​edge device 211A receives the distributed allocation completion notification from the management server 231, the process proceeds to step SP233 and transmits allocation data to the switching destination AI edge device 211B.

[0086] When the assignment destination control unit 221B of the AI ​​edge device 211B receives the assignment data from the assignment source AI edge device 211A, the process proceeds to step SP253, where it performs AI processing using the container 46A based on the assignment data, and transmits the obtained analysis data to the assignment source AI edge device 211A. After that, the assignment destination control unit 221B proceeds to the next step SP254, where it ends the assignment destination processing procedure RT25.

[0087] When the allocation source control unit 221A of the AI ​​edge device 211A receives this analysis data, it stores it in the memory unit 22 (Figure 2) or transmits it to a specified destination, and then proceeds to the next step SP234 to terminate the allocation source processing procedure RT23.

[0088] [2-3. Effects, etc.] In the above configuration, in the processing system 201 according to the second embodiment, when the resource usage rate of the AI ​​edge device 211A executing AI processing increases and satisfies the resource condition, the management server 231 receives an allocation request from the AI ​​edge device 211A and selects another AI edge device 211B to distribute and execute the AI ​​processing.

[0089] That is, the management server 231 selects the AI ​​edge device 211B with the lowest VPU usage rate as the allocation destination, and transmits the container 46A that was being executed on the allocation source AI edge device 211A to the allocation destination AI edge device 211B for deployment. Next, when allocation data is transmitted from the allocation source AI edge device 211A, the allocation destination AI edge device 211B executes AI processing based on this allocation data, and transmits the obtained analysis data to the allocation source AI edge device 211A.

[0090] As a result, in the processing system 201, when resource usage increases in the AI ​​edge device 211A, the container 46A can be immediately deployed by the AI ​​edge device 211B selected as the allocation destination to share and execute part of the AI ​​processing, thereby distributing the processing. As a result, in the processing system 201, delays in AI processing and data overflows in the AI ​​edge device 211A can be prevented in advance, and AI processing can proceed smoothly.

[0091] Furthermore, in the processing system 201, the management server 231 stores a status management table 281 (FIG. 10) in the memory unit 262 and sequentially updates the status management table 281 based on status information transmitted from each AI edge device 211. Therefore, in the processing system 201, when the management server 231 receives an allocation request from an AI edge device 211, it can check the status of each resource in the status management table 281 and select the most appropriate allocation destination.

[0092] According to the above configuration, in the processing system 201 according to the second embodiment, when the AI ​​edge device 211A transmits an allocation request due to an increase in resource usage, the AI ​​edge device 211B with a low resource usage rate is selected as the allocation destination under the management of the management server 231, and the container 46A is deployed and then the AI ​​edge device 211B shares part of the AI ​​processing. Therefore, in the processing system 201, the AI ​​edge device 211B to which the allocation destination is switched can share the AI ​​processing in a relatively short time, and the generation of analysis data can be continued while preventing delays and overflows in the AI ​​processing in the AI ​​edge device 211A from which the allocation was switched.

[0093] 3. Third embodiment [3-1. Processing system configuration] As shown in FIG. 13 corresponding to FIG. 1, a processing system 301 according to the third embodiment is composed of an installation site 302 and a management center 303 instead of the installation site 2 and the management center 3, and a wide area network 4.

[0094] The installation site 302 differs from the installation site 2 (FIG. 1) according to the first embodiment in that it is provided with AI edge devices 311 (311A, 311B, and 311C) and sensor devices 312 (312A, 312B, and 312C) instead of the AI ​​edge device 11 and sensor device 12. Furthermore, at the installation site 302, each AI edge device 311 and each sensor device 312 is connected to a network 310 configured separately from the network 10.

[0095] This network 310 is a wireless network that complies with standards such as IEEE802.15.4, and uses a radio frequency band of, for example, 920 MHz, and can transmit and receive various data in packets of relatively small size, for example, a maximum of 127 bytes.

[0096] The AI ​​edge device 311 (FIG. 2) differs from the AI ​​edge device 11 according to the first embodiment in that it has a control unit 321 and a sensor connection unit 324 instead of the control unit 21 and the sensor connection unit 24, but is otherwise configured in the same manner. Like the control unit 21 according to the first embodiment, the control unit 321 has a CPU, VPU, ROM, RAM, etc. (not shown), and can perform various processes by executing various programs with the CPU and VPU. The sensor connection unit 324 is a wireless interface that complies with standards such as IEEE802.15.4, and can transmit and receive information to and from the network 310.

[0097] The sensor device 312, like the sensor device 12 in the first embodiment, has various sensors not shown, generates data based on the results detected by the various sensors, and transmits this data to the AI ​​edge device 311 via the network 310.

[0098] The management center 303 differs from the management center 3 (FIG. 1) according to the first embodiment in that it has a management server 331 instead of the management server 31. As shown in FIG. 14 corresponding to FIG. 4, the management server 331 differs in that it has a control unit 361 and a storage unit 362 instead of the control unit 61 and the storage unit 62, but is otherwise configured in the same way.

[0099] The control unit 361, like the control unit 61 according to the first embodiment, has a CPU, a ROM, a RAM, etc. (not shown). This control unit 361 can perform various processes by using the CPU to execute programs read from the ROM, the storage unit 362, etc. while using the RAM as a work area.

[0100] Moreover, the control unit 361 reads out and executes a predetermined program from the memory unit 362, thereby forming multiple functional blocks therein, such as a state management unit 71, a switching management unit 72, and an allocation management unit 273. As in the first embodiment, the state management unit 71 and the switching management unit 72 each manage the state of each AI edge device 311 and switch the operating state of each AI edge device 311. As in the second embodiment, the allocation management unit 273 manages the allocation of AI processing to each AI edge device 311.

[0101] The storage unit 362, like the storage unit 62 according to the first embodiment, has a non-volatile storage medium such as an SSD or a hard disk drive, and stores various programs, setting information, etc. The storage unit 362 is provided with a state management table 381 and a container storage unit 82.

[0102] As shown in Fig. 15, which corresponds to Fig. 5 and Fig. 11, the status management table 381 has a configuration in which the status management table 81 (Fig. 5) and the status management table 281 (Fig. 11) are integrated. That is, the status management table 381 stores information on the status of each AI edge device 311 for each item, such as the IP address, the utilization rate of various resources (CPU, VPU, memory, and storage), the operating state, the connected sensors, and the operating containers. The container storage unit 82 stores the containers 46 executed by each AI edge device 311, as in the first embodiment.

[0103] In addition, in the processing system 301, as in the first and second embodiments, a table update process (Figure 6) is performed to sequentially update the status management table 381 (Figure 15) based on the status information transmitted from each AI edge device 311.

[0104] [3-2. Management process] Incidentally, in the processing system 301, with regard to the management of each AI edge device 311 by the management server 331, processing is performed that combines the first and second embodiments.

[0105] That is, in the processing system 301, if an abnormality occurs in an operating AI edge device 311, the processing is handed over to another AI edge device 311 in a standby state and executed, as in the first embodiment. Also, in the processing system 301, if the resource usage rate of an AI edge device 311 executing AI processing increases, the processing system 301 distributes the AI ​​processing to be executed by the other AI edge devices 311.

[0106] For this reason, in the processing system 301, the AI ​​edge device 311 executes the switching allocation request process shown in Fig. 16, and the management server 331 executes the switching allocation management process shown in Fig. 17. In the following, the AI ​​edge device 311A ​​will be described as an example.

[0107] Specifically, the control unit 321 (FIG. 2) of the AI ​​edge device 311A ​​starts the switching allocation request process RT36 (FIG. 16) by reading a predetermined program from the storage unit 22, and proceeds to the first step SP361.

[0108] In step SP361, the control unit 321 determines whether or not an abnormality has been detected in the AI ​​edge device 311A. If a positive result is obtained here, this indicates that the AI ​​processing that has been executed in the AI ​​edge device 311A ​​so far needs to be switched to be executed by another AI edge device 311. In this case, the control unit 321 proceeds to the next step SP362.

[0109] In step SP362, the control unit 321 executes, as a subroutine, the switching source processing procedure RT13 (FIG. 7) similar to that in the first embodiment, and proceeds to the next step SP365. This allows the control unit 321 to switch to another AI edge device 311 (e.g., AI edge device 311B) and have it execute the AI ​​processing, that is, to take over the AI ​​processing, through the switching management processing of the management server 331, as described below.

[0110] On the other hand, if a negative result is obtained in step SP361, this indicates that the AI ​​processing by the AI ​​edge device 311A ​​should be continued. In this case, the control unit 321 proceeds to the next step SP363.

[0111] In step SP363, the control unit 321 determines whether or not the AI ​​edge device 311 has insufficient resources, specifically, whether or not the resource conditions in the second embodiment are satisfied. If a positive result is obtained here, this indicates that part of the AI ​​processing being executed in the AI ​​edge device 311A ​​needs to be assigned to another AI edge device 311 for execution, that is, that the AI ​​processing needs to be distributed and executed. In this case, the control unit 321 proceeds to the next step SP364.

[0112] In step SP364, the control unit 321 executes, as a subroutine, the allocation source processing procedure RT23 (FIG. 12) similar to that in the second embodiment, and proceeds to the next step SP365. As a result, the control unit 321 can have another AI edge device 311 (e.g., the AI ​​edge device 311B) share part of the AI ​​processing through the allocation management process of the management server 331, as will be described later.

[0113] On the other hand, if a negative result is obtained in step SP363, this indicates that the AI ​​processing by the AI ​​edge device 311A ​​should be continued. In this case, the control unit 321 proceeds to the next step SP365 and ends the switching allocation request processing RT36. In step SP365, the control unit 321 ends the switching allocation request processing RT36. Incidentally, the control unit 321 is configured to repeatedly execute the switching allocation request processing RT36 at predetermined intervals.

[0114] Moreover, the control unit 361 (FIG. 14) of the management server 331 starts the switching allocation management process RT37 (FIG. 17) by reading a predetermined program from the storage unit 362, and proceeds to the first step SP371.

[0115] In step SP371, the control unit 361 determines whether or not a switching request has been received from any of the AI ​​edge devices 311. If a positive result is obtained here, this indicates that the AI ​​processing that has been executed in that AI edge device 311 (e.g., AI edge device 311A) needs to be switched to be executed by another AI edge device 311. In this case, the control unit 361 proceeds to the next step SP372.

[0116] In step SP372, the control unit 361 executes, as a subroutine, a switching management processing procedure RT14 (FIG. 7) similar to that in the first embodiment, and proceeds to the next step SP365. This enables the control unit 361 to switch the AI ​​processing that was previously executed by the AI ​​edge device 311A ​​to another AI edge device 311 (e.g., the AI ​​edge device 311B) for execution, i.e., to have it take over.

[0117] On the other hand, if a negative result is obtained in step SP371, this indicates that the AI ​​processing should be continued by each AI edge device 311. In this case, the control unit 321 proceeds to the next step SP373.

[0118] In step SP373, the control unit 321 determines whether or not an allocation request has been received from any of the AI ​​edge devices 311. If a positive result is obtained here, this indicates that part of the AI ​​processing that has been executed in that AI edge device 311 (e.g., AI edge device 311A) needs to be assigned to another AI edge device 311 for execution, i.e., the processing needs to be shared. In this case, the control unit 361 proceeds to the next step SP374.

[0119] In step SP374, the control unit 361 executes, as a subroutine, an allocation management processing procedure RT24 (FIG. 12) similar to that of the second embodiment, and proceeds to the next step SP375. This enables the control unit 361 to allocate part of the AI ​​processing that has been executed by the AI ​​edge device 311A ​​to another AI edge device 311 (e.g., the AI ​​edge device 311B) for execution, that is, to distribute the processing.

[0120] On the other hand, if a negative result is obtained in step SP373, this indicates that the AI ​​processing by each AI edge device 311 should be continued. In this case, the control unit 361 proceeds to the next step SP375. In step SP375, the control unit 361 ends the switching allocation management process RT37. Incidentally, the control unit 361 is configured to repeatedly execute the switching allocation management process RT37 at predetermined intervals.

[0121] [3-3. Effects, etc.] In the above configuration, in the processing system 301 according to the third embodiment, if the AI ​​edge device 311A ​​in operation detects an abnormality, the management server 331 selects the AI ​​edge device 311B in standby as the switchover destination and has it take over and execute the AI ​​processing. Also, in the processing system 301, if the resource usage rate of the AI ​​edge device 311A ​​executing the AI ​​processing increases, the management server 331 selects another AI edge device 311B and distributes the AI ​​processing to be executed.

[0122] As a result, in the processing system 301, if an abnormality occurs in the AI ​​edge device 311A, the container 46A can be immediately deployed and executed by the AI ​​edge device 311B selected as the switching destination, thereby allowing AI processing of the data generated by the sensor device 312 to continue.

[0123] Furthermore, in the processing system 301, when resource usage increases in the AI ​​edge device 311A, the container 46A can be immediately deployed to the AI ​​edge device 311B selected as the allocation destination to share and execute part of the AI ​​processing. As a result, the processing system 301 can prevent delays in AI processing and data overflows in the AI ​​edge device 311A, enabling the AI ​​processing to proceed smoothly.

[0124] In other respects as well, the processing system 301 can achieve the same advantageous effects as the first and second embodiments.

[0125] According to the above configuration, in the processing system 301 according to the third embodiment, when the AI ​​edge device 311A ​​detects an abnormality, the processing system 301 switches to another AI edge device 311B as the switching destination to execute AI processing under the management of the management server 331. Furthermore, in the processing system 301, when the resource usage rate of the AI ​​edge device 311A ​​increases, the processing system 301 distributes and executes the AI ​​processing to another AI edge device 311B as the allocation destination under the management of the management server 331. Therefore, in the processing system 301, when each AI edge device 311 cannot continue AI processing, the processing system 301 can immediately appropriately select another AI edge device 311 and perform switching or allocation, so that the generation and supply of analysis data can be smoothly continued with almost no interruption in AI processing.

[0126] 4. Fourth Embodiment [4-1. Processing system configuration] Compared to the third embodiment, a processing system 401 (FIG. 13) according to the fourth embodiment is composed of an installation site 402 and a management center 403 instead of the installation site 302 and the management center 303, and a wide area network 4.

[0127] The installation site 402 differs from the installation site 302 according to the third embodiment in that AI edge devices 411 (411A, 411B, and 411C) are provided instead of the AI ​​edge device 311.

[0128] The AI ​​edge device 411 (FIG. 2) differs from the AI ​​edge device 311 according to the third embodiment in that it has a control unit 421 instead of the control unit 321, but is otherwise configured similarly. Like the control unit 21 according to the first embodiment, the control unit 421 has a CPU, VPU, ROM, RAM, etc. (not shown), and can perform various processes by executing various programs with the CPU and VPU.

[0129] The management center 403 differs from the management center 303 according to the third embodiment in that it has a management server 431 instead of the management server 331. The management server 431 (FIG. 14) differs in that it has a control unit 461 and a storage unit 462 instead of the control unit 361 and the storage unit 362, but is otherwise configured in the same way.

[0130] The control unit 461, like the control unit 361 according to the third embodiment, has a CPU, a ROM, a RAM, etc. (not shown). This control unit 461 can perform various processes by using the CPU to execute programs read from the ROM, the storage unit 462, etc. while using the RAM as a work area.

[0131] Moreover, the control unit 461 reads out and executes a predetermined program from the memory unit 462, thereby forming multiple functional blocks therein, such as a state management unit 71, a switching management unit 72, and an allocation management unit 473. As in the first and third embodiments, the state management unit 71 and the switching management unit 72 manage the state of each AI edge device 311 and switch the operating state of each AI edge device 311. The allocation management unit 473 manages the allocation of AI processing to each AI edge device 411 using a method that is partially different from that of the allocation management unit 273 in the second and third embodiments.

[0132] The storage unit 462, like the storage unit 362 according to the third embodiment, has a non-volatile storage medium such as an SSD or a hard disk drive, and stores various programs and setting information that are partially different from those in the third embodiment. The storage unit 462 is provided with a status management table 481 that replaces the status management table 381 in the third embodiment, and a container storage unit 82.

[0133] Of these, the state management table 481, as shown in Fig. 18 corresponding to Fig. 15, has a processing capacity item in addition to the items provided in the state management table 381. This processing capacity is a numerical representation of the computational processing capacity of the control unit 421 of each AI edge device 411, and a larger numerical value indicates a higher processing capacity.

[0134] In addition, in the processing system 401, as in the first to third embodiments, a table update process (Figure 6) is performed to sequentially update the status management table 381 (Figure 15) based on the status information transmitted from each AI edge device 411.

[0135] [4-2. Management process] In the processing system 401, with regard to the management of each AI edge device 411 by the management server 431, processing is performed that is a combination of the first and second embodiments, but the processing is partially different from that of the third embodiment.

[0136] That is, in the case where an abnormality occurs in an AI edge device 411 in an operating state in the processing system 401, the processing is handed over to another AI edge device 411 in a standby state for execution, as in the first and third embodiments. On the other hand, in the processing system 401, when the resource usage rate of an AI edge device 411 that is executing AI processing increases, part of the AI ​​processing is assigned to another AI edge device 411 for execution (hereinafter referred to as distributed assignment processing), or the entirety of the AI ​​processing is assigned to execution (hereinafter referred to as collective assignment processing).

[0137] Among these, the distributed allocation process is almost the same as the allocation change process (FIG. 12) in the second embodiment, whereas the collective allocation process is partially different from the allocation change process in the second embodiment.

[0138] Here, specific processing will be described with reference to the sequence chart in Fig. 19, which corresponds to Fig. 12, for an example in which resource usage in the AI ​​edge device 411A increases and the AI ​​processing is assigned to and executed by the AI ​​edge device 411B. In the following, the AI ​​edge device 411A will also be referred to as the assignment source, and the AI ​​edge device 411B will also be referred to as the assignment destination.

[0139] When the control unit 421 of the allocation source AI edge device 411A determines that the resource conditions are satisfied, it executes the allocation source processing procedure RT43 by reading and executing a predetermined allocation source program from the storage unit 22. The control unit 461 of the management server 431 executes the allocation management processing procedure RT44 by reading and executing a predetermined allocation management program from the storage unit 462. The control unit 421 of the allocation destination AI edge device 411B executes the allocation destination processing procedure RT45 by reading and executing a predetermined allocation source program from the storage unit 22.

[0140] The control unit 421 of the AI ​​edge device 411A, which is the source of allocation, performs processing in the source of allocation processing procedure RT43 that is generally similar to that in the source of allocation processing procedure RT23 (Figure 12), but the allocation data prepared in step SP432 and the allocation data transmitted in step SP433 are not just a part of the AI ​​processing, but the entirety of it.

[0141] The control unit 461 of the management server 431 performs roughly the same processing as the allocation management processing procedure RT24 (FIG. 12) in the allocation management processing procedure RT44. However, the control unit 461 transmits a lump-sum allocation notification instead of a distributed allocation notification in step SP442, and transmits a lump-sum allocation completion notification instead of a distributed allocation completion notification in step SP444.

[0142] The control unit 421 of the AI ​​edge device 411B to be assigned performs processing in the assignment destination processing procedure RT45 that is generally similar to that of the assignment destination processing procedure RT25 (Figure 12), but terminates the assignment destination processing procedure RT45 without performing processing equivalent to step SP265.

[0143] Furthermore, in the processing system 401, the AI ​​edge device 411 executes a switching allocation request process shown in Fig. 20 which corresponds to Fig. 16, and the management server 431 executes a switching allocation management process shown in Fig. 21 which corresponds to Fig. 17. The following describes the AI ​​edge device 411A as an example.

[0144] Specifically, the control unit 421 (FIG. 2) of the AI ​​edge device 411A starts the switching allocation request processing RT46 (FIG. 20) by reading a predetermined program from the storage unit 22, and proceeds to the first step SP461. In steps SP461, SP462, and SP463, the control unit 421 performs the same processes as steps SP361, SP362, and SP363 of the switching allocation request processing RT36 (FIG. 16), respectively, thereby performing the switching source processing procedure RT13 (FIG. 7) as a subroutine when an abnormality occurs, and then proceeds to the next step SP468.

[0145] On the other hand, if a positive result is obtained in step SP463, the control unit 421 proceeds to the next step SP464, where it transmits an allocation request to the management server 431, as in step SP231 of the allocation source processing procedure RT23 (FIG. 12), and proceeds to the next step SP465. In response to this, the management server 431 determines whether to perform a divided allocation process or a lump-sum allocation process, as described below, and then transmits a divided allocation notification or a lump-sum allocation notification to the AI ​​edge device 411A.

[0146] In step SP465, the control unit 421 determines whether or not a divided allocation notification has been received from the management server 431. If a positive result is obtained here, this indicates that the AI ​​edge device 411A will continue the AI ​​processing and allocate part of the processing to another AI edge device 411. In this case, the control unit 421 proceeds to the next step SP466.

[0147] In step SP466, the control unit 421 performs a division allocation transition process, and proceeds to the next step SP468. Specifically, the control unit 421 executes, as a subroutine, the processes from step SP232 onwards in the allocation source processing procedure RT23 (FIG. 12) similar to that of the second embodiment.

[0148] On the other hand, if a negative result is obtained in step SP465, this indicates that the AI ​​processing in the AI ​​edge device 411A has been collectively assigned to another AI edge device 411, that is, the AI ​​processing in the AI ​​edge device 411A should be terminated. At this time, the control unit 421 proceeds to the next step SP467. In step SP467, the control unit 421 performs a collective assignment transition process, and proceeds to the next step SP468. Specifically, the control unit 421 executes, as a subroutine, the processes from step SP432 onwards in the above-mentioned assignment source processing procedure RT43 (FIG. 19).

[0149] In step SP468, the control unit 421 ends the switching allocation request process RT 46. Incidentally, the control unit 421 is configured to repeatedly execute the switching allocation request process RT 46 at predetermined intervals, as in the third embodiment.

[0150] Moreover, the control unit 461 (FIG. 14) of the management server 431 starts the switching allocation management process RT47 (FIG. 21) by reading a predetermined program from the storage unit 462, and proceeds to the first step SP471.

[0151] In steps SP471, SP472 and SP473, the control unit 461 performs processing similar to steps SP371, SP372 and SP373 of the switching allocation management processing RT37 (FIG. 17), respectively, thereby performing the switching management processing procedure RT14 (FIG. 7) as a subroutine when a switching request is received, and then proceeds to the next step SP478.

[0152] On the other hand, if a positive result is obtained in step SP473, the control unit 461 proceeds to the next step SP474, performs allocation method determination processing, and proceeds to the next step SP475. Specifically, the control unit 461 performs calculation processing in which the allocation management unit 473 (FIG. 14) refers to the usage rate and processing capacity of each resource in each AI edge device 411 stored in the state management table 481 and applies each value to a predetermined formula. The allocation management unit 473 then determines whether distributed allocation processing or collective allocation processing should be adopted for the AI ​​processing being executed in the AI ​​edge device 411A that is the allocation source.

[0153] In step SP475, the control unit 461 judges whether or not to perform distributed allocation processing, i.e., whether or not the result of the immediately preceding allocation method determination processing is distributed allocation processing. If a positive result is obtained here, the control unit 461 proceeds to the next step SP476, performs distributed allocation management processing, and then proceeds to the next step SP478. Specifically, the control unit 461 executes, as a subroutine, the processing from step SP242 onwards of the allocation management processing procedure RT24 (FIG. 12) similar to that of the second embodiment. This allows the control unit 461 to assign part of the AI ​​processing that has been executed by the AI ​​edge device 411A so far to another AI edge device 411 (e.g., the AI ​​edge device 411B) to be executed, i.e., distribute the processing.

[0154] On the other hand, if a negative result is obtained in step SP475, the control unit 461 proceeds to the next step SP477, performs a collective allocation management process, and proceeds to the next step SP478. Specifically, the control unit 461 executes the processes from step SP442 onwards in the above-mentioned allocation management process procedure RT44 (FIG. 19) as a subroutine. This allows the control unit 461 to allocate all of the AI ​​processing that has been executed by the AI ​​edge device 411A up until now to another AI edge device 411 (for example, the AI ​​edge device 411B) for execution, that is, to execute the processing collectively.

[0155] On the other hand, if a negative result is obtained in step SP473, this indicates that the AI ​​processing by each AI edge device 411 should be continued. In this case, the control unit 461 proceeds to the next step SP478. In step SP478, the control unit 461 ends the switching allocation management process RT47. Incidentally, the control unit 461 is configured to repeatedly execute the switching allocation management process RT47 at predetermined intervals.

[0156] [4-3. Effects, etc.] In the above configuration, in the processing system 401 according to the fourth embodiment, if the AI ​​edge device 411A in operation detects an abnormality, the management server 431 selects the AI ​​edge device 411B in standby as the switchover destination and has it take over and execute the AI ​​processing. Also, in the processing system 401, if the resource usage rate of the AI ​​edge device 411A executing the AI ​​processing increases, the management server 431 selects another AI edge device 411B and has it execute the AI ​​processing collectively or in a distributed manner.

[0157] As a result, in the processing system 401, if an abnormality occurs in the AI ​​edge device 411A, the container 46A can be immediately deployed and executed by the AI ​​edge device 411B selected as the switching destination, thereby allowing AI processing of the data generated by the sensor device 312 to continue.

[0158] Furthermore, in the processing system 401, when resource usage increases in the AI ​​edge device 411A, the container 46A can be immediately deployed by the AI ​​edge device 411B selected as the allocation destination to execute all or part of the AI ​​processing. Therefore, in the processing system 401, delays in AI processing and data overflows in the AI ​​edge device 411A can be prevented in advance, and the AI ​​processing can proceed smoothly.

[0159] In particular, in the processing system 401, the management server 431 takes into consideration the processing capacity and resource usage rate of each AI edge device 411, and can appropriately determine whether the AI ​​processing should be divided and executed by multiple AI edge devices 411, or whether it should be executed collectively by one AI edge device 411. Therefore, in the processing system 401, the AI ​​processing can be executed efficiently as a whole system, and a large amount of analysis data can be generated in a relatively short time.

[0160] In other respects as well, the processing system 401 can achieve the same effects as the first to third embodiments.

[0161] According to the above configuration, in the processing system 401 according to the fourth embodiment, when the AI ​​edge device 411A detects an abnormality, the processing system 401 switches to another AI edge device 411B as a switching destination to execute AI processing under the management of the management server 431. Furthermore, in the processing system 401, when the resource usage rate of the AI ​​edge device 411A increases, the processing system 401 switches to another AI edge device 411B as an allocation destination to execute part or all of the AI ​​processing under the management of the management server 431. Therefore, in the processing system 401, when each AI edge device 411 cannot continue AI processing, the processing system 401 can immediately appropriately select another AI edge device 411 and perform switching or allocation, so that the generation and supply of analysis data can be smoothly continued with almost no interruption in AI processing.

[0162] 5. Other embodiments In the above-described first embodiment, a case has been described in which, after the AI ​​edge device 11B is selected as the switching destination in the state switching process (FIG. 7), the management server 31 transmits the container 46A to the AI ​​edge device 11B to be deployed thereto. However, the present invention is not limited to this. For example, the same container 46A as that of the AI ​​edge device 11A may be deployed to the AI ​​edge device 11B that belongs to the same group as the AI ​​edge device 11A and is connected to the same sensor device 12A. In this case, after the AI ​​edge device 11B is selected as the switching destination, the AI ​​edge device 11B can immediately start AI processing using the container 46A. The same applies to the third and fourth embodiments.

[0163] In the above-described embodiment, in the state switching process (FIG. 7), the management server 31 selects the AI ​​edge device 11B on the condition that the AI ​​edge device 11B belongs to the same group as the AI ​​edge device 11A of the switching source and is connected to the same sensor device 12A. However, the present invention is not limited to this. For example, when a group is formed of three or more AI edge devices 11 and connected to the same sensor device 12A, the resource usage rate of each device may be acquired in advance as in the second embodiment, and the AI ​​edge device 11 may be selected based on this. Alternatively, for example, the free storage capacity of each AI edge device 11 may be acquired in advance, and an AI edge device 11 having a free capacity larger than the capacity required for deploying the container 46 may be selected. Alternatively, for example, when multiple containers 46 are being executed in the AI ​​edge device 11A of the switching source, multiple AI edge devices 11 may be selected as the switching destination, and the containers 46 may be appropriately allocated and executed. The same applies to the third and fourth embodiments.

[0164] Furthermore, in the above-mentioned first embodiment, when the temperature of the VPU exceeds a predetermined threshold and an abnormality is detected, the AI ​​edge device 11 that executes the AI ​​processing is switched from the AI ​​edge device 11A that is the source of the AI ​​processing to the AI ​​edge device 11B that is the destination of the AI ​​processing. However, the present invention is not limited to this. For example, when the control unit 21 cannot access the storage unit 22, or when the connection with the sensor device 12A is cut off and not restored, the switching process may be performed when various abnormalities are detected. Alternatively, for example, a maintenance worker operates the management terminal device 32 (FIG. 1) to check the status of each AI edge device 11, and when there is an AI edge device 11 that is likely to be broken, the management terminal device 32 may send a switching request similar to step SP31 (FIG. 7) to the management server 31, and the subsequent state switching process may be performed using this as a trigger. The same applies to the second, third, and fourth embodiments.

[0165] Furthermore, in the above-mentioned first embodiment, a case has been described in which the state switching process (FIG. 7) is triggered by detecting an abnormality in the AI ​​edge device 11A that is the switching source and transmitting a switching request from the AI ​​edge device 11A to the management server 31. However, the present invention is not limited to this. For example, the management server 31 may periodically transmit information to each AI edge device 11 and perform alive monitoring processing to receive a response thereto, and may use failure to receive the response as a trigger to perform the state switching process as a trigger based on a judgment on the management server 31 side. Also, in the second embodiment, the management server 231 may determine whether each AI edge device 211 satisfies the resource condition, and then start the allocation change processing as necessary. The same applies to the third and fourth embodiments.

[0166] Furthermore, in the above-mentioned second embodiment, the usage rates of the CPU, VPU, memory, and storage are stored in the status management table 281 ( FIG. 11 ) as the resource usage rates in each AI edge device 211. However, the present invention is not limited to this, and some of the CPU, VPU, memory, and storage may be omitted, or various other items may be added, such as the communication volume per unit time in the communication unit 23. Alternatively, for example, by providing various sensors in the AI ​​edge device 211, the temperature of the CPU and VPU, the power supply voltage supplied from a predetermined power supply circuit (not shown), or power consumption may be included in the status information, and the items may be added to the status management table 281. The same applies to the third and fourth embodiments.

[0167] Furthermore, in the second embodiment described above, in step SP241 of the allocation change process (FIG. 12), the AI ​​edge device 211 with the lowest VPU usage rate is selected as the allocation destination. However, the present invention is not limited to this. For example, the usage and capacity of the CPU, VPU, memory, and storage required for each container 46 may be stored in advance, and the optimal allocation destination may be selected based on the usage and capacity of the CPU, VPU, memory, and storage in the newly allocated container 46 and the value of each resource stored in the state management table 281, for example, the free storage capacity. For example, when the free storage capacity of the AI ​​edge device 211 is larger than the storage capacity required to execute the container 46, the AI ​​edge device 211 may be selected as the allocation destination. Alternatively, for example, a predetermined simulation process may be performed, and the AI ​​edge device 211 that can minimize the usage of the CPU and VPU when the container 46 is executed may be selected as the allocation destination. The same applies to the third and fourth embodiments.

[0168] Furthermore, in the second embodiment described above, the case where only one AI edge device 211 is selected as the allocation destination in the allocation change process (FIG. 12) has been described. However, the present invention is not limited to this, and for example, two or more AI edge devices 211 may be selected as the allocation destination. The same applies to the third and fourth embodiments.

[0169] Furthermore, in the above-mentioned first embodiment, the case where the sensor device 12 is connected to the AI ​​edge device 11 by a communication means such as USB or Bluetooth (registered trademark) has been described. However, the present invention is not limited to this, and a network connecting each AI edge device 11 and each sensor device 12 may be constructed. In this case, for example, a wired LAN conforming to a standard such as IEEE802.3, a wireless LAN conforming to a standard such as IEEE802.11, or a wireless network conforming to a standard such as IEEE802.15.4 as in the third embodiment can be constructed. Alternatively, each sensor device 12 may be connected to the network 10 together with the AI ​​edge device 11 and the router 13. In these cases, since each AI edge device 11 can be connected to all the sensor devices 12, the sensor item may be omitted from the state management table 81 (FIG. 5). The same applies to the third and fourth embodiments.

[0170] Furthermore, in the above-mentioned first embodiment, the case where various sensors such as a temperature sensor and a humidity sensor are provided in the sensor device 12 has been described. However, the present invention is not limited to this, and various elements such as an imaging unit (so-called camera) having a lens and an imaging element to capture a moving image or a still image, and a recording unit having a microphone or the like to record a sound may be provided in the sensor device 12 to generate various data. The same applies to the second to fourth embodiments. For example, in the third embodiment, the network 310 may be a wireless LAN conforming to IEEE802.11, and the sensor device 312 may be a device called a network camera or IP camera, that is, a device that can be connected to the network 310 and has an imaging unit. In this case, an image (such as a moving image or a still image) captured by the imaging unit of the sensor device 312 can be supplied as data to the AI ​​edge device 311 via the network 310.

[0171] Furthermore, in the above-mentioned first embodiment, the case where the management server 31 is installed in the management center 3 has been described. However, the present invention is not limited to this, and for example, the management server 31 may be installed in the installation site 2 and connected to the network 10. The same applies to the second to fourth embodiments.

[0172] Furthermore, in the above-mentioned first embodiment, a case has been described in which the pre-switching program is stored in advance in the storage unit 22 of the AI ​​edge device 11, and the control unit 21 reads and executes it. However, the present invention is not limited to this, and the pre-switching program may be acquired (downloaded) and executed, for example, from the management server 31 or a predetermined server device (not shown) connected to the wide area network 4. The same applies to the second, third, and fourth embodiments.

[0173] Furthermore, in the above-mentioned first embodiment, a case has been described in which one installation site 2 and one management center 3 are provided in the processing system 1 (FIG. 1). However, the present invention is not limited to this, and for example, two or more installation sites 2 may be provided, and two or more management centers 3 may be provided. The same applies to the second to fourth embodiments.

[0174] Furthermore, the present invention is not limited to the above-described embodiments and other embodiments. That is, the scope of application of the present invention extends to embodiments in which the above-described embodiments and the other embodiments are combined in part or in whole in any manner, or to embodiments in which parts are extracted.

[0175] Furthermore, in the above-described embodiment, a case has been described in which a plurality of AI edge devices 11 are connected to a management server 31 serving as a management device, and the processing system 1 is configured as a processing system by a state management unit 71 serving as an acquisition unit, a control unit 61 serving as a detection unit and execution unit, and a switching management unit 72 serving as a selection unit. However, the present invention is not limited to this, and the processing system may be configured by an acquisition unit, a detection unit, a selection unit, and an execution unit having various other configurations. [Industrial Applicability]

[0176] The present invention can be used, for example, in a processing system that performs AI processing using multiple AI edge devices. [Explanation of symbols]

[0177] 1, 201, 301, 401...processing system, 11, 211, 311, 411...AI edge device, 12, 212, 312...sensor device, 21, 221, 321, 421...control unit, 22, 222...memory unit, 31, 231, 331, 431...management server, 46...container, 61, 261, 361, 461...control unit, 62, 262, 362, 462...memory unit, 71...status management unit, 72...switching management unit, 81, 281, 381, 481...status management table, 82...container memory unit, 273, 473...allocation management unit.

Claims

1. A management device that manages a plurality of AI edge devices that execute AI (Artificial Intelligence) processing based on data supplied from a sensor device, An acquisition unit that acquires information representing a state of each of the AI ​​edge devices; A detection unit that detects an abnormality in the AI ​​edge device; A selection unit that selects a switching destination AI edge device that is another AI edge device that should execute a target AI processing, which is the AI ​​processing that was executed in the switching source AI edge device that is the AI ​​edge device in which the abnormality was detected; A transmission unit that transmits an instruction to the switching destination AI edge device to execute the target AI processing in the switching destination AI edge device; Equipped with The detection unit detects the abnormality of the switching source AI edge device by receiving a switching request from the switching source AI edge device to request that the target AI process be switched to another AI edge device. A management device comprising:

2. A management device that manages a plurality of AI edge devices that execute AI processing based on data supplied from a sensor device, An acquisition unit that acquires information representing a state of each of the AI ​​edge devices; A management unit that manages a container that virtualizes an application executed on the AI ​​edge device; A detection unit that detects an abnormality in the AI ​​edge device; A selection unit that selects a switching destination AI edge device that is another AI edge device that should execute a target AI processing, which is the AI ​​processing that was executed in the switching source AI edge device that is the AI ​​edge device in which the abnormality was detected; A transmission unit that transmits an instruction to the switching destination AI edge device to execute the target AI processing in the switching destination AI edge device; Equipped with The transmission unit transmits the container to the switching destination AI edge device so that the container in which the application is virtualized is deployed and executed in the switching destination AI edge device when the switching destination AI edge device does not have the container for executing the target AI processing. A management device comprising:

3. A management device that manages a plurality of AI edge devices that execute AI processing based on data supplied from a sensor device, An acquisition unit that acquires information representing a state of each of the AI ​​edge devices; A detection unit that detects an abnormality in the AI ​​edge device; A selection unit that selects a switching destination AI edge device that is another AI edge device that should execute a target AI processing, which is the AI ​​processing that was executed in the switching source AI edge device that is the AI ​​edge device in which the abnormality was detected; A transmission unit that transmits an instruction to the switching destination AI edge device to execute the target AI processing in the switching destination AI edge device; Equipped with The selection unit is an AI edge device that is previously associated with one set of two or more devices, and selects one of the remaining AI edge devices previously associated with the switching source AI edge device as the switching destination AI edge device. A management device comprising:

4. A management device that manages a plurality of AI edge devices that execute AI processing based on data supplied from a sensor device, An acquisition unit that acquires information representing a state of each of the AI ​​edge devices; A detection unit that detects an abnormality in the AI ​​edge device; A selection unit that selects a switching destination AI edge device that is another AI edge device that should execute a target AI processing, which is the AI ​​processing that was executed in the switching source AI edge device that is the AI ​​edge device in which the abnormality was detected; A transmission unit that transmits an instruction to the switching destination AI edge device to execute the target AI processing in the switching destination AI edge device; Equipped with The selection unit selects another AI edge device capable of acquiring the data necessary for the target AI processing as the switching destination AI edge device. A management device comprising:

5. A management device that manages a plurality of AI edge devices that execute AI processing based on data supplied from a sensor device, An acquisition unit that acquires information representing a state of each of the AI ​​edge devices; A storage unit that stores a storage capacity required for executing the AI ​​processing in the AI ​​edge device in association with the AI ​​processing; A detection unit that detects an abnormality in the AI ​​edge device; A selection unit that selects a switching destination AI edge device that is another AI edge device that should execute a target AI processing, which is the AI ​​processing that was executed in the switching source AI edge device that is the AI ​​edge device in which the abnormality was detected; A transmission unit that transmits an instruction to the switching destination AI edge device to execute the target AI processing in the switching destination AI edge device; Equipped with The acquisition unit acquires information indicating the free capacity of the storage in each of the AI ​​edge devices in addition to information on the AI ​​processing being executed in each of the AI ​​edge devices, The selection unit selects, as the switching destination AI edge device, the other AI edge device in which the free capacity of the storage is larger than the capacity of the storage required to execute the target AI processing. A management device comprising:

6. A management device that manages a plurality of AI edge devices that execute AI processing based on data supplied from a sensor device, An acquisition unit that acquires information representing a state of each of the AI ​​edge devices; A detection unit that detects an abnormality in the AI ​​edge device; A selection unit that selects a switching destination AI edge device that is another AI edge device that should execute a target AI processing, which is the AI ​​processing that was executed in the switching source AI edge device that is the AI ​​edge device in which the abnormality was detected; A transmission unit that transmits an instruction to the switching destination AI edge device to execute the target AI processing in the switching destination AI edge device; Equipped with The acquisition unit acquires information representing a resource usage state in each of the AI ​​edge devices, The detection unit detects a resource shortage in the AI ​​edge device in addition to the abnormality in the AI ​​edge device, When the resource shortage is detected, the selection unit determines whether the target AI processing should be distributed and executed by an assignment source AI edge device, which is the AI ​​edge device that detected the resource shortage, and one or more other AI edge devices, which are assignment destination AI edge devices, to which the target AI processing should be assigned, or whether the target AI processing should be executed collectively by only the assignment destination AI edge device, and selects the number of assignment destination AI edge devices based on the result of the determination; When it is determined that the target AI processing should be executed in a distributed manner, the transmission unit transmits the instruction to execute the target AI processing in a distributed manner by the assignment source AI edge device and the assignment destination AI edge device, and when it is determined that the target AI processing should be executed collectively, the transmission unit transmits the instruction to execute the target AI processing collectively by one of the assignment destination AI edge devices. A management device comprising:

7. A program for operating an information processing device as a management device described in any one of claims 1 to 6, the program causing the information processing device to function as each part of the management device.

8. A processing system in which a plurality of AI edge devices that perform AI processing based on data supplied from a sensor device and a management device that manages the AI ​​edge devices are connected, The management device is a management device according to any one of claims 1 to 6. A processing system comprising:

9. A processing system in which a plurality of AI edge devices that perform AI processing based on data supplied from a sensor device and a management device that manages the AI ​​edge devices are connected, The management device is an information processing device that runs the program according to claim 7 and functions as each unit included in the management device. A processing system comprising:

Images