JP6911004B2 - Monitoring model update method, monitoring system, and monitoring device - Google Patents

Description

The present invention relates to a monitoring model update method, a monitoring system, and a monitoring device.

In recent years, for example, IoT data is collected by an IoT (Internet of Things) sensor installed in a machine at a manufacturing site, and based on the IoT data and a monitoring model, the occurrence of defective products manufactured by this machine is monitored in real time. The need to do is increasing. Such real-time monitoring is required to follow changes in site conditions (product specification changes, equipment changes, seasonal changes, etc.). For example, Patent Document 1 supports this work when updating a monitoring model for following changes in the situation at the site in a condition monitoring device that diagnoses mechanical abnormalities at the site based on IoT data and a diagnostic model. The technology to be used is disclosed.

Japanese Unexamined Patent Publication No. 2015-203936

However, the above-mentioned conventional technique accepts changes in the monitoring model input by the user when updating the monitoring model, evaluates the performance of the changed monitoring model, and determines that the performance is sufficient. The model is updated with. As described above, in the prior art, when updating the monitoring model in real-time monitoring, there is a problem that the model cannot be updated by quickly following the change in the situation at the site.

This application has been made in view of the above-mentioned problems, and when updating a monitoring model in real-time monitoring, a monitoring model update method, a monitoring system, and a monitoring device that can quickly follow changes in the situation at the site and update the model. The purpose is to provide.

In order to solve such a problem, in the present invention, for example, in the monitoring model update method executed by the monitoring system, the data regarding the monitoring target collected by the sensor and the abnormality of the monitoring target are determined in advance. Based on the monitoring model, the monitoring step for determining the abnormality of the monitoring target in real time, and the time-series data including the data and the determination result corresponding to the data by the monitoring step based on a predetermined generation rule. A monitoring model generation step that periodically generates update candidates for the monitoring model based on the collected model generation data, and a monitoring model that stores the update candidates generated by the monitoring model generation unit in the storage unit. It has a memory step.

According to the present invention, it is possible to provide a monitoring model update method, a monitoring system, and a monitoring device that can quickly follow changes in the situation at the site and update the model when updating the monitoring model in real-time monitoring.

The figure which shows the functional structure example of the monitoring system of Example 1. FIG. The figure which shows the hardware configuration example of the monitoring system of Example 1. FIG. The flowchart which shows the whole processing example in the monitoring system of Example 1. The flowchart which shows the real-time monitoring processing example of Example 1. The figure for demonstrating the process example in which the monitoring log is generated in Example 1. FIG. The figure which shows the display screen example of the monitoring result of Example 1. FIG. The flowchart which shows the monitoring model generation processing example of Example 1. The figure for demonstrating the process example in which the model information is created in Example 1. FIG. The flowchart which shows the monitoring model evaluation processing example of Example 1. The figure for demonstrating the monitoring model evaluation processing example of Example 1. FIG. The figure for demonstrating the model accuracy calculation example of the monitoring model of Example 1. FIG. The flowchart which shows the monitoring model update processing example of Example 1. The figure for demonstrating the monitoring model update processing example of Example 1. FIG. The flowchart which shows the tuning process example of the model generation parameter of Example 1. The figure for demonstrating the tuning process example of the model generation parameter of Example 1. FIG. The figure for demonstrating the update example of the data transmission rule of Example 1. FIG. The flowchart which shows the monitoring model deletion processing example of Example 1. The figure for demonstrating the monitoring model deletion processing example of Example 1. FIG. The figure which shows the confirmation screen example of the model information of Example 1. FIG. The hardware diagram which shows the configuration example of the monitoring system of Example 2. FIG.

Examples of the present invention will be described in detail below with reference to the drawings. The following examples are not limited to the present invention. In the following embodiment, a monitoring system that monitors the occurrence of defective products in a product manufactured by this machine in real time based on the sensor data acquired by the IoT sensor installed in the machine at the manufacturing site will be given as an example. However, the present invention is not limited to defective product monitoring, and can be widely applied to a monitoring system that monitors whether or not sensor data meets the conditions specified in the monitoring model.

In each drawing for explaining the following examples, the configuration requirements having the same or similar functions are indicated by the same reference numbers, and the description below will be omitted. In addition, examples and modifications can be combined in part or in whole within the scope of the technical idea of the present invention and within the range consistent with it.

<Functional configuration of monitoring system>
FIG. 1 is a diagram showing a functional configuration example of the monitoring system of the first embodiment. As shown in FIG. 1, the monitoring system 1S of the first embodiment is, for example, based on the sensor data acquired by the IoT sensor provided in the machine at the manufacturing site, the product to be monitored manufactured by this machine. It is a system that monitors the occurrence of defective products in real time. As shown in FIG. 1, the monitoring system 1S includes a monitoring device 10 and an inspection device 20.

The monitoring device 10 includes a real-time monitoring unit 12a, a monitoring model generation unit 12b, a monitoring model evaluation unit 12c, a monitoring model deletion unit 12d, a model update request unit 12e, a monitoring model update unit 12f, and model generation parameter tuning. It has a portion of 12 g. Further, the monitoring device 10 has a monitoring model evaluation DB (Data Base) 11-1 and a model information storage DB 11-2. The monitoring model evaluation DB 11-1 and the model information storage DB 11-2 may be built on the same storage device or may be built on different storage devices.

The real-time monitoring unit 12a of the monitoring device 10 monitors the time-series sensor data 11a collected by the sensor 1 (see FIG. 2) in real time for each data according to the monitoring model 12a1 preset in the real-time monitoring unit 12a. analyse. The sensor data 11a is stored in the monitoring model evaluation DB11-1.

For example, as a result of comparing the manufacturing conditions of the product corresponding to one record of the sensor data 11a with the monitoring model 12a1, the real-time monitoring unit 12a determines that the product is "OK" if it is a non-defective product, and determines that the product is defective. If there is, real-time monitoring that determines "NG" is performed. The real-time monitoring unit 12a transmits the monitoring result 21a to the inspection device 20.

The inspection device 20 has an inspection log generation unit 22a and a monitoring result display unit 25.

The inspection device 20 stores the monitoring result 21a received from the monitoring device 10 in the storage unit 21 (see FIG. 2) and displays it on the monitoring result display unit 25. The monitoring result display unit 25 is a GUI (Graphical User Interface) or the like on the output screen. The inspector confirms the monitoring result displayed on the monitoring result display unit 25, and determines whether the product is "OK" or "NG" for the product determined to be "NG". I do. The inspector operates the input unit 26 (see FIG. 2) such as the keyboard, mouse, and touch panel of the inspection device 20, and inputs the defective product determination result again.

The inspection log generation unit 22a of the inspection device 20 generates an inspection log 21b that reflects the re-defective product determination result input from the input unit 26 in the monitoring result 21a. Further, the inspection log generation unit 22a reflects the defective product determination result input from the input unit 26 again in the monitoring result 21a displayed on the monitoring result display unit 25.

The inspection log 21b generated by the inspection log generation unit 22a is transmitted to the monitoring device 10. The real-time monitoring unit 12a of the monitoring device 10 collects the inspection log 21b and the sensor data 11a received from the inspection device 20 in a certain range of records or period according to the data transmission rule 12a3 preset in the real-time monitoring unit 12a. Model generation data 11ab is generated. The inspection log 21b is stored in the monitoring model evaluation DB11-1.

The monitoring model generation unit 12b analyzes the model generation data 11ab generated by the real-time monitoring unit 12a, and calculates the defective product generation condition for each model generation data 11ab. The monitoring model generation unit 12b generates model information 11c including a defective product generation condition, a model information creation date and time, a period in which model generation data 11ab is collected, and stores the model information 11c in the model information storage DB 11-2.

The monitoring model evaluation unit 12c reads each model information 11c stored in the model information storage DB 11-2 on a regular basis (for example, once a month), and the sensor stored in the monitoring model evaluation DB 11-1. Based on the data 11a and the inspection log 21b, each model information 11c as a monitoring model is evaluated. The monitoring model evaluation unit 12c adds the evaluation result of each model information 11c as a monitoring model to each model information 11c.

The monitoring model deletion unit 12d periodically reads each model information 11c from the model information storage DB 11-2. The monitoring model deletion unit 12d follows a preset deletion rule (for example, the accuracy of the model is lower than a certain value, the model creation date and time is older than a certain period, etc.), and the deletion rule of each read model information 11c The model information 11c corresponding to is deleted from the model information storage DB 11-2.

The model update request unit 12e determines whether or not the monitoring model 12a1 preset in the real-time monitoring unit 12a corresponds to the update rule according to the monitoring model update rule 12a2 preset in the real-time monitoring unit 12a. When the model update request unit 12e determines that the monitoring model 12a1 preset in the real-time monitoring unit 12a corresponds to the update rule, the model update request unit 12e requests the monitoring model update unit 12f to update the monitoring model 12a1.

The monitoring model update unit 12f responds to the update request from the model update request unit 12e, follows the preset monitoring model update rule, and based on the evaluation result of the model information 11c by the monitoring model evaluation unit 12c, the model information 11c Select a monitoring model from. Then, the monitoring model updating unit 12f sets the selected monitoring model in the real-time monitoring unit 12a. When a new monitoring model 12a1 is set in the real-time monitoring unit 12a, the monitoring model updating unit 12f increments the "model application count" of the record of the corresponding model information 11c by +1. The monitoring model update rule is, for example, to select the model information 11c having the best model accuracy in the same month when the update request is generated by the model update request unit 12e.

The model generation parameter tuning unit 12g reads the data transmission rule 12a3 set in the real-time monitoring unit 12a, and updates each model information 11c stored in the model information storage DB 11-2 (model application count or model accuracy). Analyze the optimal data transmission rules based on. The model generation parameter tuning unit 12g creates a new data transmission rule based on the analyzed optimum data transmission rule, and sets it in the real-time monitoring unit 12a.

<Functional configuration of monitoring system>
FIG. 2 is a diagram showing a hardware configuration example of the monitoring system of the first embodiment. The monitoring device 10 and the inspection device 20 are realized by executing a predetermined program on the computer. FIG. 2 is an example in which the monitoring system 1S, the monitoring device 10, and the inspection device 20 whose functional configuration examples are shown in FIG. 1 are illustrated from the side of the hardware configuration.

The monitoring device 10 has a monitoring model evaluation DB 11-1, a model information storage DB, a memory 12, a CPU (Central Processing Unit) 13, and communication IFs (Inter Face) 14-1, 14-2. , These are connected via a bus.

The monitoring model evaluation DB11-1 and the model information storage DB11-2 are databases constructed in the storage area of the storage device. The monitoring model evaluation DB11-1 stores the sensor data 11a and the inspection log 21b. The model information storage DB 11-2 stores the model information 11c.

The monitoring device 10 has each part that realizes each processing function by loading and executing a predetermined program in the memory 12 by the CPU 13. The monitoring device 10 includes a real-time monitoring unit 12a, a monitoring model generation unit 12b, a monitoring model evaluation unit 12c, a monitoring model deletion unit 12d, a model update request unit 12e, a monitoring model update unit 12f, and model generation parameter tuning. It has a portion of 12 g.

The communication IF 14-1 is an interface for connecting the monitoring device 10 to the network 30a. The monitoring device 10 receives the time-series sensor data 11a generated by the sensor 1 via the network 30a. The communication IF 14-2 is an interface for connecting the monitoring device 10 to the network 30b. The monitoring device 10 communicates with the inspection device 20 via the network 30b. The network 30a and the network 30b may be the same network or different networks.

The inspection device 20 has a storage unit 21, a memory 22, a CPU 23, a communication IF 24, a monitoring result display unit 25, and an input unit 26, which are connected via a bus.

The storage unit 21 is a non-volatile storage device, and stores the monitoring result 21a and the inspection log 21b.

The inspection device 20 has each part that realizes each processing function by loading and executing a predetermined program in the memory 22 by the CPU 23. The inspection device 20 has an inspection log generation unit 22a.

The communication IF 24 is an interface for connecting the inspection device 20 to the network 30b. The inspection device 20 communicates with the monitoring device 10 via the network 30b.

<Processing in the monitoring device of Example 1>
FIG. 3 is a flowchart showing an example of overall processing in the monitoring system of the first embodiment. The overall process in the monitoring system of the first embodiment is a process that is periodically repeated.

As shown in FIG. 3, in step S1, the monitoring system 1S executes the real-time monitoring process. Details of the real-time monitoring process will be described later with reference to FIG. Subsequently, in step S2, the monitoring system 1S executes the monitoring model generation process. Details of the monitoring model generation process will be described later with reference to FIG. 7.

Subsequently, in step S3, the monitoring system 1S executes the monitoring model evaluation process. Details of the monitoring model evaluation process will be described later with reference to FIG. Subsequently, in step S4, the monitoring system 1S executes the monitoring model update process. Details of the monitoring model update process will be described later with reference to FIG.

Subsequently, in step S5, the monitoring system 1S executes the tuning process of the model generation parameter. Details of the tuning process of the model generation parameters will be described later with reference to FIG. Subsequently, in step S6, the monitoring system 1S executes the monitoring model deletion process. The details of the monitoring model deletion process will be described later with reference to FIG. When the monitoring model deletion process is completed, the monitoring system 1S ends the entire process in the monitoring device.

<Real-time monitoring process of Example 1>
FIG. 4 is a flowchart showing an example of real-time monitoring processing according to the first embodiment. FIG. 4 shows a flowchart detailing step S1 of FIG. As shown in FIG. 4, first, in step S11, the monitoring system 1S reads the sensor data 11a collected by the sensor 1 in the real-time monitoring unit 12a of the monitoring device 10.

Subsequently, in step S12, the monitoring system 1S analyzes the sensor data 11a read in step S11 according to the preset monitoring model 12a1 in the real-time monitoring unit 12a. In step S12, the real-time monitoring unit 12a monitors whether the value of the sensor data 11a corresponds to the defective product generation condition preset in the real-time monitoring unit 12a. When the value of the sensor data 11a does not correspond to the defective product generation condition preset in the real-time monitoring unit 12a, the real-time monitoring unit 12a sets the monitoring result 21a corresponding to the sensor data 11a as "OK" and generates a defective product. When the condition is met, the monitoring result 21a corresponding to the sensor data 11a is set as "NG".

Subsequently, in step S13, the monitoring system 1S displays the monitoring result 21a by the real-time monitoring unit 12a in step S12 on the monitoring result display unit 25 of the inspection device 20.

When the monitoring result in step S12 is "OK" (step S14: YES), the monitoring system 1S ends the real-time monitoring process and shifts the process to step S2 in FIG. On the other hand, when the monitoring result in step S12 is "NG" (step S14: NO), the monitoring system 1S shifts the process to step S15.

In step S15, the monitoring system 1S reads the monitoring result 21a by the real-time monitoring unit 12a in step S12 in the inspection log generation unit 22a of the inspection device 20. Subsequently, in step S16, the monitoring system 1S generates an inspection log 21b in which the inspection log generation unit 22a reflects the re-defective product determination result input by the inspector in the monitoring result 21a. Further, in step S16, the monitoring system 1S reflects the defective product determination result input from the input unit 26 again in the inspection log generation unit 22a in the monitoring result 21a displayed on the monitoring result display unit 25. When the step S16 is completed, the monitoring system 1S ends the real-time monitoring process and shifts the process to step S2 of FIG.

<Monitoring log generation process>
FIG. 5 is a diagram for explaining an example of a process in which a monitoring log is generated in the first embodiment. FIG. 5 is a data diagram showing that the inspection log 21b is generated from the sensor data 11a via the monitoring result 21a in the real-time monitoring process shown in FIG.

The sensor 1 transmits the sensor data 11a to the real-time monitoring unit 12a of the monitoring device 10 each time data is generated. When the real-time monitoring unit 12a reads the received sensor data 11a, it immediately analyzes it. The real-time monitoring unit 12a generates a monitoring result 21a as a result of analyzing the sensor data 11a.

In the example of FIG. 5, the sensor data 11a includes at least "data acquisition date and time", "product ID", "weight", and "processing speed". In the real-time monitoring unit 12a, for each sensor data 11a corresponding to the product identified by the "product ID", the manufacturing conditions of "weight" and "machining speed" are the conditions for generating defective products of the product defined by the monitoring model 12a1. Perform real-time monitoring to analyze whether or not it corresponds to. If the manufacturing conditions of "weight" and "machining speed" correspond to the defective product conditions of the product defined in the monitoring model 12a1, "NG" is set, and if not, "OK" is set. Then, the real-time monitoring unit 12a outputs the monitoring result 21a including the "data acquisition date and time", the "product ID", and the "system determination result" as the result of the real-time monitoring.

Further, the inspection log generation unit 22a displays the monitoring result 21a output from the real-time monitoring unit 12a on the monitoring result display unit 25. The inspector confirms the monitoring result 21a displayed on the monitoring result display unit 25. The inspector re-inspects the product of the "system determination result" and the "product ID" determined to be "NG" in the monitoring result 21a, and inputs the result from the input unit 26. The inspection log generation unit 22a adds columns of "inspector" and "inspection result" to the monitoring result 21a, and generates an inspection log 21b in which the result of the re-inspection input by the inspector is added.

In the example of FIG. 5, the inspection log 21b includes a "data acquisition date and time", a "product ID", a "system determination result", an "inspector", and an "inspection result". The “data acquisition date and time” and the “product ID” are the same as the record of the corresponding monitoring result 21a.

In the example of FIG. 5, among the monitoring results 21a, the product of "product ID" "A012" which is "system judgment result" "NG" is re-inspected by the inspector ("inspector" "Sato"). "Inspection result" is "OK". Therefore, the inspection log generation unit 22a generates an inspection log 21b in which the "inspector" "Sato" and the "inspection result" "OK" of the records of the "product ID" "A012" are added, and outputs the inspection log 21b to the monitoring device 10. ..

<Monitoring result confirmation screen>
FIG. 6 is a diagram showing an example of a display screen of the monitoring result of the first embodiment. As shown in FIG. 6, the GUI 51a on the screen of the monitoring result display unit 25 of the inspection device 20 sets the "product ID", "data acquisition date and time", and "system determination result" of each record included in the monitoring result 21a. indicate. Further, the GUI 51a displays the "inspector" and the "inspection result" associated with each record included in the monitoring result 21a so that the inspector can input and display the re-inspection result of the product. Further, the GUI 51a displays the "product ID", the "data acquisition date and time", the "determination result" of the latest monitoring result 21a, and the record including these so that they can be easily identified.

The inspector confirms the "system judgment result" of the monitoring result 21a displayed on the GUI 51a, and re-determines the defective product for the product judged as "NG", and checks the result of the "inspection" in the corresponding line. Enter in "Person" and "Test result". The inspection log 21b is generated by reflecting the information of the "inspector" and the "inspection result" input from the GUI 51a in the monitoring result 21a.

<Monitoring model generation process of Example 1>
FIG. 7 is a flowchart showing an example of the monitoring model generation process of the first embodiment. FIG. 7 shows a flowchart detailing step S2 of FIG. The monitoring model generation process is a process of generating update candidates for the monitoring model in advance.

As shown in FIG. 7, first, in step S21, the monitoring system 1S reads the sensor data 11a received from the sensor 1 and the inspection log 21b received from the inspection device 20 in the real-time monitoring unit 12a of the monitoring device 10. The real-time monitoring unit 12a transmits each of the read sensor data 11a and the inspection log 21b to the monitoring model generation unit 12b in a unit collected according to the data transmission rule 12a3 set in advance in the real-time monitoring unit 12a.

Subsequently, in step S22, the monitoring system 1S transmits the sensor data 11a and the inspection log 21b (hereinafter referred to as model generation data 11ab) transmitted from the real-time monitoring unit 12a in accordance with the data transmission rule 12a3 in the monitoring model generation unit 12b. Read. The details of the model generation data 11ab will be described later.

Subsequently, in step S23, the monitoring system 1S analyzes the model generation data 11ab received from the real-time monitoring unit 12a in the monitoring model generation unit 12b, and calculates the defective product occurrence condition and the like. Then, the monitoring system 1S adds the creation date and time of the record of the model information 11c including the analysis result such as the defective product occurrence condition and the target period of the model generation data 11ab in the monitoring model generation unit 12b. To generate.

Subsequently, in step S24, the monitoring system 1S stores the record of the model information 11c created in step S23 in the model information storage DB 11-2 in the monitoring model generation unit 12b. When the step S24 is completed, the monitoring system 1S ends the monitoring model generation process and shifts the process to step S3 of FIG.

<Model information generation process>
FIG. 8 is a diagram for explaining a process example in which model information is created in the first embodiment. FIG. 8 is a data diagram showing that model information 11c is generated from the sensor data 11a and the inspection log 21b via the model generation data 11ab in the monitoring model generation process shown in FIG. 7.

The real-time monitoring unit 12a of the monitoring device 10 transmits the sensor data 11a received from the sensor 1 and the inspection log 21b received from the inspection device 20 periodically (for example, every time the inspection log 21b is generated) to the real-time monitoring unit 12a. The data is adjusted to the range of data according to the preset data transmission rule 12a3 and summarized. Then, the real-time monitoring unit 12a generates the model generation data 11ab summarized in the range of this data. The range of this data includes, for example, a data period such as the latest M time, ..., The latest N time data, or the number of data such as the latest X data, ..., the latest Y data.

Specifically, it is assumed that the data transmission rule 12a3 is preset with the data transmission rule 12a3 that collectively transmits data in the latest M time and the latest N time. In this case, the real-time monitoring unit 12a collects the sensor data 11a whose data acquisition date and time is the latest M time as a sensor data group of the latest M time, and inspects the record of the inspection log 21b whose data acquisition date and time is the latest M time. Collect as a group of log records. Then, the real-time monitoring unit 12a sets up a sensor data group for the latest M time and an inspection log record group for the latest M time.

Similarly, the real-time monitoring unit 12a sets up a sensor data group for the latest N hours and an inspection log record group for the latest N hours. In this way, the real-time monitoring unit 12a includes a set of the sensor data group of the latest M time and the inspection log record group of the latest M time, and a set of the sensor data group of the latest N time and the inspection log record group of the latest N time. The model generation data 11ab including the above is generated and transmitted to the monitoring model generation unit 12b.

Then, the monitoring model generation unit 12b calculates the defective product occurrence condition and the like for the set of the sensor data group of the latest M time and the inspection log record group of the latest M time in the received model generation data 11ab. Then, the monitoring model generation unit 12b records the model information 11c including the analysis result such as the defective product occurrence condition and the data period of "latest M time" with the creation date and time of the record and the ID "A" of the record. Is added and generated, and stored in the model information storage DB 11-2.

The monitoring model generation unit 12b processes the set of the sensor data group of the latest N time and the inspection log record group of the latest N time in the same manner as the set of the sensor data group of the latest M time and the inspection log record group of the latest M time. Then, a record of model information 11c (record ID “B”) is generated and stored in the model information storage DB 11-2. In this way, the monitoring model generation unit 12b periodically (for example, once a day) generates a record of model information 11c. The set of values in each column of "ID", "model creation date and time", "data period", and "defective product occurrence condition" of each record of model information 11c is an update candidate of the monitoring model 12a1.

<Monitoring model evaluation process of Example 1>
FIG. 9 is a flowchart showing an example of the monitoring model evaluation process of the first embodiment. FIG. 9 shows a flowchart detailing step S3 of FIG. As shown in FIG. 9, first, in step S31, the monitoring system 1S reads the sensor data 11a from the sensor 1 and the inspection log 21b from the inspection device 20 in the monitoring model evaluation unit 12c of the monitoring device 10.

Subsequently, in step S32, the monitoring system 1S stores the sensor data 11a and the inspection log 21b read in step S31 in the monitoring model evaluation DB 11-1 in the monitoring model evaluation unit 12c. For example, the monitoring model evaluation unit 12c reads the sensor data 11a and the inspection log 21b for a certain period (for example, one month) and stores them in the monitoring model evaluation DB 11-1.

Subsequently, in step S33-1 and step S33-2, the monitoring system 1S periodically (for example, once a month) in the monitoring model evaluation unit 12c, performs sensor data from the monitoring model evaluation DB 11-1. Read 11a and inspection log 21b.

Subsequently, in step S34, the monitoring system 1S reads all the monitoring models stored in the model information storage DB 11-2 in the monitoring model evaluation unit 12c, calculates the accuracy of the read monitoring model, and calculates the accuracy. Is written in the record of the corresponding model information 11c in the model information storage DB 11-2. Details of the calculation of the accuracy of the monitoring model will be described later. When the step S34 is completed, the monitoring system 1S ends the monitoring model evaluation process and shifts the process to step S4 of FIG.

FIG. 10 is a diagram for explaining a monitoring model evaluation processing example of the first embodiment. FIG. 11 is a diagram for explaining a model accuracy calculation example of the monitoring model of the first embodiment. A specific example of calculating the model accuracy of the monitoring model in the monitoring model evaluation process of the first embodiment shown in FIG. 9 will be described with reference to FIG.

As shown in FIG. 10, the monitoring model evaluation unit 12c stores the sensor data 11a and the inspection log 21b in the monitoring model evaluation DB 11-1 at regular intervals. Then, the monitoring model evaluation unit 12c periodically reads all the monitoring models stored in the model information storage DB 11-2.

Then, the monitoring model evaluation unit 12c determines the accuracy of all the monitoring models read from the model information storage DB 11-2, and the above-mentioned sensor data 11a and inspection log 21b for a certain period of time stored in the monitoring model evaluation DB 11-1. Evaluate using.

The accuracy evaluation of the monitoring model will be described with reference to FIG. As shown in FIG. 11, sensor data of "data acquisition date and time" "2017/11/15 16:58:42", "product ID" "A011", "weight" "490", "machining speed" "10" 11a is determined to be "non-defective" (OK) according to the monitoring model of "ID" "A" (defective product generation condition: weight> 500 & processing speed> 35). Further, in the record of the inspection log 21b of the corresponding "data acquisition date and time" "2017/11/15 16:58:42" and "product ID" "A011", the "system determination result" is "OK". Therefore, the determination result according to the monitoring model A and the "system determination result" recorded in the inspection log 21b both match with "non-defective" (OK).

In the records of the inspection log 21b of "data acquisition date and time" "2017/11/15 16:58:42" and "product ID" "A011", information is stored in "inspector" and "inspection result". Since there is no such thing, the "system judgment result" is used.

On the other hand, the sensor data 11a of "data acquisition date and time" "2017/11/15 17:10:00", "product ID" "A012", "weight" "510", "machining speed" "45" is "ID". According to the monitoring model of "A", it is determined as "defective" (NG). On the other hand, in the record of the inspection log 21b of the corresponding "data acquisition date and time" "2017/11/15 17:10:00" and "product ID" "A012", "inspection result" is "OK". Therefore, the determination result according to the monitoring model A and the "inspection result" recorded in the inspection log 21b do not match.

In the records of the inspection log 21b of "data acquisition date and time" "2017/11/15 17:10:00" and "product ID" "A012", information is stored in "inspector" and "inspection result". Therefore, the "inspection result" is used in preference to the "system judgment result".

In this way, for each monitoring model, the result of determining a plurality of sensor data 11a for a certain period according to this monitoring model and the "system determination result" or "inspection result" stored in the record of the corresponding inspection log 21b ( If there is an "inspection result", priority is given to the "inspection result"). Then, the percentage at which the comparison results of the two match is calculated. As shown in FIG. 10, the calculated percentage is recorded in the "model accuracy (correct answer rate)" of the record of the model information 11c corresponding to the monitoring model to be evaluated together with the "model evaluation date and time".

For example, for the monitoring model A, 100 sensor data 11a and the corresponding inspection log 21b are compared, and if 85 items match, the "model accuracy (correct answer rate)" is "85%". Each time the monitoring model is newly evaluated, the "model evaluation date and time" and the "model accuracy (correct answer rate)" are additionally recorded in the record of the additional model information 11c.

<Monitoring model update process of Example 1>
FIG. 12 is a flowchart showing an example of the monitoring model update process of the first embodiment. FIG. 12 shows a flowchart detailing step S4 of FIG. As shown in FIG. 12, first, in step S41, the monitoring system 1S reads the sensor data 11a from the sensor 1 in the real-time monitoring unit 12a.

Subsequently, in step S42, the monitoring system 1S determines in the real-time monitoring unit 12a whether or not it applies to the monitoring model update rule 12a2 set in advance in the real-time monitoring unit 12a. When the monitoring model update rule 12a2 set in the real-time monitoring unit 12a is applied (step S42: YES), the monitoring system 1S shifts the process to step S43. On the other hand, when the monitoring model update rule 12a2 set in the real-time monitoring unit 12a is not applied (step S42: NO), the monitoring system 1S ends the monitoring model update process and shifts the process to step S5 of FIG.

In step S43, the monitoring system 1S requests the monitoring model update unit 12f to update the monitoring model in the model update request unit 12e. Subsequently, in step S44, the monitoring system 1S reads all the model information 11c from the model information storage DB 11-2 in the monitoring model update unit 12f.

Subsequently, in step S45, the monitoring system 1S selects the monitoring model to be updated according to the evaluation results of the monitoring model (“model evaluation date and time” and “model accuracy”) in the monitoring model updating unit 12f. Subsequently, in step S46, the monitoring system 1S converts the monitoring model selected in step S45 into a format that can be set in the real-time monitoring unit 12a in the monitoring model updating unit 12f.

Subsequently, in step S47, the monitoring system 1S sets the new monitoring model 12a1 whose format has been converted in step S46 in the real-time monitoring unit 12a in the monitoring model updating unit 12f. When the monitoring system 1S finishes step S47, the monitoring model update process ends, and the process shifts to step S5 of FIG.

FIG. 13 is a diagram for explaining a monitoring model update processing example of the first embodiment. The monitoring model update process of FIG. 12 will be described with reference to FIG. 13 with reference to a specific example. For example, as shown in FIG. 13, it is assumed that the real-time monitoring unit 12a is set with the monitoring model 12a1 of "ID = A, defective product generation condition: weight> 500 & processing speed> 35". Further, it is assumed that the monitoring model update rule 12a2 of "defect occurrence frequency> 10 cases / hour" is set in the real-time monitoring unit 12a.

The real-time monitoring unit 12a counts the number of defects in the last hour. The model update request unit 12e is a monitoring model update unit when, for example, the number of defects exceeds the threshold value of “10 cases / hour” according to the monitoring model update rule 12a2 of “defect occurrence frequency> 10 cases / hour”. We request 12f to update the monitoring model 12a1. When the number of defects exceeds the threshold value of the monitoring model update rule 12a2, it is considered that the accuracy of the monitoring model 12a1 is lowered and it does not follow the change in the situation at the site.

When the monitoring model update unit 12f receives the update request of the monitoring model 12a1 from the model update request unit 12e, the "model evaluation date and time" and "model" of each record of the model information 11c stored in the model information storage DB 11-2 Refer to the "Accuracy (correct answer rate)" set, and select and read the monitoring model that corresponds to the preset monitoring model selection rule for updating. The preset monitoring model selection rule for update is, for example, to select the monitoring model having the highest "model accuracy (correct answer rate)" in the same month as when the update is requested.

In the example of FIG. 13, assuming that it is “November” that the model update request unit 12e requests the monitoring model update unit 12f to update the monitoring model 12a1, the monitoring model update unit 12f determines the “model evaluation date and time”. "November" is "November" "ID" is "ID" "ID" which is the best "90%" among the records of model information 11c from "A" to "D" Select the monitoring model corresponding to the record of "D".

Then, the monitoring model update unit 12f increments the "model application count" of the record of the model information 11c corresponding to the selected monitoring model. As a result, the "model application count" indicates the actual number of times the corresponding monitoring model was actually applied. Then, the monitoring model updating unit 12f uses the selected monitoring model of "ID" and "D" as a new monitoring model 12a11 to replace the monitoring model 12a1 of "ID" and "A" set in the real-time monitoring unit 12a. Set in the real-time monitoring unit 12a. By setting the new monitoring model 12a11 in the real-time monitoring unit 12a, it is possible to reduce the frequency of false determinations of "system judgment" and "OK" in the real-time monitoring unit 12a as "NG" in real time. The accuracy of real-time monitoring by the monitoring unit 12a can be improved.

<Modified example of monitoring model update processing of Example 1>
In the first embodiment, as described above, the monitoring system 1S automatically updates the monitoring model 12a1 when the condition of the monitoring model update rule 12a2 preset in the real-time monitoring unit 12a is satisfied. Not limited to this, regardless of the monitoring model update rule 12a2, the "model accuracy" is set to the real-time monitoring unit 12a on a regular basis (for example, every time the evaluation process of the monitoring model is executed) with the best monitoring model. The monitoring model 12a1 may be updated automatically.

Alternatively, an appropriate operator selects from the confirmation result of the screen display of the model information 11c stored in the model information storage DB 11-2 based on information such as "model evaluation date and time" and "model accuracy (correct answer rate)". In the monitoring model, the monitoring model 12a1 set in the real-time monitoring unit 12a may be updated.

<Tuning process of model generation parameters of Example 1>
FIG. 14 is a flowchart showing an example of tuning processing of the model generation parameter of the first embodiment. FIG. 14 shows a flowchart detailing step S5 of FIG. As shown in FIG. 14, first, in step S51-1 and step S51-2, the monitoring system 1S stores the data transmission rule 12a3 and the model information set in the real-time monitoring unit 12a in the model generation parameter tuning unit 12g. All model information 11c stored in DB 11-2 is read.

Subsequently, in step S52, the monitoring system 1S analyzes the optimum data transmission rule in the model generation parameter tuning unit 12g, and creates a new data transmission rule based on the analysis result. Subsequently, in step S53, the monitoring system 1S sets the new data transmission rule 12a31 created in step S52 in the real-time monitoring unit 12a in the model generation parameter tuning unit 12g. When step S53 is completed, the monitoring system 1S ends the tuning process of the model generation parameter, and shifts the process to step S6 of FIG.

FIG. 15 is a diagram for explaining an example of tuning processing of the model generation parameter of the first embodiment. FIG. 16 is a diagram for explaining an example of updating the data transmission rule of the first embodiment. The tuning process of the model generation parameter of FIG. 14 will be described with reference to FIGS. 15 and 16 with reference to a specific example.

As shown in FIG. 15, the model generation parameter tuning unit 12g first reads the data transmission rule 12a3 set in the real-time monitoring unit 12a and the model information 11c stored in the model information storage DB 11-2. Next, the model generation parameter tuning unit 12g analyzes the optimum data transmission rule based on the update process (for example, "model application count" or "model accuracy (correct answer rate)") recorded in the model information 11c. .. In the example of FIG. 15, the "data period" and "6 hours" of "ID" and "D", in which the "model application count" is "1", which is the largest, and the "model accuracy (correct answer rate)" is the best, "90%". "(6 hours) is analyzed as the optimal modeling data range.

In this case, as shown in FIG. 16A, the data transmission rule 12a3 before the update is the latest 4 hours, 6 hours, as in "the latest 4 hours data, the latest 6 hours data, the latest 8 hours data". , And the sensor data 11a and the inspection log 21b summarized in three data periods of 8 hours. The data range for optimal model creation and the analyzed "6 hours" are set to the median value "data for the last 6 hours" of the three data periods in the data transmission rule 12a3 before the update in FIG. 16 (a). Applicable. That is, the data range for model creation is set in the data transmission rule 12a3 as the median value "data for the last 6 hours".

Therefore, while maintaining this "data of the last 6 hours" as the median value, the "data of the last 4 hours" and the "data of the last 8 hours" are centered between the "data of the last 6 hours". Update with the values "data for the last 5 hours" and "data for the last 7 hours". That is, as shown in FIG. 16A, the updated data transmission rule 12a3 is updated as "data of the latest 5 hours, data of the latest 6 hours, data of the latest 7 hours".

Alternatively, as shown in FIG. 16B, the data transmission rule 12a3 before the update is the latest 2 hours, 4 hours, as in "the latest 2 hours data, the latest 4 hours data, the latest 6 hours data". And it is assumed that the sensor data 11a and the inspection log 21b summarized in the data period of 6 hours are transmitted. The data range of the model creation and the analyzed "6 hours" correspond to the maximum value "data of the latest 6 hours" among the three data periods in the data transmission rule 12a3 before the update of FIG. 16 (b). .. That is, the optimum model transmission rule is set in the data transmission rule 12a3 as the maximum value "data of the last 6 hours".

Therefore, this "data of the last 6 hours" is changed to be the median value, and the "data of the last 2 hours" and "data of the last 4 hours" are the difference values of the data periods before and after. Based on "2 hours", it is updated with "data of the last 4 hours" and "data of the last 8 hours", respectively. That is, as shown in FIG. 16A, the updated data transmission rule 12a3 is updated as "data of the latest 4 hours, data of the latest 6 hours, data of the latest 8 hours".

If the analyzed optimal model creation data range is set as the minimum value in the data transmission rule 12a3 before update, the minimum value is changed to be the center value, and the following is described above. It is the same as "when the data range of the optimum model creation is set in the data transmission rule 12a3 as the maximum value".

By updating the data transmission rule 12a3 in this way, the data transmission rule 12a3 approaches the optimum model transmission rule for each update, so that a more accurate monitoring model can be generated.

The above is merely an example of updating the data transmission rule 12a3, and the data period (or the number of data) for collecting the data included in the data transmission rule 12a3 is not limited to the above-mentioned "3" and can be changed as appropriate. ..

<Monitoring model deletion process of Example 1>
FIG. 17 is a flowchart showing an example of the monitoring model deletion process of the first embodiment. FIG. 17 shows a flowchart detailing step S6 of FIG. As shown in FIG. 17, first, in step S61, the monitoring system 1S reads all the records of the model information 11c from the model information storage DB 11-2 in the monitoring model deletion unit 12d. Subsequently, in step S62, the monitoring system 1S corresponds to the deletion rule among the records of the model information 11c stored in the model information storage DB 11-2 according to the deletion rule set in advance in the monitoring model deletion unit 12d. Delete the record you want. When the step S62 is completed, the monitoring system 1S ends the monitoring model deletion process, returns the process to FIG. 3, and ends the entire process in the monitoring system 1S.

FIG. 18 is a diagram for explaining a monitoring model deletion processing example of the first embodiment. A specific example of the monitoring model deletion process of the first embodiment shown in FIG. 17 will be described with reference to FIG.

As shown in FIG. 18, first, the monitoring model deletion unit 12d periodically reads all the records of the model information 11c from the model information storage DB 11-2. It is assumed that the monitoring model deletion rule is set in advance in the monitoring model deletion unit 12d. The monitoring model deletion rule includes that the "model accuracy (correct answer rate)" of the record of the model information 11c is a certain value or less, the "model creation date and time" is older than a certain period, and the like. Alternatively, as the monitoring model deletion rule, it may be adopted that the "model creation date and time" is older than a certain period of time and the "model application count" is more than a certain number of times.

The monitoring model deletion unit 12d deletes the record corresponding to the monitoring model deletion rule among the records of the model information 11c read from the model information storage DB 11-2. In the example shown in FIG. 18, it is assumed that ““ model accuracy (correct answer rate) ”is 80% or less” is set as the monitoring model deletion rule. The monitoring model deletion unit 12d has the “ID” shown in FIG. Assuming that the record of the model information 11c of "A" to "D" is read, the records of "ID", "A" and "C" correspond to the monitoring model deletion rule, and therefore the records are deleted.

In this way, by deleting the record of the model information 11c corresponding to the monitoring model deletion rule from the model information storage DB 11-2, the capacity of the model information storage DB 11-2 can be effectively used, and the highly effective update candidate can be used. Many monitoring models can be prepared.

<Confirmation screen of model information of Example 1>
FIG. 19 is a diagram showing an example of a confirmation screen for model information of the first embodiment. In the example shown in FIG. 19, the GUI 52a on the screen of the display device (not shown) connected to the monitoring device 10 is the “ID” of each record included in the model information 11c stored in the model information storage DB 11-2. The values of "model creation date and time", "model creation period", "model evaluation date and time", and "model accuracy" are displayed as a monitoring model and its evaluation information. Further, the GUI 52a displays the "ID", "model creation date and time", "model accuracy" of the latest monitoring model corresponding to the record of the latest model information 11c, and the record including these so that they can be easily identified. ..

By referring to the GUI 52a, the operator of the monitoring device 10 executes the monitoring model generation process (see FIG. 7), the monitoring model evaluation process (see FIG. 9), the monitoring model deletion process (see FIG. 17), and the like to obtain model information. It is possible to confirm the change of the model information 11c including the addition of the record, the addition of the information of the model accuracy, the deletion of the record, and the like.

Further, by including the "defective product occurrence condition" in the display of the GUI 52a, the operator of the monitoring device 10 may be able to confirm the details of the update candidate of the monitoring model.

<Effect of Example 1>
According to the above-described first embodiment, by creating update candidates for the monitoring model 12a1 set in the real-time monitoring unit in advance, it is possible to quickly follow changes in the on-site conditions including product specification changes, equipment changes, and seasonal changes. The monitoring model 12a1 can be updated in a short time.

Further, the accuracy of the created monitoring model 12a1 as an update candidate is evaluated in advance, and when updating the monitoring model 12a1, the update candidate of the monitoring model is selected based on the accuracy, so that the accuracy is higher. It can be updated to a higher monitoring model.

Further, since the generation rule for generating the update candidate of the monitoring model such as the data transmission rule 12a3 approaches the optimum rule by being updated in a timely manner, it is possible to generate the update candidate of the monitoring model with higher accuracy. ..

FIG. 20 is a hardware diagram showing a configuration example of the monitoring system of the second embodiment. The monitoring system 2S of the second embodiment is different in that the monitoring device 10 (see FIG. 2) of the first embodiment is separated into the monitoring device 10B and the model update device 40. The monitoring system 2S of the second embodiment may include a plurality of monitoring devices 10B, each of which is arranged at each base. Further, the monitoring system 2S of the second embodiment may include a plurality of inspection devices 20B, each of which is arranged at each base. In FIG. 20, the same reference numerals are given to the same configurations as those in the first embodiment, and the description thereof will be omitted.

As shown in FIG. 20, the monitoring system 2S of the second embodiment includes one or more monitoring devices 10B, an inspection device 20B, and a model update device 40.

The monitoring device 10B has a storage unit 11, a memory 12, a CPU 13, and communication IFs 14-1, 14-2, and 14-3, which are connected via a bus. The communication IF14-3 is an interface for connecting the monitoring device 10B to the network 30c. The monitoring device 10B communicates with the model update device 40 via the network 30c.

The storage unit 11 stores the sensor data 11a received from the sensor 1 and the inspection log 21b received from the inspection device 20B. The monitoring device 10B has a real-time monitoring unit 12a on the memory 12.

The inspection device 20B has a storage unit 21, a memory 22, a CPU 23, communication IFs 24 and 24-2, a monitoring result display unit 25, and an input unit 26, which are connected via a bus. .. The communication IF24-2 is an interface for connecting the inspection device 20B to the network 30d. The inspection device 20B communicates with the model update device 40 via the network 30d.

The networks 30b, 30c, and 30d may be the same network, any two of them may be the same network and the other one may be different, or all of them may be different networks.

The monitoring result 21a and the inspection log 21b are stored in the storage unit 21. The inspection device 20B has an inspection log generation unit 22a on the memory 22.

The model update device 40 has a monitoring model evaluation DB11-1, a model information storage DB11-2, a memory 42, a CPU 43, and communication IF44-1, 44-2, 44-3, which are buses. It is connected via.

The communication IF44-1 is an interface for connecting the model update device 40 to the network 30a. The model update device 40 receives the sensor data 11a from the sensor 1 via the network 30a. Further, the communication IF44-2 is an interface for connecting the model update device 40 to the network 30c. The model update device 40 communicates with the monitoring device 10B via the network 30c. Further, the communication IF44-3 is an interface for connecting the model update device 40 to the network 30d. The model update device 40 communicates with the inspection device 20B via the network 30d.

The monitoring model evaluation DB 11-1 stores the sensor data 11a received from the sensor 1 and the inspection log 21b received from the inspection device 20B. The model information 11c is stored in the model information storage DB 11-2. The model update device 40 has a monitoring model generation unit 12b, a monitoring model evaluation unit 12c, a monitoring model deletion unit 12d, a model update request unit 12e, a monitoring model update unit 12f, and model generation parameter tuning on the memory 42. It has 12 g of a portion.

Each part of the model update device 40 on the memory 42 may be appropriately distributed and mounted on different devices.

In the second embodiment, for example, the monitoring device 10B is arranged in each of the business establishments such as factories distributed in various places to inspect the defective product in real time, and the sensor data 11a and the inspection log 21b are subjected to the model updating device 40. The model information 11c is generated from the aggregated sensor data 11a and the inspection log 21b. Then, based on the model information 11c for each base, the monitoring model 12a1 set in the monitoring device 10B for each base is updated with the monitoring model generated by the model update device 40.

<Effect of Example 2>
According to the above second embodiment, the bases for real-time monitoring are dispersed, and even if there is a difference in the situation change at the site for each base, the monitoring model is updated by quickly following the situation change for each base. be able to.

The present invention is not limited to the above-described examples, but includes various modifications. For example, the above-described embodiment has been described in detail in order to explain the present invention in an easy-to-understand manner, and is not necessarily limited to the one having the described configuration. Further, as long as there is no contradiction, it is possible to replace a part of the configuration of one embodiment with the configuration of another embodiment, or to add the configuration of another embodiment to the configuration of one embodiment. In addition, it is possible to add, delete, replace, integrate, and divide a part of the configuration of each embodiment.

1: Sensor, 1S, 2S: Monitoring system, 10,10B: Monitoring device, 11a: Sensor data, 11ab: Model generation data, 11c: Model information, 12a: Real-time monitoring unit, 12a1: Monitoring model, 12a2: Monitoring model Update rule, 12a3: Data transmission rule, 12b: Monitoring model generation unit, 12c: Monitoring model evaluation unit, 12d: Monitoring model deletion unit, 12e: Model update request unit, 12f: Monitoring model update unit, 12g: Model generation parameter tuning Units 20, 20B: Inspection device, 21a: Monitoring result, 21b: Inspection log, 22a: Inspection log generation unit, 40: Model update device, 51a, 52a: GUI

Claims (12)

It is a monitoring model update method executed by the monitoring system.
Based on the data related to the monitoring target collected by the sensor and the monitoring model set in advance for determining the abnormality of the monitoring target, the monitoring step of determining the abnormality of the monitoring target in real time, and the monitoring step.
Based on a predetermined generation rule, a plurality of model generation data in which the data and the time-series data including the determination result corresponding to the data by the monitoring step are collected are generated, and the monitoring target is generated from the model generation data. A monitoring model generation step that periodically generates update candidates for the monitoring model, including information on the determination accuracy for determining an abnormality, and a monitoring model generation step.
A monitoring model storage step for storing the update candidate generated by the monitoring model generation step in the monitoring model storage unit, and a monitoring model storage step.
An update step for updating the preset monitoring model using the update candidate stored in the monitoring model storage unit as a new monitoring model, and
When the monitoring model is updated by the update step, it has a change step of changing the predetermined generation rule based on the model generation data used by the update candidate among the plurality of model generation data. A monitoring model update method characterized by. The update step
When the determination result by the monitoring step corresponds to the update condition for updating the monitoring model, the monitoring model is selected from the update candidates stored in the monitoring model storage unit based on a predetermined selection condition. The monitoring model updating method according to claim 1, wherein the monitoring model is updated. Further having a step of adjusting the data range of the time series data,
The monitoring model according to claim 1, wherein in the monitoring model generation step, the update candidate is generated based on the model generation data in which the time series data is collected in the adjusted data range. How to update. The monitoring model update method according to claim 3, wherein the data range is a data period of the time series data. The monitoring model update method according to claim 3, wherein the data range is the number of data of the time series data. The first aspect of claim 1, wherein the update candidate corresponding to a predetermined deletion condition among the update candidates stored in the monitoring model storage unit is further provided with a deletion step of deleting the update candidate corresponding to the predetermined deletion condition from the monitoring model storage unit. How to update the monitoring model. The predetermined generation rule is
The claim is characterized in that, when the update candidate stored in the monitoring model storage unit is set as the monitoring model, it is determined based on the determination accuracy for determining the abnormality of the monitoring target in the monitoring step. The monitoring model update method according to 1. The predetermined generation rule is
The monitoring model update method according to claim 1, wherein the update candidate stored in the monitoring model storage unit is determined based on the number of actual results set in the past as the monitoring model. The monitoring model update method according to claim 1, further comprising a display step for displaying the update candidate generated by the monitoring model generation step. With one or more monitoring units that determine the abnormality of the monitoring target in real time based on the data related to the monitoring target collected by the sensor and the monitoring model set in advance for determining the abnormality of the monitoring target. ,
Based on a predetermined generation rule, a plurality of model generation data in which the data and the time-series data including the determination result corresponding to the data by the monitoring unit are collected are generated, and the monitoring target is generated from the model generation data. A monitoring model generation unit that periodically generates update candidates for the monitoring model, including information on the determination accuracy for determining an abnormality.
A monitoring model storage unit that stores the update candidates generated by the monitoring model generation unit,
An update unit that updates the preset monitoring model, using the update candidate stored in the monitoring model storage unit as a new monitoring model,
When the monitoring model is updated by the update unit, it has a change unit that changes the predetermined generation rule based on the model generation data used by the update candidate among the plurality of model generation data. A monitoring system featuring. The update part
When the determination result by the monitoring unit corresponds to the update condition for updating the monitoring model, the monitoring model is selected from the update candidates stored in the monitoring model storage unit based on a predetermined selection condition. The monitoring system according to claim 10, wherein the monitoring system is updated. Based on the time-series data related to the monitoring target collected by the sensor and the monitoring model set in advance to determine the abnormality of the monitoring target, the monitoring unit that determines the abnormality of the monitoring target in real time,
Based on a predetermined generation rule, a plurality of model generation data in which the data and the time-series data including the determination result corresponding to the data by the monitoring unit are collected are generated, and the determination accuracy for determining the abnormality of the monitoring target is determined. A monitoring model generator that periodically generates update candidates for the monitoring model that includes the information in
A monitoring model storage unit that stores the update candidates generated by the monitoring model generation unit,
An update unit that updates the preset monitoring model, using the update candidate stored in the monitoring model storage unit as a new monitoring model,
When the monitoring model is updated by the update unit, it has a change unit that changes the predetermined generation rule based on the model generation data used by the update candidate among the plurality of model generation data. A monitoring device characterized by.

Images