Support systems and programs

Abstract

We support the implementation and operational management of AI models into existing monitoring systems. [Solution] The support system of the embodiment supports the implementation and operation management of an AI model for an existing monitoring system that performs predetermined processing based on sensor information acquired from sensor devices that detect the state of a monitored object. The support system stores model management information that shows the relationship between the monitoring system and the AI ​​model implemented in the monitoring system. It includes a distribution unit that distributes program modules for operating the AI ​​model to each monitoring system and distributes the corresponding AI model to each monitoring system based on the model management information, and a learning data management unit that stores the sensor information received from each monitoring system as learning data in a predetermined storage area and outputs the sensor information to an AI management system that generates or updates an AI model by performing learning processing.

Description

Technical Field

[0001] Embodiments of the present invention relate to a technology for assisting the implementation of an AI model into an existing monitoring system and for assisting the operation and management of the implemented AI model.

Background Art

[0002] The IoT (Internet of Things) system is widely known as a technology for collecting sensor data and grasping the state of a target. As an application of this technology, for example, there is a monitoring system that collects a large amount of sensor data from factories and facilities and detects or predicts failures.

[0003] Existing monitoring systems have a determination function such as failure detection introduced for each base where factories or facilities are located. This determination function performs determination processing such as normal or abnormal based on sensor information acquired from sensor devices that detect the state of the monitoring target.

[0004] On the other hand, in recent years, the AI - transformation of the determination function has been progressing, and there is a demand for a technology to assist the AI - transformation of existing monitoring systems that do not have a determination function using an AI model.

Prior Art Documents

Patent Documents

[0005]

Patent Document 1

Patent Document 2

Patent Document 3

Summary of the Invention

Problems to be Solved by the Invention

[0006] Provide a support system that assists the implementation of an AI model into an existing monitoring system and assists the operation and management of the implemented AI model. [Means for solving the problem]

[0007] The support system of the embodiment is a support system that implements an AI model that performs predetermined processing based on sensor information acquired from sensor equipment that detects the state of a monitored object, and supports the operation management of the AI ​​model. The support system includes a first storage unit that stores model management information indicating the relationship between the monitoring system and the AI ​​model implemented in the monitoring system; a distribution unit that distributes program modules that operate the AI ​​model to perform predetermined processing based on the sensor information to each of the monitoring systems and distributes the corresponding AI model to each of the monitoring systems based on the model management information; and a learning data management unit that stores the sensor information received from each of the monitoring systems as learning data in a predetermined storage area and outputs the sensor information to an AI management system that generates or updates an AI model by performing learning processing. [Brief explanation of the drawing]

[0008] [Figure 1] This is a diagram showing the configuration of the support system and monitoring system according to the first embodiment. [Figure 2] This diagram illustrates the mode of cooperation between the monitoring system and the AI ​​management system using the support system of the first embodiment. [Figure 3] This figure shows an example of learning data management information in the first embodiment. [Figure 4] This figure shows an example of model management information in the first embodiment. [Figure 5] This flowchart shows the process flow for implementing the AI ​​model of the first embodiment and updating the implemented AI model. [Figure 6] This figure shows an example of how property information associated with sensor information in the first embodiment can be used. [Figure 7] This figure shows another embodiment of a monitoring system to which the support system of the first embodiment is applied. [Best Mode for Carrying Out the Invention]

[0009] The embodiments will be described below with reference to the drawings.

[0010] (First Embodiment) Figures 1 to 7 are diagrams illustrating the first embodiment. Figure 1 is a configuration diagram of the support system 100 and monitoring system 700 of this embodiment. In the example in Figure 1, the support system 100 assists multiple monitoring systems 701, 702, and 703 in the implementation and operation management of AI models.

[0011] The monitoring system 700 (701, 702, 703) is an IoT-based computer system that performs predetermined processing based on sensor information acquired from sensor devices s that detect the state of the monitored object. For example, it can monitor equipment and machinery at locations such as factories and facilities, and perform processing such as anomaly detection and predictive maintenance.

[0012] The monitoring system 700 of this embodiment is equipped with a mechanism for centrally managing the status of each location scattered domestically and / or overseas, and is configured so that the status of multiple locations can be centrally monitored through the monitoring management system 500. In the example in Figure 1, the management systems 300A, 300B, and 300C installed at the locations to be monitored are each connected to the monitoring management system 500 via a network (IP network, dedicated line, etc.). In other words, the monitoring systems 701, 702, and 703 are each constructed as independent computer systems for each location, and the monitoring management system 500 is formed as a common monitoring management platform for the monitoring systems 701, 702, and 703.

[0013] Therefore, each of the multiple monitoring systems 700 (701, 702, 703) consists of a site-side management system 300 (300A, 300B, 300C) and a cloud-side monitoring management system 500. The monitoring management system 500 receives processing results and / or alert information based on those processing results output from the site-side management systems 300 (300A, 300B, 300C). The monitoring management system 500 can collect processing results and / or alert information based on those processing results from each site and monitor them centrally. The display of processing results and the display of alert information based on those processing results are performed by the monitoring control unit 510 of the monitoring management system 500.

[0014] In the non-AI-enabled monitoring systems 700 (701, 702, 703), a monitoring management unit (not shown) is provided in the management device 310 (310A, 310B, 310C), and sensor information is input from sensor devices s. The monitoring management unit then performs predetermined processing based on the sensor information and outputs the processing results and / or alert information based on the processing results. The sensor information management unit 312 outputs the sensor information acquired from the sensor devices s to the monitoring management unit, and the storage unit 313 records the sensor information, processing results, and alert information based on the processing results in chronological order. The sensor information management unit 312 can also transmit sensor information to the monitoring management system 500 via the network, and the monitoring management system 500 can receive processing results from the management device 310 via the network and output alert information based on the processing results. In other words, sensor information from sensor devices s, processing results, and alert information can be stored and accumulated in the storage area provided in the monitoring management system 500.

[0015] The support system 100 of this embodiment supports the AI ​​implementation of the monitoring and management unit, and also supports the operation and management of the AI ​​model after the AI ​​implementation. In the example shown in Figure 1, the management device 310A of the monitoring system 701 is equipped with an AI model control unit 311A ​​instead of the monitoring and management unit. The AI ​​model control unit 311A ​​is an AI model execution area that implements the AI ​​model z and causes the AI ​​model z to perform predetermined processing. The sensor information management unit 312 outputs sensor information to the AI ​​model control unit 311A ​​(AI model). The AI ​​model control unit 311A ​​also outputs processing results based on the sensor information and / or alert information based on the processing results. Note that the sensor information management unit 312 and the storage unit 313 are the same in the monitoring systems 702 and 703, so the same reference numerals are used and their explanation is omitted.

[0016] The monitoring systems 702 and 703 also have AI-powered monitoring and management units. The management system 300B includes a management device 310B and is composed of an AI model control unit 311B, a sensor information management unit 312, and a storage unit 313. The management device 310B is connected to multiple sensor devices s1, s2, and s3 and acquires sensor information from each of the sensor devices s1, s2, and s3. The AI ​​model control unit 311B implements three AI models: AI model a, AI model b, and AI model c. The multiple AI models a, b, and c may each be an AI model that performs independent processing, or they may be an AI model that performs predetermined processing using the processing results of other AI models as input information in addition to sensor information. Alternatively, they may be an AI model that takes sensor information from multiple different sensor devices s as input.

[0017] The management system 300C includes a management device 310C, which is composed of an AI model control unit 311C, a sensor information management unit 312, and a memory unit 313. The management device 310C is connected to a plurality of sensor devices s10 and s11, and acquires respective sensor information from the sensor devices s10 and s11. The AI model control unit 311C implements two AI models, x and y. The management device 310C implements the AI models x and y for each of the plurality of monitoring targets. The AI model control unit 311C separates the sensor information of the sensor device s10 corresponding to the AI model x and the sensor information of the sensor device s11 corresponding to the AI model y, and causes the AI models x and y to perform predetermined processing.

[0018] Here, the sensor information will be described. The sensor information is quantitative information of a monitoring (measurement) target acquired using a sensor device such as a photographing device, a detector, a sensor, a measuring device, etc., and can include, for example, images (still images, moving images), temperature, humidity, vibration, sound, etc. The sensor device s can include an imaging device (camera or infrared camera), a temperature sensor, a humidity sensor, a vibration sensor, a sound sensor, etc.

[0019] Also, the predetermined processing based on the sensor information is exemplified by processing such as abnormality detection and sign detection of facilities and devices, but is not limited thereto. For example, it may perform a determination process of good products / defective products for manufactured products. The AI models exemplified in this embodiment are merely examples, and can include determination functions and the like necessary for the monitoring system 700. Also, the AI model may be a customized AI model, or may be a known AI model such as object detection or image recognition, and an AI model appropriately updated by a learning process.

[0020] Furthermore, the processing results include, for example, judgment results indicating normal or abnormality, judgment results indicating whether or not there are signs of a predetermined event, and judgment results indicating good or defective products. In addition, alert information includes, for example, information indicating warnings regarding judgment results, and includes information for displaying warnings, etc., and warning messages, etc. The AI ​​model control unit 311 of each monitoring system 700 can transmit processing results and alert information based on processing results to the monitoring management system 500 via the network. The support system 100 can obtain processing results and alert information based on processing results through the monitoring management system 500 or directly from the AI ​​model control unit 311.

[0021] While the example illustrates monitoring three different locations, it is not limited to this configuration. The support system 100 can be applied to multiple monitoring systems 700 that monitor multiple locations, providing support for the implementation of AI models and the operational management of AI models at locations that have not yet been AI-enabled.

[0022] Furthermore, in this embodiment, as an example of a monitoring system 700, a system configuration is shown in which the system is divided into a site-side management system 300 and a cloud-side monitoring management system 500, with the monitoring management system 500 grasping the monitoring status of multiple sites. However, this is not the only example, and the support system 100 can be applied to a known IoT-based monitoring system. For example, the support system 100 can also be applied to an independent monitoring system 700 that is completed by the site-side management system 300.

[0023] <Explanation of the support system> As shown in Figure 1, the support system 100 is comprised of a linkage system 110, an AI management system 120, and a data management system 130. The support system 100 is connected to the monitoring system 700 (701, 702, 703) via a network (IP network, dedicated line, etc.).

[0024] The collaborative system 110 includes a storage unit 113 that stores model management information indicating the relationship between each monitoring system 700 and the AI ​​model implemented in each monitoring system 700. The model management information is managed by the distribution management unit 112. For example, it provides a screen for inputting model management information, and allows registration control of model management information via the screen from an administrator terminal (not shown). The distribution unit 111 distributes platform modules (program modules, corresponding to AI model control units) that operate the AI ​​models and perform predetermined processing based on sensor information to each monitoring system 700 via the network. The distribution unit 111 also distributes the corresponding AI models to each monitoring system 700 via the network based on the model management information. The collaborative system 110 can also distribute the platform modules and AI models as a single combined distribution data to each monitoring system 700.

[0025] The AI ​​management system 120 includes an AI model management unit 121 and an AI model learning unit 122. The AI ​​model management unit 121 controls the learning process of the AI ​​model learning unit 122 and provides the generated or updated AI models to the collaborative system 110 (distribution unit 111). The AI ​​model learning unit 122 performs learning processing using sensor information as learning data and generates or updates AI models. The AI ​​models provided by the AI ​​management system 120 are stored in the storage unit 110 of the collaborative system 110, but it is also possible to control the system to retain only AI models that can be distributed.

[0026] The AI ​​management system 120 can be configured to include a monitoring function (monitoring unit). The timing for updating the AI ​​model is arbitrary, but for example, an AI model update trigger can be generated based on processing results received from each monitoring system 700. The AI ​​model management unit 121 may be configured to update (learn) the corresponding AI model based on the update trigger generated by the monitoring function. The monitoring conditions for generating the update trigger can be set arbitrarily.

[0027] The data management system 130 includes an information collection unit 131, a learning data management unit 132, and a storage unit 133. The information collection unit 131 collects sensor information from each monitoring system 700 via the network. The sensor information can be provided to the support system 100 by the sensor information management unit 312 of each monitoring system 700, either indirectly through the monitoring management system 500 or directly to the support system 100.

[0028] The learning data management unit 132 stores sensor information received from each monitoring system 700 (701, 702, 703) as learning data in the storage unit 133, and outputs the sensor information to the AI ​​management system 120 during the learning process. The storage unit 133 stores the acquired sensor information as a dataset based on the correspondence with each monitoring system 700, which is the source of the sensor information. In other words, the learning data management unit 132 manages to form a dataset for each sensor information and can provide the AI ​​management system 120 with a learning dataset with sensor information. At this time, it may be configured to manage the sensor information and the processing results based on that sensor information as learning data (training data) in association.

[0029] Figure 2 is a diagram illustrating the coordination between the support system 100 and the monitoring system 700 (701, 702, 703) and the AI ​​management system 120.

[0030] As shown in Figure 2, for each monitoring system 701, 702, and 703, the data management system 130 functions as a collection unit for sensor information and processing results (or alerts) of the monitored systems, and also functions as a supply unit that provides training data to the AI ​​management system 120. Similarly, the collaborative system 110 functions as a distribution unit for AI models managed by the AI ​​management system 120 for each monitoring system 701, 702, and 703.

[0031] In other words, the support system 100 provides data linkage (data collection unit, data provision unit) and AI model distribution (AI model linkage) functions between the single AI management system 120 and each of the monitoring systems 701, 702, and 703.

[0032] Here, the AI ​​management system 120 can be built using, for example, a method for managing the lifecycle of an AI model, such as MLOps (Machine Learning Operations). MLOps refers to a series of processes that manage the entire lifecycle of an AI model, from development to deployment and operation.

[0033] On the other hand, simply applying MLOps would not be sufficient to convert the existing monitoring system 700 into an AI-powered system. In other words, a new monitoring system, including the AI ​​management system 120, would need to be designed and built for each location where an AI model is to be implemented. However, the support system 100 of this embodiment can link the existing monitoring systems 701, 702, and 703 with the AI ​​management system 120, supporting the implementation of AI models in each of the existing monitoring systems 701, 702, and 703, as well as the operational management of the AI ​​models. In short, it is possible to support the implementation and operational management of AI models while utilizing the existing AI management system 120 built using MLOps methods and while suppressing the system configuration of the existing monitoring system 700.

[0034] In other words, AI functionality can be retrofitted to existing monitoring systems 700 that are not AI-enabled, allowing for easy migration to AI-enabled systems. In particular, the integration system 110 and the data management system 130 enable integration with multiple IoT-based monitoring systems 700 without modifying (or altering) the existing AI management system 120, which is built using MLOps methods.

[0035] Figure 3 shows an example of learning data management information. The learning data management information is information that shows the correspondence between each monitoring system 700, which is the source of sensor information, and the sensor information, and is also information that manages the sensor information acquired from each monitoring system 700 as a dataset. It includes a dataset ID, AI model ID, learning data acquisition source information, and property information (property information 1, 2, ...). The learning data management unit 132 provides a screen for inputting learning data management information, and the registration of learning data management information can be controlled via the screen from an administrator terminal (not shown).

[0036] The dataset ID is information that identifies sensor information and is assigned to each training data source. The AI ​​model ID is unique information that identifies the AI ​​model. Sensor information is associated with training data source information and property information. Training data source information is information that allows us to understand which sensor device in which monitoring system the sensor information was acquired from. For example, dataset ID "DS001" is associated with sensor information acquired from the management system 300X, management device 310X, and sensor device sx of monitoring system 700, and property information is also associated with it. After the AI ​​model is generated, the AI ​​model ID is associated with dataset ID "DS001". Also, dataset ID "DS002" is associated with the AI ​​model "seg1", and the management system 300A, management device 310A, and sensor device s of monitoring system 701 are registered as training data source information, and property information is further associated with the sensor information acquired from sensor device s. Sensor information serves as training data for generating or updating AI models, and a set of sensor information acquired from one sensor device s is managed as a single dataset. Alternatively, the system may be configured to manage sets of sensor information acquired from multiple sensor devices s as a single dataset. Furthermore, a single dataset can be used to generate or update multiple AI models; in this case, a single dataset can be managed to be associated with multiple AI models.

[0037] Property information (attribute information) includes sensor type, installation location, monitored object, and sensing type (monitoring attribute). Sensor type is the type of sensor device, and includes cameras, temperature sensors, vibration sensors, etc. Installation location indicates where the sensor device is installed in the monitoring system 700, and includes specific locations such as XX factory or XX facility. Monitored object indicates the specific equipment or device to be monitored, and includes equipment such as septic tanks or manufacturing lines, or equipment such as compressors. Sensing type is information indicating what the sensor device is sensing within the monitored object. For example, if the monitored object is a septic tank, the condition of the "water surface" is sensed by the sensor device (sensor type: camera). If the monitored object is a compressor, the vibration condition of the "upper part of the device" is sensed by the sensor device (sensor type: vibration sensor).

[0038] In this embodiment, the data management system 130 manages the sensor information so that a dataset is formed for each sensor information, and associates the sensor information group (learning data) with learning data acquisition source information (for example, a sensor number assigned to each data acquisition source) and property information (for example, sensor type, installation location such as factory name, monitoring (measurement) target, sensing type, etc.). The data management system 130 stores learning data management information, including the acquisition source of the sensor information (learning data) and the property information of the sensor information, in the storage unit 133.

[0039] The learning data management unit 132 associates sensor information acquired from each location (monitoring systems 701, 702, 703) where the AI ​​model is implemented with the corresponding dataset (learning data source) based on the learning data management information, and also associates property information. The learning data management unit 132 then manages and stores the sensor information used for the AI ​​model's learning process based on the learning data management information, and outputs the sensor information collected from multiple locations as learning data to the AI ​​management system 120. The AI ​​management system 120 associates the dataset identified by the learning data management information with the learning data of the AI ​​model to be generated or updated, performs the learning process, and can generate and / or update the AI ​​model. When the AI ​​management system 120 generates an AI model for the first time, it can generate an AI model ID and store it in the learning data management information. Therefore, after the AI ​​model is generated, the AI ​​model ID will be associated with the dataset ID, and for example, the correspondence "AI model α was created with dataset Y, and the sensor information included in dataset Y is X, W, and V" will be recorded in the learning data management information.

[0040] Figure 4 shows an example of model management information. Model management information includes various types of information such as the type of data to be distributed, the processing method for the distribution destination, the AI ​​model ID, and the distribution destination information.

[0041] Model management information manages which monitoring system implements which AI model, what processing the implemented (or completed) AI model performs, and controls the distribution of AI models. The distribution data type identifies, for example, whether to distribute an AI model or a platform module (AI model control unit). The distribution destination processing indicates the predetermined processing performed by the AI ​​model, and includes processing names such as image recognition, object detection, sound recognition, anomaly detection, and predictive detection. The model name (AI model ID) is identification information for the AI ​​model that is applied in common with the training data management information. Distribution destination information (distribution destination information 1, distribution destination information 2) is information about the location where the AI ​​model is implemented, and is information for setting which AI model is distributed to which management device in which monitoring system. Note that multiple management devices 310 may be installed within the same monitoring system 700, and the same or different AI models may be implemented in each management device 310, so the distribution destination of the AI ​​model is set for each monitoring system 700 and each management device 310.

[0042] The linked system 110 stores model management information in its storage unit 113 that shows the relationship between each monitoring system 700 and the AI ​​model implemented in each monitoring system 700, and the distribution unit 111 distributes the corresponding AI model to each monitoring system 700 based on the model management information.

[0043] Here, we will explain the platform module. This platform module is a program module that forms a platform for running the AI ​​model on the management device 310 at each location, and corresponds to the AI ​​model control unit 311. The platform module is always distributed to each monitoring system 700 that is not AI-enabled, and is installed on the management device 310 as the AI ​​model control unit 311 that runs the AI ​​model. On the other hand, the platform module can be updated as needed, and the updated platform module can be distributed as appropriate to the monitoring system 700 on which the AI ​​model control unit 311 is installed (i.e., the AI ​​model is implemented).

[0044] Figure 5 is a flowchart showing the process flow for implementing the AI ​​model of this embodiment and updating the implemented AI model.

[0045] The support system 100 performs the learning data management information registration process (S101). As described above, the data management system 130 accepts input of learning data management information through the administrator terminal. Although the input method has been described via a screen, it is not limited to this, and for example, it may be configured to import learning data management information that has been created separately.

[0046] The support system 100 performs association processing of sensor information acquired from each location. The monitoring system 700 acquires sensor information from the sensor equipment (S201) and transmits the sensor information to the support system 100 (data management system 130) (S202). Based on the learning data management information, the data management system 130 associates the sensor information with the dataset and property information corresponding to the learning data acquisition source information (S102). The sensor information is stored in the storage unit 133.

[0047] The support system 100 provides training data to the AI ​​management system 120. Based on the training data management information, the data management system 130 extracts the training data (sensor information, training data including sensor information) held in the storage unit 133 and outputs the training data set to the AI ​​management system 120 (S103). The training data used here can be obtained from the management device 310, specifically the processing results and sensor information based on sensor information from the monitoring and management unit that was built in the management device 310 before the AI ​​was implemented. In other words, the data management system 130 can be configured to further associate the sensor information, to which the dataset and property information are associated, with the processing results or alert information based on the processing results from the monitoring and management unit that was built in the management device 310 before the AI ​​was implemented. The AI ​​management system 120 performs training processing and generates an AI model (S104). The AI ​​management system 120 outputs the generated AI model to the linkage system 110.

[0048] At this time, the AI ​​management system 120 generates an AI model ID for the first AI model generated and stores the generated AI model ID in the corresponding dataset of the training data management information (S105). In addition, the support system 100 (distribution management unit 112) receives input of model management information as the AI ​​model is generated and performs registration processing of the model management information (S105).

[0049] The support system 100 distributes the generated AI model to the monitoring system 700. The distribution unit 111 distributes a pre-created platform module to the relevant monitoring system 700 based on the model management information (S106) and has it installed in the management device 310 (generating the AI ​​control unit 311 in the management device 310 (S203)). Subsequently, the distribution unit 111 distributes the generated AI model to the relevant monitoring system 700 (S107). The monitoring system 700 incorporates the distributed AI model into the AI ​​control unit 311 (S204).

[0050] The monitoring system 700 causes the AI ​​model to perform predetermined processing. The management device 310 acquires sensor information through the sensor information management unit 312 (S205). At this time, the management device 310 controls the sensor information to be output from the sensor information management unit 312 to the AI ​​model control unit 311. The AI ​​model control unit 311 causes the AI ​​model to perform predetermined processing based on the input sensor information (S206). The management device 310 outputs the processing result, transmits the processing result, and transmits the sensor information (S207, S208, S209).

[0051] The support system 100 performs association processing of sensor information acquired from each location. The data management system 130 associates the sensor information with a dataset (information from which learning data was acquired) and property information based on the learning data management information (S108). The sensor information is stored in the storage unit 133. At this time, the data management system 130 can acquire processing results or alert information based on processing results output from the AI ​​model implemented together with the sensor information. The system can be configured to further associate the corresponding processing results or alert information based on processing results with the sensor information to which the dataset and property information have been associated.

[0052] The support system 100 provides training data to the AI ​​management system 120. The data management system 130 extracts the training data (sensor information, training data including sensor information) held in the memory unit 133 based on the training data management information and outputs the training data to the AI ​​management system 120 (S109). The AI ​​management system 120 associates the dataset identified by the training data management information with the training data of the AI ​​model to be updated, performs training processing, and updates the AI ​​model (S110). The AI ​​management system 120 outputs the updated AI model to the linkage system 110.

[0053] As described above, the timing of AI model updates is arbitrary. When the AI ​​management system 120 performs an AI model update process (learning process), it requests the data management system 130 to provide the learning data related to the AI ​​model to be updated, and the data management system 130 can provide the relevant learning data to the AI ​​management system 120 in response to this request.

[0054] The support system 100 distributes the updated AI model to the monitoring system 700. The distribution unit 111 distributes the updated AI model to the corresponding monitoring system 700 based on the model management information (S111).

[0055] The support system 100 of this embodiment includes a linkage system 110 and a data management system 130, and as shown in Figure 2, it can link the existing monitoring systems 701, 702, and 703 with the AI ​​management system 120 to support the implementation of AI models and the operation management of AI models. Therefore, AI functionality can be retrofitted to existing monitoring systems 700 that are not AI-enabled, and users can easily switch to AI-enabled systems.

[0056] In particular, the integration system 110 and the data management system 130 enable the existing AI management system 120, which is built using MLOps methods, to be integrated with monitoring systems 700 on multiple IoT platforms without any modifications.

[0057] Figure 6 shows an example of how property information linked to sensor information in this embodiment can be used.

[0058] The example in Figure 6 extracts training data (each dataset) with the same property information for sensor type, monitored target, and sensing type (monitoring attribute), and updates the AI ​​model using sensor information acquired at other locations. In Figure 6, training data 1 and training data 11 are sensor information collected by different monitoring systems 700, but their property information for sensor type, monitored target, and sensing type is identical. Therefore, in addition to training data 1 which is associated with AI model z, training data 11, which has the same property information, is also used as training data for AI model z.

[0059] Another example of utilizing the commonality of property information is the ability to control the distribution (application) of the updated AI model z to other locations that share the same property information. In the example shown in Figure 6, the monitoring system 700, to which the AI ​​model a associated with the training data 11 is implemented, can be controlled to distribute the updated AI model z instead of AI model a.

[0060] In this case, even if the AI ​​model z has not been updated by the learning process using the learning data 1 and the learning data 11, the system may be controlled to distribute the updated AI model z to the monitoring system 700 on which the AI ​​model a associated with the learning data 11 is implemented. That is, the updated AI model z, which has been trained using sensor information (learning data 1) having the same property information, can be controlled to distribute to the monitoring system 700 in place of the AI ​​model a associated with the learning data 11.

[0061] The data management system 130 in this embodiment is configured externally to the AI ​​management system 120 and manages learning data management information, which includes the source of sensor information (learning data) and property information (attribute information) of the sensor information. It associates a dataset (learning data source information) and property information with each sensor information received from each monitoring system 700, and manages the sensor information (dataset) identified by the learning data management information as learning data.

[0062] The data management system 130 (learning data management unit 132) can use property information (sensor type, monitored target, monitoring attribute) to extract sensor information with the same property information from multiple different monitoring systems 700 and provide it to the AI ​​management system 120 as learning data. The AI ​​management system 120 can apply the sensor information from multiple different monitoring systems 700 as learning data to generate and / or update an AI model.

[0063] Furthermore, the distribution unit 111 can refer to the property attribute information of the sensor information used in the training process of the generated or updated AI model, extract the monitoring system 700 on which the AI ​​model associated with the sensor information having the same property information is implemented, and control the distribution of the same generated or updated AI model to one or more of the extracted monitoring systems 700.

[0064] By utilizing property information in this way, the management and operation of AI models can be streamlined.

[0065] In particular, as described above, the data management system 130 of this embodiment manages sensor information (datasets) identified by learning data management information including the source of sensor information (learning data) and property information (attribute information) of the sensor information as learning data, and provides the learning data to the AI ​​management system 120. For example, when managing collected sensor information, if property information (attribute information) does not exist, one AI management system 120 would be required for each AI model to determine, such as one AI management system 120 that generates and updates an AI model based on the premise that the input is "image from the first monitoring system" and another AI management system 120 that generates and updates an AI model based on the premise that the input is "temperature sensor from the second monitoring system". However, the support system 100 of this embodiment is equipped with a data management system 130, and property information (attribute information) is associated with and managed with the sensor information, so one AI management system 120 can support multiple monitoring systems (multiple different sensor types (each AI model corresponding to multiple different determination targets)).

[0066] While the description above illustrates an example of generating or updating an AI model by extracting training data (each dataset) with the same sensor type, monitored target, and sensing type (monitoring attribute) from the property information, this is not the only possible approach. For example, the system may be configured to extract training data (each dataset) and generate or update an AI model based on any one of the sensor type, monitored target, or sensing type (monitoring attribute), or any combination thereof.

[0067] Figure 7 shows another embodiment of the monitoring system 700 to which the support system 100 of this embodiment is applied.

[0068] In the example shown in Figure 1, an AI model control unit 311 is built on the management device 310 installed at the base station, and the AI ​​model is implemented there. On the other hand, as described above, the monitoring system 700 of this embodiment has a system configuration that includes a cloud-side monitoring management system 500, so it is possible to configure the monitoring management system 500 to implement the AI ​​models (AI model control units 311) of each base station (each monitoring system 700).

[0069] Therefore, as shown in Figure 7, the monitoring system 700 can be configured to include a management device 310 installed at the monitored site to receive sensor information from sensor equipment, and an AI model control unit 511A (AI model control unit 311A) which is connected to each management device 310 of multiple monitoring systems via a network, and in which a platform module is installed and an AI model is implemented for each monitoring system 700. The AI ​​model control unit 511A is an AI model execution area provided on the monitoring management system 500 side that corresponds to the AI ​​model control unit 311A.

[0070] The monitoring management system 500 acquires sensor information from the management device 310 of each monitoring system 700 and inputs the sensor information to the corresponding AI model control units 511 (511A, 511B, 511C). The monitoring management system 500 causes each AI model control unit 511 to execute predetermined processing by the AI ​​model. The processing results by the AI ​​models of the multiple monitoring systems 700 and / or alert information based on the processing results are controlled to be viewable via the network.

[0071] In the example shown in Figure 7, the distribution destination for model management information is the monitoring and management system 500, and the distribution destination for the AI ​​models is the AI ​​model control units 511(511A, 511B, 511C) at each location built within the monitoring and management system 500, or in other words, the AI ​​model execution areas where the AI ​​model control units 511(511A, 511B, 511C) are implemented.

[0072] As described above, each function constituting the support system 100 can be realized by a program. Computer programs prepared in advance to realize each function are stored in an auxiliary storage device, and a control unit such as a CPU reads the program stored in the auxiliary storage device into the main memory. The control unit then executes the program read into the main memory, thereby enabling the operation of each part.

[0073] Furthermore, the above program can also be provided to a computer in a state where it is recorded on a computer-readable recording medium. Examples of computer-readable recording media include optical discs such as CD-ROMs and Blu-ray® Disc Rewritables, phase-change optical discs such as DVD-ROMs, magneto-optical discs such as MO (Magneto Optical), magnetic discs such as floppy disks and hard disks, and memory cards such as SD memory cards and USB flash drives. Hardware devices such as integrated circuits (IC chips such as ROMs and RAMs) that are specially designed and configured for the purposes of the present invention are also included as recording media. Furthermore, the present invention, including the above-mentioned program, is not limited to being executed on a von Neumann computer architecture, but may also be executed on so-called non-von Neumann computer architectures, such as neurocomputers based on the mechanisms of brain neural circuits or quantum computers that apply quantum mechanics to information processing.

[0074] Although embodiments of the present invention have been described, these embodiments are presented as examples only and are not intended to limit the scope of the invention. This novel embodiment can be implemented in various other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their variations are included in the scope and spirit of the invention, as well as in the claims of the invention and its equivalents. [Explanation of Symbols]

[0075] 100 Support Systems 110 Integration System 111 Distribution Department 112 Distribution Management Department 113 Storage section 120 AI Management Systems 121 AI Model Management Department 122 AI Model Learning Department 130 Data Management System 130 131 Information Gathering Department 132 Learning Data Management Department 133 Storage section 300, 300A, 300B, 300C Management System 311,311A,311B,311C AI model control section 312 Sensor Information Management Department 313 Storage section 500 Monitoring and Management Systems 510 Monitoring and Control Unit 511, 511A, 511B, 511C AI Model Control Unit 700, 701, 702, 703 Monitoring System

Claims

1. An existing monitoring system that performs predetermined processing based on sensor information acquired from sensor devices that detect the state of a monitored object, implements an AI model that performs the predetermined processing, and a support system that assists in the operation and management of the AI ​​model, The monitoring system and a first storage unit that stores model management information indicating the relationship between the monitoring system and the AI ​​model implemented in the monitoring system, A distribution unit distributes a program module that operates an AI model to perform predetermined processing based on the sensor information to each of the monitoring systems, and also distributes the corresponding AI model to each of the monitoring systems based on the model management information. The AI ​​management system stores the sensor information received from each of the aforementioned monitoring systems as learning data in a predetermined storage area and generates or updates an AI model by performing learning processing, and includes a learning data management unit that outputs the sensor information, A support system characterized by including the following.

2. The system includes a second storage unit that stores learning data management information, which includes information about the source of the sensor information and attribute information of the sensor information. The learning data management unit associates the attribute information with the sensor information received from each monitoring system. The learning data management unit extracts sensor information with the same attribute information from among the sensor information received from multiple different monitoring systems as learning data and provides it to the AI ​​management system, The support system according to claim 1, characterized in that the distribution unit refers to the attribute information, extracts the monitoring system corresponding to the sensor information having the same attribute information, and controls the distribution of the same generated or updated AI model to one or more of the extracted monitoring systems.

3. The support system according to claim 2, characterized in that the attribute information includes one of the following: sensor type, target to be monitored, or monitoring attribute.

4. The support system includes the AI ​​management system, The aforementioned AI management system is An AI model learning unit that uses the aforementioned sensor information as learning data to perform learning processing and generates or updates an AI model, An AI model management unit controls the learning process of the AI ​​model learning unit and provides the generated or updated AI model to the distribution unit. An AI model monitoring unit generates an AI model update trigger based on the processing results of the predetermined process received from the monitoring system and outputs it to the AI ​​model management unit. The support system according to claim 1, characterized by having the following features.

5. The aforementioned monitoring system A management device installed at a monitored location, which receives sensor information from the sensor equipment and includes an AI model execution area where the program module is installed and the AI ​​model is implemented; A monitoring management system that acquires processing results and / or alert information based on predetermined processing by an AI model from each of the management devices of multiple monitoring systems via a network, and controls each of the processing results and / or alert information of the multiple monitoring systems to be viewable via the network, The support system according to claim 1, characterized by including the following:

6. The aforementioned monitoring system A management device installed at the monitored location, which receives sensor information from the sensor device, A monitoring management system that connects to the respective management devices of multiple monitoring systems via a network, includes an AI model execution area where the program module is installed and the AI ​​model is implemented for each monitoring system, acquires sensor information from the management device of each monitoring system, causes the AI ​​model to execute the predetermined processing in the corresponding AI model execution area, and controls the processing results of the predetermined processing of the multiple monitoring systems and / or alert information based on the processing results to be viewable via the network, The support system according to claim 1, characterized by including the following:

7. An existing monitoring system that performs predetermined processing based on sensor information acquired from sensor equipment that detects the state of a monitored object is equipped with an AI model that performs the predetermined processing, and a program that is executed by a computer that supports the operation and management of the AI ​​model, wherein the computer... The monitoring system and a first function that stores model management information indicating the relationship between the monitoring system and the AI ​​model implemented in the monitoring system, A second function which distributes a program module that operates the AI ​​model to perform the predetermined processing based on the sensor information to each of the monitoring systems, and which distributes the corresponding AI model to each of the monitoring systems based on the model management information, A third function that stores the sensor information received from each of the aforementioned monitoring systems as learning data in a predetermined storage area, and outputs the sensor information to an AI management system that performs learning processing to generate or update an AI model, A program characterized by its ability to achieve this.

8. A method for implementing an AI model that performs predetermined processing based on sensor information acquired from sensor devices that detect the state of a monitored object, and for supporting the operation and management of the AI ​​model, wherein a computer... The steps include storing model management information that shows the relationship between the monitoring system and the AI ​​model implemented in the monitoring system, The steps include: distributing a program module that operates the AI ​​model and performs the predetermined processing based on the sensor information to each monitoring system, and distributing the corresponding AI model to each monitoring system based on the model management information; The steps include: storing the sensor information received from each of the aforementioned monitoring systems as learning data in a predetermined storage area, and outputting the sensor information to an AI management system that performs learning processing to generate or update an AI model; A method characterized by performing the following:

Images