Detection device, detection method, and computer-readable recording medium

Abstract

The present invention relates to a detection device, a detection method, and a computer-readable recording medium for improving the accuracy of a learning model. A server device (10) collects measurement data measured by a field device (20), inputs the collected measurement data to a learned detection model that predicts a predetermined phenomenon on the basis of the input of the measurement data, and acquires an output result of the learned detection model. When the output result of the learned detection model is different from the determination result of the operator (W) which confirms that the field device (20) for which the predetermined phenomenon has been predicted is different, the tag value input by the operator (W) and the collected measurement data are used to perform re-learning of the learned detection model and generate a re-learning model.

Description

Technical Field

[0001] The present invention relates to a detection device, a detection method and a computer-readable recording medium. Background Art

[0002] The following technology is known, namely, unsupervised machine learning is performed based on normal data collected in advance, a normal state model is created by extracting features of the normal data, and abnormal conditions of the workshop are detected based on the learned normal state model and input data.

[0003] Patent Document 1: Japanese Patent Application Laid-Open No. 11-338848

[0004] However, in the above-mentioned technology, it is difficult to improve the accuracy of the machine learning model (appropriately referred to as "learning model") for detecting the signs of abnormal conditions (appropriately referred to as "abnormal signs"). For example, in the above-mentioned technology, the pre-collected data is not labeled, so it is impossible to calculate the evaluation index of the learning model, and it is difficult to determine whether the learning model used is appropriate. Summary of the invention

[0005] The present invention has been made in view of the above-mentioned circumstances, and an object of the present invention is to improve the accuracy of a learning model for detecting a sign of abnormality.

[0006] The present invention provides a detection device, which comprises: a collection unit, which collects measurement data measured by a measuring device; a detection unit, which inputs the collected measurement data into a learning model that has been learned to predict a specified phenomenon based on the input of the measurement data, and obtains the output of the learned learning model, i.e., a detection result; and an execution unit, which, when the output result of the learned learning model is different from the judgment result of a user of the measuring device that confirms that the specified phenomenon is predicted, uses a label value input by the user and the collected measurement data to perform re-learning of the learned learning model to generate a re-learned model.

[0007] In addition, the present invention provides a detection method, which is a detection method in which a computer performs the following processing, namely, measurement data measured by a measuring device is collected, the collected measurement data is input into a learning model that has been learned to predict a specified phenomenon based on the input of the measurement data, and an output result of the learned learning model is obtained. When the output of the learned learning model, i.e., the detection result, is different from a judgment result of a user of the measuring device that confirms that the specified phenomenon is predicted, the learned learning model is re-learned using a label value input by the user and the collected measurement data to generate a re-learned model.

[0008] In addition, the present invention provides a computer-readable recording medium, which records a detection program, and the detection program enables a computer to perform the following processing, namely, to collect measurement data measured by a measuring device, input the collected measurement data into a learning model that has been learned to predict a specified phenomenon based on the input of the measurement data, obtain the output of the learned learning model, that is, the detection result, and when the output result of the learned learning model is different from the judgment result of a user of the measuring device that confirms that the specified phenomenon is predicted, use the label value input by the user and the collected measurement data to perform re-learning of the learned learning model to generate a re-learned model.

[0009] Effects of the Invention

[0010] According to the present invention, there is an effect that the accuracy of the learning model can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 It is a diagram showing a configuration example and a processing example of the abnormality sign detection system according to the embodiment.

[0012] Figure 2 It is a block diagram showing a configuration example of each device of the abnormality sign detection system according to the embodiment.

[0013] Figure 3 It is a diagram showing a detection result storage unit of a server device according to an embodiment.

[0014] Figure 4 It is a diagram showing a determination result storage unit of the server device according to the embodiment.

[0015] Figure 5 It is a diagram showing an evaluation index storage unit of a server device according to an embodiment.

[0016] Figure 6 It is a diagram showing a detection model storage unit of a server device according to an embodiment.

[0017] Figure 7 This is a diagram showing an example of a detection model involved in an embodiment.

[0018] Figure 8 It is a diagram showing a specific example 1 of the display screen of the operator terminal according to the embodiment.

[0019] Fig. 9 It is a diagram showing a specific example 2 of the display screen of the operator terminal involved in the embodiment.

[0020] Fig.10 It is a diagram showing a specific example 3 of the display screen of the operator terminal involved in the embodiment.

[0021] Fig.11 This is a flowchart showing an example of the overall processing flow of the abnormality sign detection system according to the embodiment.

[0022] Fig.12 This is a flowchart showing an example of the flow of the detection model building process of the abnormality sign detection system according to the embodiment.

[0023] Fig.13 This is a flowchart showing an example of the flow of the predicted-abnormality-sign output process of the predicted-abnormality-sign detection system according to the embodiment.

[0024] Fig.14 It is a diagram showing a hardware configuration example according to an embodiment of the present invention. DETAILED DESCRIPTION

[0025] Hereinafter, a detection device, a detection method, and a computer-readable recording medium according to an embodiment of the present invention will be described in detail with reference to the accompanying drawings. In addition, the present invention is not limited to the embodiment described below.

[0026] [Implementation Method]

[0027] Hereinafter, the structure and processing of the abnormality sign detection system 100 involved in the embodiment, the structure and processing of each device of the abnormality sign detection system 100, and the processing flow of the abnormality sign detection system 100 will be described in order, and finally the effects of the embodiment will be described.

[0028] [1. Configuration and processing of abnormality sign detection system 100]

[0029] use Figure 1 The configuration and processing of the abnormality sign detection system 100 according to the embodiment will be described in detail. Figure 1 1 is a diagram showing a structural example and a processing example of an abnormal sign detection system 100 involved in an embodiment. Below, the structural example of the entire abnormal sign detection system 100, the processing example of the entire abnormal sign detection system 100, and the problems of the abnormal sign detection system 100P involved in the reference technology are described in order, and finally the effect of the abnormal sign detection system 100 is described. In addition, in the embodiment, the remote monitoring of factory production using a workshop instrument as a device set in the workshop is described as an example, but the equipment and the field of use are not limited, and it can also be applied to remote monitoring of environmental measurements such as power monitors, wind power generation, water supply and drainage monitors, and monitoring rivers.

[0030] (1-1. Example of Overall Configuration of Abnormality Sign Detection System 100)

[0031] use Figure 1The overall configuration example of the abnormality sign detection system 100 is described. The abnormality sign detection system 100 includes a server device 10, a field instrument 20, and an operator terminal 30. Here, the server device 10, the field instrument 20, and the operator terminal 30 are connected to each other via a predetermined communication network (not shown) by wire or wireless means so as to be able to communicate. In addition, various communication networks such as the Internet and a dedicated line can be used for the predetermined communication network.

[0032] (1-1-1. Server Device 10)

[0033] The server device 10 is a detection device set in a cloud environment. Figure 1 The abnormality sign detection system 100 shown may include a plurality of server devices 10. In addition, the server device 10 may be implemented in an in-house environment, an edge environment, or the like.

[0034] (1-1-2. On-site instrument 20)

[0035] The field instrument 20 is a workshop instrument constituting a workshop. Figure 1 In the example of FIG. 2 , the field instrument 20 is implemented by a Coriolis flow meter.

[0036] (1-1-3. Operator terminal 30)

[0037] The operator terminal 30 is used by the operator W who performs maintenance and inspection of the workshop and field equipment 20, and accepts the operation of the operator W. The operator terminal 30 is implemented by, for example, a smart phone, a tablet terminal, a desktop PC (Personal Computer), a notebook PC, a smart watch, a mobile phone, a PDA (Personal Digital Assistant), etc. Figure 1 In the example of , the operator terminal 30 is implemented by a smartphone.

[0038] (1-2. Processing Example of the Entire Abnormality Sign Detection System 100)

[0039] use Figure 1 The following describes an example of the overall processing of the abnormality sign detection system 100. The processing of steps S1 to S11 described below may be performed in a different order. In addition, there may be omitted processing in the processing of steps S1 to S11 described below.

[0040] (1-2-1. Measurement data acquisition process)

[0041] The field device 20 acquires measurement data (step S1). For example, as measurement data, the field device 20 acquires sensor data such as phase difference data, frequency data, temperature data, drive current data, fluid pressure data, and density data.

[0042] (1-2-2. Measurement data transmission process)

[0043] The field device 20 transmits the measurement data to the server device 10 (step S2 ). For example, the field device 20 transmits sensor data such as phase difference data, frequency data, temperature data, drive current data, fluid pressure data, and density data to the server device 10 as the acquired measurement data.

[0044] (1-2-3. Abnormality sign detection and processing)

[0045] The server device 10 detects signs of abnormality in the workshop (appropriately referred to as "sense of disharmony") based on the measurement data (step S3). For example, the server device 10 detects signs of abnormality in the workshop using a teacher-less learning model that is a learned detection model DM that outputs signs of abnormality in the workshop based on the input of measurement data.

[0046] Here, the abnormal sign refers to a state that has the possibility of causing an abnormal state that has a negative impact on the workshop, such as equipment failure, abnormal value, etc. In addition, in addition to the above-mentioned abnormal signs, the server device 10 can also detect the period of the normal state, the sign of improvement that indicates the progress, that is, the sign of improvement in the intermediate area of ​​the jump from the abnormal state to the normal state, etc.

[0047] In addition, the server device 10 may detect a sign of abnormality using one detection model DM, or may detect a sign of abnormality using a plurality of detection models DM.

[0048] (1-2-4. Test result transmission process)

[0049] The server device 10 transmits the detection result to the operator terminal 30 (step S4 ). For example, as the detection result, the server device 10 transmits to the operator terminal 30 the abnormality sign of the workshop, the detection time, the identification information of the field device 20 related to the detection, and the like.

[0050] (1-2-5. Alarm notification processing)

[0051] The operator terminal 30 notifies the operator W of an alarm (step S5). For example, when receiving the detection result from the server device 10, the operator terminal 30 displays the abnormal sign of the workshop, the detection time, the identification information of the field instrument 20 associated with the detection, etc. on the monitor screen, and urges the operator W to confirm whether there is an abnormal sign of the workshop.

[0052] (1-2-6. Detection result determination process)

[0053] The operator W determines the authenticity of the detection result (step S6). For example, the operator W visually confirms the presence or absence of abnormal signs of the field instrument 20 notified as the detection result to determine the authenticity of the detection result. At this time, for example, when the field instrument 20 is a Coriolis flowmeter, if the measured fluid contains residue that cannot be seen in the normal state, the operator W determines that the abnormal sign is "true", and if no difference from the normal state is confirmed, the operator determines that the abnormal sign is "false".

[0054] (1-2-7. Judgment result input processing)

[0055] The worker W inputs the determination result to the worker terminal 30 (step S7 ). For example, the worker W inputs the authenticity of the detection result by operating an input screen displayed on the monitor screen of the worker terminal 30 .

[0056] (1-2-8. Judgment result transmission process)

[0057] The operator terminal 30 transmits the determination result to the server device 10 (step S8 ). For example, the operator terminal 30 transmits the authenticity of each detection result input by the operator W to the server device 10 as the determination result.

[0058] (1-2-9. Evaluation index calculation process)

[0059] The server device 10 calculates the evaluation index based on the determination result (step S9). For example, the server device 10 calculates the evaluation index for each detection model DM within a specified period, and sends the calculated evaluation index to the operator terminal 30. Here, the evaluation index refers to a numerical value indicating the accuracy of the machine learning model, such as the correct answer rate, the fit rate, the reproducibility rate, the specificity, the F value, etc.

[0060] Furthermore, the server device 10 generates a time series graph indicating output data, measurement data, and the like of abnormality signs within a predetermined period for each detection model DM, and transmits the generated time series graph to the operator terminal 30 .

[0061] (1-2-10. Model relearning process)

[0062] The server device 10 relearns the detection model DM (step S10). For example, the server device 10 relearns the detection model DM, which is determined to have an incorrect detection result, among the plurality of detection models DM using the measurement data, and updates various parameters. At this time, if the evaluation index of the detection model DM is less than a specified value, the server device 10 may relearn the detection model DM using the measurement data and update various parameters.

[0063] (1-2-11. Model determination process)

[0064] The server device 10 determines the detection model DM (step S11). For example, when the evaluation index of the plurality of detection models DM is greater than or equal to a predetermined value, the server device 10 determines the detection model DM as a model used in the subsequent abnormality prediction processing, such as the detection model DM described below. Fig.13 In this case, the server device 10 may determine as the adopted model one or more detection models DM whose evaluation index is greater than or equal to a predetermined value among the plurality of detection models DM. In addition, when there are a plurality of detection models DM determined as adopted models, the server device 10 may select one or more detection models DM as the adopted model. Fig.13 In the abnormal sign detection processing and detection result sending processing, all detection results as outputs of multiple detection models DM may be notified to the operator W, or only the detection results as outputs of the detection model DM having the largest evaluation index among the multiple detection models DM may be notified to the operator W.

[0065] Furthermore, the server device 10 may determine one or more detection models DM selected by the operator W based on the evaluation index and the time series graph among the plurality of detection models DM as the adopted model.

[0066] (1-3. Effects of abnormality sign detection system 100)

[0067] Next, the effects of the predicted abnormality sign detection system 100 will be described after describing the outline and problems of the predicted abnormality sign detection system 100P according to the reference technique.

[0068] (1-3-1. Overview of abnormality sign detection system 100P)

[0069] The following processing is performed in the abnormal sign detection system 100P involved in the reference technology. First, the abnormal sign detection system 100P performs unsupervised learning based on normal data collected in advance, extracts the characteristics of the normal data and stores them as a normal state model. Second, the abnormal sign detection system 100P dynamically detects a normal judgment reference value that can regard the input data as normal based on unknown input data and the normal state model. Third, the abnormal sign detection system 100P calculates a judgment value for abnormal judgment of the input data based on the input data and the normal state model. Fourth, the abnormal sign detection system 100P detects abnormal signs of the input data by detecting an abnormal state that deviates from the normal state based on the normal judgment reference value and the judgment value.

[0070] (1-3-2. Problems with abnormality sign detection system 100P)

[0071] The following problems exist with the abnormal sign detection system 100P involved in the reference technology. First, with the abnormal sign detection system 100P, the pre-collected data is not labeled, so the evaluation index of the machine learning model cannot be calculated. Second, with the abnormal sign detection system 100P, the machine learning model cannot be easily changed at the initial stage of introduction, so it takes a long time to study the machine learning model until it is introduced into the workshop.

[0072] (1-3-3. Overview of abnormality sign detection system 100)

[0073] Regarding the abnormal sign detection system 100 involved in the implementation method, the following processing is performed. First, the server device 10 collects the measurement data sent by the field instrument 20 and inputs the measurement data into the detection model DM, thereby detecting the abnormal sign of the workshop. At this time, the server device 10 prepares a plurality of unsupervised learning models obtained from the pre-learning data set, learns the time series data used for re-learning of the model based on the obtained learned model, and notifies the operator W of the output abnormal sign. Second, the server device 10 accepts the judgment result for the detection result input by the operator W. At this time, when the abnormal sign is notified, the operator W uses the application on the operator terminal 30 such as a smartphone or PC to input whether the above-mentioned abnormal sign, that is, the sense of disharmony, is correct (true abnormality: true positive, false abnormality: false positive). Here, "true positive" refers to the situation where the abnormal sign is detected by the server device 10 (there is a sense of disharmony) and the operator W confirms the abnormal sign (sense of disharmony: positive). In addition, a “false positive” refers to a situation where a sign of abnormality is detected by the server device 10 (there is a sense of discomfort), but the operator W cannot confirm the sign of abnormality (sense of discomfort: error).

[0074] (1-3-4. Effects of abnormality sign detection system 100)

[0075] The abnormal sign detection system 100 has the following effects. First, by calculating the evaluation index, the abnormal sign detection system 100 can compare which detection model DM is better among multiple detection models DM, so even if there is a change in the normal range or the operation process, it is possible to select an appropriate detection model DM that has been learned. Second, by using multiple detection models DM for abnormal sign detection at the same time and comparing them with the judgment results from the operator W on site, the detection model DM can be updated, so there is no need to conduct a detailed inspection before introduction to the workshop.

[0076] As described above, the abnormality sign detection system 100 can update the detection model DM obtained by unsupervised learning and construct the optimal detection model DM, thereby improving the accuracy of the detection model DM for detecting abnormality signs in the workshop. In addition, the abnormality sign detection system 100 can use the optimal model among multiple models.

[0077] Furthermore, the abnormal sign detection system 100 has the following effects. With regard to the abnormal sign detection system 100, the operator W can easily operate it intuitively, so the workshop user can easily introduce the abnormal sign detection system 100 into the workshop. In addition, in order to select a learning model, a lot of domain knowledge and empirical rules are required, and the supplier of the learning model requires such knowledge and empirical rules. With regard to the abnormal sign detection system 100, the on-site data of the workshop user is used as the basis for model selection, so the workshop user can hide data and domain knowledge from the supplier. In addition, for changes in the operating process and the introduction of different instruments, the learning model must also be changed, but in the abnormal sign detection system 100, the learning model can be rebuilt by performing the same operation, so the workshop user does not need to negotiate with the supplier again and rebuild the learning model. In addition, with regard to the abnormal sign detection system 100, it is possible to model the basis for experienced people to judge what is abnormal, so it can be flexibly used to eliminate the lack of successors and educate inexperienced operators W. Furthermore, regarding the abnormality-sign detection system 100 , the supplier does not need to provide consultation or provision of an abnormality-sign detection solution corresponding to an application or a plant user, and thus the solution can be made universal.

[0078] [2. Configuration and processing of each device of abnormality sign detection system 100]

[0079] use Figure 2 right Figure 1 The configuration and processing of each device included in the shown abnormality sign detection system 100 will be described. Figure 2 This is a block diagram showing a configuration example of each device of the abnormality sign detection system 100 according to the embodiment. Next, based on the description of the overall configuration example of the abnormality sign detection system 100 according to the embodiment, the configuration example and processing example of the server device 10, the configuration example and processing example of the field device 20, and the configuration example and processing example of the operator terminal 30 according to the embodiment are described in detail.

[0080] (2-1. Example of Overall Configuration of Abnormality Sign Detection System 100)

[0081] use Figure 2 right Figure 1 The overall structure of the abnormality sign detection system 100 shown in FIG. Figure 2 As shown, the abnormality sign detection system 100 includes a server device 10, a field instrument 20, and an operator terminal 30. The server device 10 is connected to the field instrument 20 and the operator terminal 30 in a communicable manner using a communication network N implemented by the Internet, a dedicated line, etc. In addition, the server device 10 is set in a cloud environment. In addition, the field instrument 20 is set in a workshop. In addition, the server device 10 can be set in an internal environment of a company, an edge environment, etc.

[0082] (2-2. Configuration Example and Processing Example of Server Device 10)

[0083] use Figure 2 The configuration example and processing example of the server device 10 are described. The server device 10 includes an input unit 11, an output unit 12, a communication unit 13, a storage unit 14, and a control unit 15. The configuration example and processing example of the server device 10 are described below, but the field device 20 can also perform the same processing. That is, the abnormality sign detection system 100 can be realized even with only the field device 20.

[0084] (2-2-1. Input unit 11)

[0085] The input unit 11 controls input of various information to the server device 10. For example, the input unit 11 is realized by a mouse, a keyboard, etc., and receives input of various information to the server device 10.

[0086] (2-2-2. Output unit 12)

[0087] The output unit 12 is responsible for outputting various information from the server device 10. For example, the output unit 12 is realized by a display or the like, and displays various information stored in the server device 10.

[0088] (2-2-3. Communications Department 13)

[0089] The communication unit 13 manages data communication with other devices. For example, the communication unit 13 performs data communication with a communication device via a router, etc. The communication unit 13 can also perform data communication with a terminal of an operator (not shown).

[0090] (2-2-4. Storage Unit 14)

[0091] The storage unit 14 stores various information that is referenced when the control unit 15 performs an action, and various information that is acquired when the control unit 15 performs an action. The storage unit 14 includes a detection result storage unit 14a, a judgment result storage unit 14b, an evaluation index storage unit 14c, and a detection model storage unit 14d. Here, the storage unit 14 can be implemented by, for example, a semiconductor memory element such as a RAM (Random Access Memory) or a flash memory, or a storage device such as a hard disk or an optical disk. In addition, Figure 2 In the example of FIG. 1 , the storage unit 14 is provided inside the server device 10 , but may be provided outside the server device 10 , or a plurality of storage units may be provided.

[0092] (2-2-4-1. Detection Result Storage Unit 14a)

[0093] The detection result storage unit 14a stores the detection results. For example, the detection result storage unit 14a stores the detection results detected by the detection unit 15b of the control unit 15 described later. Figure 3 An example of data stored in the detection result storage unit 14 a will be described. Figure 3 FIG. 1 is a diagram showing an example of the detection result storage unit 14a of the server device 10 according to the embodiment. Figure 3 In the example of FIG. 1 , the detection result storage unit 14 a has items such as “workshop”, “detection result”, “detection model”, “detection time”, and “on-site instrument”.

[0094] "Workshop" indicates identification information for identifying the workshop of the monitored object, such as the identification number and identification mark of the workshop. "Detection result" indicates identification information for identifying the detection result, such as the identification number and identification mark of the abnormality sign detected by the detection unit 15b. "Detection model" indicates identification information for identifying the machine learning model, such as the identification number and identification mark of the detection model DM that outputs the abnormality sign. "Detection time" indicates the time when the detection result is detected, such as expressed by year, month, day, hour, minute and second. "Field instrument" indicates identification information for identifying the field instrument 20 that obtains the measurement data, such as the identification number and identification mark such as the instrument name and model of the field instrument 20.

[0095] That is, in Figure 3The following example is shown, that is, for the workshop identified by "Workshop PS001", "Detection result #1" is {detection model: "Detection model DM001", detection time: "10:10:00", field instrument: "FD001"}, "Detection result #2" is {detection model: "Detection model DM002", detection time: "10:10:00", field instrument: "FD001"}, "Detection result #3" is {detection model: "Detection model DM003", detection time: "14:30:00", field instrument: "FD001"}, and... data are stored in the detection result storage unit 14a.

[0096] (2-2-4-2. Determination Result Storage Unit 14b)

[0097] The determination result storage unit 14b stores the determination result. For example, the determination result storage unit 14b stores the determination result input by the operator W and sent from the operator terminal 30. Figure 4 An example of data stored in the determination result storage unit 14 b will be described. Figure 4 FIG. 1 is a diagram showing an example of the determination result storage unit 14b of the server device 10 according to the embodiment. Figure 4 In the example of FIG. 1 , the determination result storage unit 14 b has items such as “workshop”, “detection result”, and “determination result”.

[0098] "Workshop" indicates identification information for identifying the workshop to be monitored, such as the identification number or identification mark of the workshop. "Detection result" indicates identification information for identifying the detection result, such as the identification number or identification mark of the abnormal sign detected by the detection unit 15b. "Judgment result" indicates the judgment result corresponding to the detection result, such as the judgment of the authenticity of the abnormal sign confirmed by the operator W.

[0099] That is, in Figure 4 The following example is shown, that is, for the workshop identified by "Workshop PS001", the judgment of the authenticity of the abnormal sign corresponding to "Detection result #1" is "Judgment result #1", the judgment of the authenticity of the abnormal sign corresponding to "Detection result #2" is "Judgment result #2", the judgment of the authenticity of the abnormal sign corresponding to "Detection result #3" is "Judgment result #3", and so on are stored in the judgment result storage unit 14b.

[0100] (2-2-4-3. Evaluation Index Storage Unit 14c)

[0101] The evaluation index storage unit 14c stores the evaluation index. For example, the evaluation index storage unit 14c stores the evaluation index of each detection model DM calculated by the execution unit 15c of the control unit 15 described later. Figure 5 An example of data stored in the evaluation index storage unit 14 c will be described. Figure 5 FIG. 1 is a diagram showing an example of the evaluation index storage unit 14c of the server device 10 according to the embodiment. Figure 5 In the example of , the evaluation index storage unit 14c has items such as "detection model", "evaluation index", and "calculation time".

[0102] "Detection model" indicates identification information for identifying a learning model, such as an identification number or identification mark of a detection model DM that outputs an abnormality sign. "Evaluation index" indicates an evaluation index that is a numerical value indicating the accuracy of a learning model, such as the correct answer rate, fit rate, reproducibility, specificity, F value, etc. of the detection model DM. "Calculation time" indicates the time when the evaluation index is calculated, such as expressed by year, month, day, hour, minute, and second.

[0103] That is, in Figure 5 The following example is shown, that is, the detection model DM identified by "detection model DM001" is {evaluation index 1 (correct answer rate): "80%", evaluation index 2 (suitability rate): "75%", calculation time: "10:30:00", ...}, the detection model DM identified by "detection model DM002" is {evaluation index 1 (correct answer rate): "75%", evaluation index 2 (suitability rate): "70%", calculation time: "10:30:00", ...}, the detection model DM identified by "detection model DM003" is {evaluation index 1 (correct answer rate): "90%", evaluation index 2 (suitability rate): "85%", calculation time: "15:00:00", ...}, ... data are stored in the evaluation index storage unit 14c.

[0104] (2-2-4-4. Detection Model Storage Unit 14d)

[0105] The detection model storage unit 14d stores the detection model DM. For example, the detection model storage unit 14d stores the detection model DM used by the detection unit 15b of the control unit 15 described later. Figure 6 An example of data stored in the detection model storage unit 14 d will be described. Figure 6 FIG. 1 is a diagram showing an example of the detection model storage unit 14d of the server device 10 according to the embodiment. Figure 6 In the example of , the detection model storage unit 14d has items such as "detection model".

[0106] The “detection model” is model data of a machine learning model, for example, data including model parameters, hyperparameters, etc., which are execution data of an algorithm for executing the detection model DM.

[0107] That is, in Figure 6 , an example is shown in which data of “detection model DM001 ”, “detection model DM002 ”, “detection model DM003 ”, . . . are stored in the detection model storage unit 14 d as model data of the detection model DM used by the detection unit 15 b .

[0108] Here, using Figure 7 A specific example of a detection model DM for binary classification using measurement data collected from a Coriolis flowmeter as input data and abnormality sign data as output data will be described. Figure 7 In the description, an example of a detection model DM using a neural network is used, but it may also be an isolation forest, a One-class SVM (Support-Vector Machine), etc., and is not particularly limited.

[0109] exist Figure 7 In the example, the detection model DM accepts input of sensor data such as phase difference data, frequency data, temperature data, drive current data, fluid pressure data, density data, etc. at the input layer, performs analysis of the input sensor data at the middle layer, and outputs abnormal sign data such as "abnormal sign" and "no abnormal sign" at the output layer.

[0110] The detection model DM learns the normal time data through Hotelling's T2 theory, local deviation factor method, neural network system such as autoencoder, self-organizing map, one-class support vector machine, classification tree system such as isolation forest, change point detection, k-nearest neighbor method, k-means method, and unsupervised learning of density compatibility clustering, and constructs a learning model that outputs the result of classifying different actions as abnormal signs, that is, disharmony. In addition, the detection model DM uses any one or more of the sensor data obtained inside the field instrument 20 at the input layer.

[0111] (2-2-5. Control unit 15)

[0112] Back to Figure 2 The control unit 15 is in charge of the overall control of the server device 10. The control unit 15 includes a collection unit 15a, a detection unit 15b, and an execution unit 15c. Here, the control unit 15 can be implemented by electronic circuits such as CPU (Central Processing Unit), MPU (Micro Processing Unit), ASIC (Application Specific Integrated Circuit), FPGA (Field Programmable Gate Array) and the like.

[0113] (2-2-5-1. Collecting Unit 15a)

[0114] The collecting unit 15a collects various information. In addition, the collecting unit 15a can store the collected various information in the storage unit 14. For example, the collecting unit 15a collects measurement data measured by a measurement device.

[0115] To illustrate a specific example, as the measurement data sent by the Coriolis flowmeter, i.e., the field instrument 20, the collection unit 15a collects the measurement data in a time series, such as "measurement data 001" {phase difference data: "phase difference data 001", frequency data: "frequency data 001", temperature data: "temperature data 001", drive current data: "drive current data 001", fluid pressure data: "fluid pressure data 001", density data: "density data 001"}, "measurement data 002" {phase difference data: "phase difference data 002", frequency data: "frequency data 002", temperature data: "temperature data 002", drive current data: "drive current data 002", fluid pressure data: "fluid pressure data 002", density data: "density data 002"}, and so on.

[0116] (2-2-5-2. Detection Unit 15b)

[0117] The detection unit 15b performs processing related to the detection result. In addition, the detection unit 15b stores various information in the storage unit 14. In addition, the detection unit 15b obtains various information stored in the storage unit 14. In addition, the detection unit 15b obtains various information collected by the collection unit 15a. Next, the phenomenon prediction detection processing and the detection result transmission processing are described.

[0118] (Phenomenon prediction detection processing)

[0119] The detection unit 15b obtains the output result of the detection model DM and detects the prediction of the phenomenon. In addition, the detection unit 15b stores the detection result in the detection result storage unit 14a. For example, the detection unit 15b inputs the collected measurement data to the detection model DM, which is a learning model that predicts a specified phenomenon based on the input of the measurement data, and obtains the detection result as the output of the learned detection model DM, and detects the detection result as the prediction of the phenomenon. In addition, the detection unit 15b obtains the detection results of the outputs of multiple learned detection models DM that predict the specified phenomenon, and detects the detection results as the prediction of the phenomenon.

[0120] Here, the learned detection model DM is a learning model that has been learned by unsupervised learning using each measurement data as a normal value among a plurality of measurement data collected in a predetermined period as training data.

[0121] In addition, the prediction of a phenomenon refers to an "abnormal prediction" which is a sign of an abnormal state of the monitored object, a "normal prediction" which is a sign of a normal state of the monitored object, an "improvement prediction" which is a sign of improvement of the monitored object, that is, an intermediate area of ​​transition from an abnormal state to a normal state, etc.

[0122] As an "abnormality prediction", the detection unit 15b detects a sign of an abnormal state such as "there is a sign of abnormality" corresponding to a sign of a failure or abnormal value of the field device 20 based on the measurement data obtained by the field device 20 in the workshop. At this time, the detection unit 15b can detect a sign of no abnormal state such as "there is no sign of abnormality", that is, a sign of a normal state described later.

[0123] Furthermore, as a “normal prediction”, the detection unit 15 b may detect a sign of a normal state, such as a “normal sign” of a transition to a normal operation of the field device 20 in the future, based on the measurement data acquired by the field device 20 in the workshop.

[0124] In addition, as an "improvement prediction", the detection unit 15b can also detect a sign of improvement such as "there is a sign of improvement" of the current abnormality of the field instrument 20 or a transition from an abnormality sign state to an improvement of the abnormal value in the future based on the measurement data obtained by the field instrument 20 in the workshop. At this time, the detection unit 15b can detect a situation where there is no sign of improvement such as "there is no sign of improvement".

[0125] To illustrate a specific example, the detection unit 15b inputs the measurement data collected from the field instrument 20 which is a Coriolis flowmeter, namely, "measurement data 001" {phase difference data: "phase difference data 001", frequency data: "frequency data 001", temperature data: "temperature data 001", drive current data: "drive current data 001", fluid pressure data: "fluid pressure data 001", density data: "density data 001"} as input data for "detection model DM001", "detection model DM002" and "detection model DM003", respectively, and obtains {detection model DM001: "There is a sign of abnormality", detection model DM002: "There is a sign of abnormality", detection model DM003: "No sign of abnormality"} as output data. In addition, the detection unit 15b inputs the measurement data collected from the field instrument 20 which is a Coriolis flowmeter, namely, "measurement data 002" {phase difference data: "phase difference data 002", frequency data: "frequency data 002", temperature data: "temperature data 002", drive current data: "drive current data 002", fluid pressure data: "fluid pressure data 002", density data: "density data 002"} as input data for "detection model DM001", "detection model DM002", and "detection model DM003", respectively, and obtains {detection model DM001: "No signs of abnormality", detection model DM002: "No signs of abnormality", detection model DM003: "There are signs of abnormality"} as output data.

[0126] (Test result sending and processing)

[0127] When the output result of the detection model DM is acquired, the detection unit 15b transmits the acquired output result as a detection result to the terminal used by the user, and notifies the user of the detection result.

[0128] When the output result of the learned detection model DM is equivalent to a sign of an abnormal state, the detection unit 15b notifies the user. For example, when an abnormal prediction is detected based on the measurement data acquired by the field instrument 20 in the workshop, the detection unit 15b sends the detection result related to the abnormal prediction to the operator terminal 30 and notifies the operator W in the workshop by displaying the detection result such as "there is a sign of abnormality".

[0129] In addition, when the output result of the learned detection model DM is equivalent to a sign of a normal state, the detection unit 15b notifies the user. For example, when a normal prediction is detected based on the measurement data acquired by the field instrument 20 in the workshop, the detection unit 15b sends the detection result related to the normal prediction to the operator terminal 30, and notifies the operator W in the workshop by displaying the detection result such as "normal sign".

[0130] In addition, when the output result of the learned detection model DM corresponds to a sign of improvement, the detection unit 15b notifies the user. For example, when an improvement prediction is detected based on the measurement data acquired by the field instrument 20 in the workshop, the detection unit 15b sends the detection result related to the improvement prediction to the operator terminal 30, and notifies the operator W in the workshop by displaying the detection result such as "there is a sign of improvement".

[0131] To illustrate a specific example, the detection unit 15b sends {detection result: "Detection result #1", detection time: "10:10:00", on-site instrument: "FD001"} as the detection result of "Detection model DM001", {detection result: "Detection result #2", detection time: "10:10:00", on-site instrument: "FD001"} as the detection result of "Detection model DM002", and {detection result: "Detection result #3", detection time: "14:30:00", on-site instrument: "FD001"} as the detection result of "Detection model DM003" to the operator terminal 30 used by the operator W in the workshop, so that the input and output unit 31 which is the monitor screen of the operator terminal 30 displays them.

[0132] (2-2-5-3. Execution unit 15c)

[0133] The execution unit 15c executes processing related to the adoption or relearning of the detection model DM. In addition, the execution unit 15c obtains various information stored in the storage unit 14. In addition, the execution unit 15c obtains various information collected by the collection unit 15a. In addition, the execution unit 15c obtains various information output by the detection unit 15b. The relearning execution processing, score calculation processing, time series graph generation processing, and detection model determination are described below.

[0134] (Relearning execution process)

[0135] In the case where the detection result of the detection model DM as the learned model is different from the judgment result obtained by the user of the device that has confirmed the prediction of the specified phenomenon, that is, the operator W, the execution unit 15c performs re-learning of the learned detection model DM using the label value input by the operator W and the collected measurement data to generate a re-learned model. In addition, the execution unit 15c performs re-learning of the learned detection model DM by assigning the label value based on the judgment result as teacher data to the collected measurement data through supervised (supervised) learning. In addition, in the case where the detection result as the output of the learned learning model is equivalent to a sign of an abnormal state, and the operator W has not confirmed the abnormal state or abnormal sign of the measuring device, the execution unit 15c performs re-learning of the learned detection model DM using the teacher data obtained by assigning the presence or absence of the abnormal state or abnormal sign state based on the judgment result as a label value to the collected measurement data.

[0136] To illustrate a specific example, the execution unit 15c inputs the "measurement data 001" collected from the field instrument 20 which is a Coriolis flowmeter as input data to the "detection model DM002", and obtains {detection model DM002: "there is a sign of abnormality"} as output data. When the output data is a false positive different from the judgment result input by the operator W, the teacher data obtained by assigning "no sign of abnormality" as a label value to the "measurement data 001" used in the above processing is used as training data, and re-learning of the "detection model DM002" is performed to update the model parameters and hyperparameters.

[0137] In addition, the execution unit 15c executes re-learning of the learned detection model DM based on the evaluation index. For example, the execution unit 15c executes re-learning of the learned detection model DM whose evaluation index is smaller than a predetermined value.

[0138] To illustrate a specific example, the execution unit 15c calculates the suitability rate as an evaluation index of the "detection model DM002" through the evaluation index calculation process described later. When the suitability rate is less than "75%", the "measurement data 001" used in the above process is used as training data to perform re-learning of the "detection model DM002" and update the model parameters and hyperparameters.

[0139] (Evaluation index calculation process)

[0140] The execution unit 15c calculates the evaluation index of the learned detection model DM based on the output result of the learned detection model DM and the judgment result of the operator W who confirmed that the measuring device predicted the specified phenomenon. In addition, the execution unit 15c calculates at least one of the correct answer rate, the fit rate, the reproducibility rate, the specificity, and the F value as an evaluation index based on the detection results of each of the multiple learned detection models DM and the judgment result of the operator W who confirmed that the device predicted the specified phenomenon. In addition, the execution unit 15c causes the operator terminal 30 used by the operator W to display the calculated evaluation index.

[0141] The execution unit 15c calculates the correct answer rate, the fit rate, the reproducibility rate, the specificity, the F value, etc. within the prescribed period as evaluation indicators. For example, the execution unit 15c calculates the fit rate as an evaluation indicator of the detection model DM that only outputs "there is a sign of abnormality" (positive), calculates the correct answer rate, the reproducibility rate, and the F value as evaluation indicators of the detection model DM that outputs "there is a sign of abnormality" (positive) and "there is no sign of abnormality" (negative), and calculates the specificity as an evaluation indicator of the detection model DM that only outputs "there is no sign of abnormality" (negative).

[0142] In the following, the correct answer rate, the fit rate, the reproducibility rate, the specificity, and the F value are described in order with respect to the evaluation indexes calculated by the execution unit 15c. In addition, in the description of each evaluation index below, the predicted existence of a prescribed phenomenon such as an abnormal sign detected by the detection unit 15b is set as "positive", the predicted non-existence of the prescribed phenomenon detected by the detection unit 15b is set as "negative", the predicted existence of the prescribed phenomenon determined to be correct by the confirmation of the operator W is set as "true positive", the predicted existence of the prescribed phenomenon determined to be wrong by the confirmation of the operator W is set as "false positive", the predicted non-existence of the prescribed phenomenon determined to be correct by the confirmation of the operator W is set as "true negative", and the predicted non-existence of the prescribed phenomenon determined to be wrong by the confirmation of the operator W is set as "false negative".

[0143] (correct answer rate)

[0144] Regarding the execution unit 15c, when the output results of the "detection model DM003" of "there are signs of abnormality" and "there are no signs of abnormality" are 85 times true positive, 15 times false positive, 75 times true negative, and 25 times false negative out of 200 times, the correct answer rate = (true positive + true negative) / (true positive + false positive + true negative + false negative) = 85 + 75 / 200 = 0.90, so "90%" is calculated as an evaluation indicator of the "detection model DM003".

[0145] (Suitability rate)

[0146] Regarding the execution unit 15c, when the output result of the "detection model DM003" "there are signs of abnormality" is true positive 85 times out of 100 times and false positive 15 times, the suitability rate = (true positive) / (true positive + false positive) = 85 / 100 = 0.85, so "85%" is calculated as the evaluation index of the "detection model DM003".

[0147] (Reproduction rate)

[0148] Regarding the execution unit 15c, when the output results of the "detection model DM002" of "with abnormal signs" and "without abnormal signs" are true positives 70 times out of 200 times, false positives 30 times, true negatives 80 times, and false negatives 20 times, the reproduction rate = (true positive) / (true positive+false negative) = 70 / 70+20 = 0.777, so "78%" is calculated as an evaluation indicator of the "detection model DM002".

[0149] (Specificity)

[0150] Regarding the execution unit 15c, when the output result of the "detection model DM001" of "no abnormal sign" is true negative 85 times out of 100 times and false negative 15 times, the specificity = (true negative) / (true negative + false negative) = 85 / 100 = 0.85, so "85%" is calculated as the evaluation indicator of the "detection model DM001".

[0151] (F value)

[0152] Regarding the execution unit 15c, when the output results of the "detection model DM002" of "there are signs of abnormality" and "there are no signs of abnormality" are true positives 70 times out of 200 times, false positives 30 times, true negatives 80 times, and false negatives 20 times, the F value (β=1) = (true positive) / {true positive+(false positive+false negative) / 2}=70 / {70+(30+20) / 2}=0.736, so "74%" is calculated as an evaluation indicator of the "detection model DM002".

[0153] (Time series graph generation process)

[0154] The execution unit 15c generates a time series graph showing the output results of each of the plurality of learned detection models DM and the history of the collected measurement data. In addition, the execution unit 15c causes the operator terminal 30 used by the operator W to display the evaluation index and the time series graph.

[0155] To illustrate a specific example, as a time series curve graph of "detection model DM001", the execution unit 15c displays "measurement data 001", "measurement data 002", "measurement data 003", ... collected from the field instrument 20 which is a Coriolis flowmeter in the form of a line graph, and generates a time series curve graph in which the detection time when "detection model DM001" is output as "there is a sign of abnormality" is colored and displayed, and the generated time series curve graph data is sent to the operator terminal 30, so that the input and output unit 31 which is the monitor screen of the operator terminal 30 displays the time series curve graph.

[0156] The execution unit 15c generates a screen (normal state screen) indicating a normal state as a detection result. The execution unit 15c causes the operator terminal 30 used by the operator W to display the generated screen so that the operator W can switch between display and non-display.

[0157] To explain a specific example, as the normal state screen of "detection model DM001", the execution unit 15c generates a time series graph indicating the current normal operation or future normal operation of the field instrument 20 as a Coriolis flowmeter, and sends the generated time series graph data to the operator terminal 30, so that the input and output unit 31 as the monitor screen of the operator terminal 30 displays the time series graph. At this time, the execution unit 15c can generate the time series graph data including buttons such as "normal state notification ON" and "normal state notification OFF", so that the operator terminal 30 can display the normal state screen in a manner that can be switched between display and non-display by the operation of the operator W.

[0158] (Model decision processing)

[0159] The execution unit 15c determines to adopt a learned detection model DM whose evaluation index is greater than or equal to a predetermined value among the plurality of learned detection models DM. Also, the execution unit 15c determines to adopt a learned detection model DM selected by the operator W among the plurality of learned detection models DM.

[0160] To illustrate a specific example, with respect to the execution unit 15c, when the evaluation index of the "detection model DM001" is "80%", the evaluation index of the detection model DM002 is "75%", and the evaluation index of the detection model DM003 is "85%", and when the reference value of the adopted evaluation index is set to "80%", the "detection model DM001" and the "detection model DM003" are determined as adopted models and used in the processing of the subsequent detection unit 15b.

[0161] In addition, the execution unit 15c causes the operator terminal 30 to display the evaluation indicators of "detection model DM001", "detection model DM002" and "detection model DM003", and the time series curve graphs of "detection model DM001", "detection model DM002" and "detection model DM003". When the operator W selects "detection model DM001" on the display screen, "detection model DM003" is determined as the adopted model and is used in the subsequent processing of the detection unit 15b.

[0162] (2-3. Configuration Example and Processing Example of Field Device 20)

[0163] use Figure 2 The following describes a configuration example and a processing example of the field device 20 . The field device 20 is a sensor device such as a thermometer, a pressure gauge, or a flow meter, and includes an acquisition unit 21 and a communication unit 22 .

[0164] (2-3-1. Acquisition Unit 21)

[0165] The acquisition unit 21 acquires measurement data. For example, as measurement data, the acquisition unit 21 acquires sensor data such as phase difference data, frequency data, temperature data, drive current data, fluid pressure data, and density data. Here, the measurement data includes sensor data acquired from each sensor, or calculation data calculated secondary based on the sensor data. For example, when the field instrument 20 is a Coriolis flowmeter, the acquisition unit 21 performs calculation processing on mass flow data, density data, etc. and outputs them. In the calculation processing process, temperature data, phase difference data, frequency data, drive current data, fluid pressure data, etc. are measured or calculated.

[0166] (2-3-2. Communications Department 22)

[0167] The communication unit 22 is responsible for data communication with other devices. For example, the communication unit 22 performs data communication with each communication device via a router, etc. In addition, the communication unit 22 can perform data communication with a terminal of an operator (not shown).

[0168] (2-4. Configuration Example and Processing Example of Worker Terminal 30)

[0169] use Figure 2 The following describes a configuration example and a processing example of the operator terminal 30. The operator terminal 30 is a terminal used by an operator W who performs maintenance and inspection of the plant and field devices 20, and includes an input / output unit 31, a transceiver unit 32, and a communication unit 33.

[0170] (2-4-1. Input / output unit 31)

[0171] The input / output unit 31 controls the input of various information to the operator terminal 30. For example, the input / output unit 31 is realized by a mouse, a keyboard, a touch panel, etc., and receives input of setting information, etc. to the operator terminal 30. In addition, the input / output unit 31 displays various information from the operator terminal 30. For example, the input / output unit 31 is realized by a display, etc., and displays setting information, etc. stored in the operator terminal 30.

[0172] The input / output unit 31 displays a confirmation result input screen for the operator W to input the determination result. In addition, details of the confirmation result input screen are described in (2-4-4. Specific example 1 of the display screen of the operator terminal 30) later. In addition, the input / output unit 31 displays a detection model determination screen that enables the operator W to select the detection model DM. In addition, details of the detection model determination screen are described in (2-4-5. Specific example 2 of the display screen of the operator terminal 30) and (2-4-6. Specific example 3 of the display screen of the operator terminal 30) later.

[0173] (2-4-2. Transceiver 32)

[0174] The transceiver 32 transmits various information. For example, the transceiver 32 transmits the determination result input by the operator W in the confirmation result input screen to the server device 10. In addition, the transceiver 32 transmits the identification information of the detection model DM selected by the operator W in the detection model determination screen to the server device 10.

[0175] The transceiver 32 receives various information. For example, the transceiver 32 receives detection results, evaluation indicators, time series graphs, etc. notified to the operator W from the server device 10.

[0176] (2-4-3. Communications Department 33)

[0177] The communication unit 33 is responsible for data communication with other devices. For example, the communication unit 33 performs data communication with each communication device via a router, etc. In addition, the communication unit 33 can perform data communication with a terminal of an operator (not shown).

[0178] (2-4-4. Specific Example 1 of Display Screen of Operator Terminal 30)

[0179] Here, using Figure 8 A specific example 1 of the display screen output by the input / output unit 31 of the operator terminal 30 will be described. Figure 8 Specific example 1 of the display screen of the operator terminal 30 according to the embodiment is shown in FIG. Next, the “detection result determination screen” displayed in the “uncomfortable feeling detection system confirmation result input screen” will be described.

[0180] (2-4-4-1. Detection result judgment screen)

[0181] like Figure 8 As shown in the example of FIG. 1 , the operator terminal 30 displays “detection result”, “detection time”, “on-site instrument”, and “judgment” for each workshop. Figure 8 In the example, regarding the detection result "detection result #1", the detection time of the sense of disharmony as an abnormality precursor is "10:10:00", the field instrument 20 that sent the measurement data is "FD001", and the operator terminal 30 displays a selection box for inputting the truth or falsehood of "detection result #1". At this time, when the result confirmed on the workshop site for "detection result #1" is "true" which is the same as the detection result, the operator W enters the judgment result by clicking the selection box of "true" in "judgment". Similarly, when the result confirmed on the workshop site for "detection result #2" is "false" which is different from the detection result, the operator W enters the judgment result by clicking the selection box of "false" in "judgment", and when the result confirmed on the workshop site for "detection result #3" is "true", the operator enters the judgment result by clicking the selection box of "true" in "judgment". Moreover, when all the judgment results are input, the operator W can send the judgment result to the server device 10 by clicking the "Send" button.

[0182] In addition, when it is impossible to determine whether the result confirmed on the shop floor is "true" or "false", the operator W can input the determination result by clicking the "unclear" selection box of "determination". At this time, the detection result selected as "unclear" is excluded from the teacher data assigned with the label value and is also excluded from the calculation object of the evaluation index.

[0183] (2-4-4-2. Display format)

[0184] like Figure 8 As shown in the example of , the operator terminal 30 can display the detection results in the order of the detection time of the abnormality sign. Figure 8 In the example of , the operator terminal 30 selects "display in detection order" as a pull-down menu, so the detection results are displayed in the order of the detection time of the abnormality sign. At this time, the operator terminal 30 can display the detection results for each detection model DM and each field device 20.

[0185] (2-4-5. Specific Example 2 of Display Screen of Operator Terminal 30)

[0186] Here, using Fig. 9A specific example 2 of the display screen output by the input / output unit 31 of the operator terminal 30 will be described. Fig. 9 Specific example 2 of the display screen of the operator terminal 30 according to the embodiment is shown in FIG. Next, the “evaluation index display screen” displayed in the “discomfort detection system detection model determination screen” will be described.

[0187] (2-4-5-1. Evaluation index display screen)

[0188] like Fig. 9 As shown in the example of , the operator terminal 30 displays "adoption", "detection model", "evaluation index 1 (correct answer rate)", "evaluation index 2 (suitability rate)", etc. for each workshop. Figure 8 In the example, the operator terminal 30 displays a list of the following evaluation indicators, that is, regarding the "detection model DM003", the "evaluation indicator 1 (correct answer rate)" is "90%", and the "evaluation indicator 2 (suitability rate)" is "85%", regarding the "detection model DM001", the "evaluation indicator 1 (correct answer rate)" is "80%", and the "evaluation indicator 2 (suitability rate)" is "75%", and regarding the "detection model DM002", the "evaluation indicator 1 (correct answer rate)" is "75%", and the "evaluation indicator 2 (suitability rate)" is "70%". At this time, when the "detection model DM003" and the "detection model DM001" whose "evaluation indicator 2 (suitability rate)" is greater than or equal to "75%" are selected as the adopted models, the operator W clicks the "adopt" selection box of the "detection model DM003" and the "adopt" selection box of the "detection model DM001". When deciding on the adoption model, the worker W can send the selection result of the adoption model to the server device 10 by clicking the “determine” button.

[0189] (2-4-5-2. Display format)

[0190] like Fig. 9 As shown in the example of , the operator terminal 30 can display the detection model DM in descending order of the evaluation index. Fig. 9 In the example of , the operator terminal 30 selects "display in order of high score" as a pull-down menu, and thus the evaluation indicators of the detection model DM are displayed in descending order of the values ​​of "evaluation indicator 1 (correct answer rate)" or "evaluation indicator 2 (fit rate)". At this time, the operator terminal 30 can display the detection results for each category of the algorithm of the detection model DM.

[0191] (2-4-6. Specific Example 3 of Display Screen of Operator Terminal 30)

[0192] Here, using Fig.10A specific example 3 of the display screen output by the input / output unit 31 of the operator terminal 30 will be described. Fig.10 Specific example 3 of the display screen of the operator terminal 30 according to the embodiment is shown in FIG. Next, the “time series data display screen” displayed in the “discomfort detection system detection model determination screen” will be described.

[0193] (2-4-6-1. Detection result judgment screen)

[0194] like Fig.10 As shown in the example of FIG. 1 , the operator terminal 30 displays a time series graph of “detection model DM001”, “detection model DM002”, etc. for each workshop. Fig.10 In the example, the operator terminal 30 displays a line graph representing the numerical values ​​of the measurement data "data 1" and "data 2" of the field device 20, together with a graph representing the detection of "disharmony 1", "disharmony 2", "disharmony 3", and "disharmony 4" as abnormal signs for the "detection model DM001". In addition, the operator terminal 30 displays a line graph representing the numerical values ​​of the measurement data "data 1" and "data 2" of the field device 20, together with a graph representing the detection of "disharmony 1" and "disharmony 2" as abnormal signs for the "detection model DM002". At this time, when the "detection model DM001" is selected as the adopted model, the operator W clicks the "adopt" selection box of the "detection model DM001". Moreover, when deciding to adopt the model, the operator W clicks the "determine" button to send the selection result of the adopted model to the server device 10.

[0195] (2-4-6-2. Display format)

[0196] like Fig.10 As shown in the example of , the operator terminal 30 can select and display the data required to determine the adopted model. Fig.10 In the example of , the operator terminal 30 selects "Display detection results and measurement data" as a pull-down menu to display "disharmony" as the detection result and "data" as the measurement data. At this time, the operator terminal 30 can display only the detection result or only the measurement data. In addition, the operator terminal 30 can superimpose the evaluation index of the detection model DM on the above-mentioned time series graph.

[0197] (2-4-6-3. Input of judgment results)

[0198] The operator W can also input the authenticity of the detection result by operating on the time series graph. For example, the operator W can input that the detection result is "true" by double-clicking the diamond in the graphic representing the "sense of disharmony" as the detection result.

[0199] [3. Processing Flow of Abnormality Sign Detection System 100]

[0200] use Figure 11 to Figure 13 The processing flow of the predicted abnormality sign detection system 100 according to the embodiment will be described. Next, after describing the overall processing flow of the predicted abnormality sign detection system 100, the detection model construction process and the predicted abnormality sign output process will be described.

[0201] (3-1. Overall Processing of Abnormality Sign Detection System 100)

[0202] use Fig.11 The overall processing flow of the abnormality sign detection system 100 according to the embodiment will be described. Fig.11 This is a flowchart showing an example of the overall flow of the abnormality sign detection system 100 according to the embodiment. In addition, the processing of steps S101 to S102 described below may be performed in a different order. In addition, there may be omitted processing in the processing of steps S101 to S102 described below.

[0203] (3-1-1. Detection model construction process)

[0204] First, the server device 10 executes a detection model construction process (step S101 ). For example, the server device 10 repeatedly executes the processes of steps S201 to S209 described later to obtain a highly accurate detection model DM.

[0205] (3-1-2. Abnormal sign output processing)

[0206] Second, the server device 10 performs abnormality sign output processing (step S102). For example, the server device 10 performs the processing of steps S301 to S303 described later using the highly accurate detection model DM constructed in the processing of step S101, thereby realizing highly accurate abnormality sign detection.

[0207] (3-2. Detection model construction process)

[0208] use Fig.12 The flow of the detection model construction process of the predicted abnormality sign detection system 100 according to the embodiment will be described. Fig.12This is a flowchart showing an example of the flow of the detection model building process of the abnormality sign detection system 100 according to the embodiment. In addition, the processes of steps S201 to S209 described below may be performed in a different order. In addition, there may be omitted processes in the processes of steps S201 to S209 described below.

[0209] (3-2-1. Measurement data collection process)

[0210] First, the server device 10 executes a measurement data collection process (step S201 ). For example, the server device 10 collects measurement data acquired by the field device 20 .

[0211] (3-2-2. Abnormality sign detection and processing)

[0212] Second, the server device 10 executes abnormality sign detection processing (step S202 ). For example, the server device 10 detects abnormality signs by inputting the collected measurement data into the detection model DM.

[0213] (3-2-3. Test result transmission process)

[0214] Thirdly, the server device 10 executes a detection result transmission process (step S203 ). For example, the server device 10 transmits the detection result to the worker terminal 30 used by the worker W and displays it on the monitor screen.

[0215] (3-2-4. Detection result determination process)

[0216] Fourth, the server device 10 performs a detection result determination process (step S204). For example, the server device 10 receives a true or false determination of the abnormality sign confirmed by the operator W via the operator terminal 30. If the detection result is true (step S204: Yes), the server device 10 proceeds to the process of step S205. On the other hand, if the detection result is false (step S204: No), the server device 10 proceeds to the process of step S206.

[0217] (3-2-5. Evaluation index calculation process)

[0218] Fifth, the server device 10 executes evaluation index calculation processing (steps S205 and S206). For example, the server device 10 calculates the suitability rate of the detection model DM as an evaluation index.

[0219] (3-2-6. Relearning execution process)

[0220] Sixthly, the server device 10 executes a relearning execution process (step S207), and returns to the process of step S202. For example, the server device 10 executes relearning of the detection model DM whose detection result is determined to be false, and updates the parameters.

[0221] (3-2-7. Continuous learning determination process)

[0222] Seventh, the operator W performs a continuous learning determination process (step S208). For example, when the evaluation index is greater than or equal to the specified value, or when it is determined that there is no obstacle in the working state, the operator W determines to end the learning. In the case of continuous learning (step S208: Yes), the server device 10 returns to the process of step S202. On the other hand, in the case of not continuing learning (step S208: No), the server device 10 enters the process of step S209.

[0223] (3-2-8. Detection model determination process)

[0224] Eighth, the server device 10 performs the detection model determination process (step S209) and ends the detection model construction process. For example, when the detection model DM whose evaluation index is a specified value is automatically selected, or when the detection model DM selected by the operator W is accepted, the server device 10 selects the detection model DM involved as the adopted model for the abnormal sign detection process.

[0225] (3-3. Abnormal sign output processing)

[0226] use Fig.13 The flow of the predicted-abnormality-sign output process of the predicted-abnormality-sign detection system 100 according to the embodiment will be described. Fig.13 This is a flowchart showing an example of the flow of the abnormality sign output process of the abnormality sign detection system 100 according to the embodiment. In addition, the processes of steps S301 to S303 described below may be executed in a different order. In addition, there may be omitted processes in the processes of steps S301 to S303 described below.

[0227] (3-3-1. Measurement data collection process)

[0228] First, the server device 10 executes a measurement data collection process (step S301 ). For example, the server device 10 collects measurement data acquired by the field device 20 .

[0229] (3-3-2. Abnormality sign detection and processing)

[0230] Second, the server device 10 performs abnormal sign detection processing (step S302). For example, the server device 10 detects abnormal signs by inputting the collected measurement data into the detection model DM as the adopted model. At this time, when there are multiple detection models DM determined as the adopted model, the server device 10 can detect abnormal signs using multiple detection models DM, or can detect abnormal signs using the detection model DM with the largest evaluation index among the multiple detection models DM.

[0231] (3-3-3. Test result transmission process)

[0232] Third, the server device 10 performs the detection result transmission process (step S303), and ends the abnormal sign output process. For example, the server device 10 transmits the detection result to the operator terminal 30 used by the operator W, and displays it on the monitor screen. At this time, when there are multiple detection models DM determined as the adopted model, the server device 10 can send all the multiple detection results to the operator terminal 30, or only notify the operator terminal 30 of the detection result outputted by the detection model DM with the largest evaluation index among the multiple detection models DM.

[0233] [4. Effects of implementation]

[0234] Finally, the effects of the embodiment will be described. Next, effects 1 to 10 corresponding to the processing according to the embodiment will be described.

[0235] (4-1. Effect 1)

[0236] First, in the process according to the above-mentioned embodiment, the server device 10 collects the measurement data measured by the field device 20, inputs the collected measurement data to the learned detection model DM that predicts the prescribed phenomenon based on the input of the measurement data, and obtains the output result of the learned detection model DM, and when the output result of the learned detection model DM and the confirmation result of the operator W who confirms that the field device 20 predicts the prescribed phenomenon are different, the learned detection model DM is re-learned using the label value input by the operator W and the collected measurement data to generate a re-learned model. Therefore, in this process, the accuracy of the learning model can be improved.

[0237] (4-2. Effect 2)

[0238] Second, in the process involved in the above-mentioned embodiment, the detection model DM that has been learned is a detection model DM that has been learned by using each measurement data as a normal value among a plurality of measurement data collected during a specified period for unsupervised machine learning of training data. In addition, the server device 10 performs re-learning of the learning model that has been learned by assigning a label value based on the judgment result as teacher data to the collected measurement data through supervised learning. Therefore, in this process, the accuracy of the learning model of unsupervised learning can be improved.

[0239] (4-3. Effect 3)

[0240] Third, in the process according to the above embodiment, when the output result of the learned detection model DM is equivalent to a sign of an abnormal state, and the abnormal state or the sign of abnormality state of the field device 20 is not confirmed according to the determination result of the operator W, the server device 10 performs re-learning of the learned detection model DM using teacher data obtained by assigning the presence or absence of the abnormal state or the sign of abnormality state based on the determination result as a label value to the collected measurement data. Therefore, in this process, the accuracy of the learned model for detecting signs of abnormality can be improved.

[0241] (4-4. Effect 4)

[0242] Fourth, in the process according to the above-mentioned embodiment, the server device 10 calculates the evaluation index of the learned detection model DM based on the detection result as the output of the learned detection model DM and the judgment result of the operator W who confirmed that the field instrument 20 predicted the specified phenomenon. Therefore, in this process, the accuracy of the learning model can be improved by evaluating the learning model based on unsupervised learning.

[0243] (4-5. Effect 5)

[0244] Fifth, in the process according to the above embodiment, the server device 10 performs relearning of the learned model based on the calculated evaluation index. Therefore, in this process, the accuracy of the learning model can be improved by relearning the learning model with a low evaluation index.

[0245] (4-6. Effect 6)

[0246] Sixth, in the process involved in the above-mentioned embodiment, the server device 10 obtains the output results of each of the plurality of learned detection models DM that predict the prescribed phenomenon, calculates the evaluation index of each of the plurality of learned detection models DM based on the output results of each of the plurality of learned detection models DM and the judgment result of the operator W who confirmed that the field instrument 20 predicted the prescribed phenomenon, and determines the selection of the learned detection model DM whose evaluation index is greater than or equal to the prescribed value among the plurality of learned detection models DM. Therefore, in this process, the accuracy of the learning model can be improved by automatically selecting a learning model with higher accuracy.

[0247] (4-7. Effect 7)

[0248] Seventh, in the process involved in the above-mentioned embodiment, the server device 10 determines the adoption of the learned detection model DM selected by the operator W among the multiple learned detection models DM. Therefore, in this process, the accuracy of the machine learning model can be improved by the operator W selecting a machine learning model with higher accuracy.

[0249] (4-8. Effect 8)

[0250] Eighth, in the process according to the above embodiment, the server device 10 calculates at least one of the correct answer rate, the fit rate, the reproducibility rate, the specificity, and the F value as an evaluation index based on the detection result of the output of the learned detection model DM and the judgment result of the operator W who confirmed that the field instrument 20 predicted the predetermined phenomenon, and displays the evaluation index on the operator terminal 30 used by the operator W. Therefore, in this process, the accuracy of the learning model can be improved by the operator W selecting a learning model with higher accuracy according to the type of the output result.

[0251] (4-9. Effect 9)

[0252] Ninth, in the process involved in the above-mentioned embodiment, the server device 10 generates a time series graph representing the history of the collected measurement data, and displays the generated time series graph on the operator terminal 30 used by the operator W, and determines the adoption of the learning model based on the displayed time series graph. Therefore, in this process, the accuracy of the machine learning model can be improved by referring to the time series data and selecting a machine learning model with higher accuracy by the operator W.

[0253] (4-10. Effect 10)

[0254] Tenth, in the processing involved in the above-mentioned embodiment, the server device 10 generates a normal status screen representing a normal status period as a detection result, and enables the operator terminal 30 used by the operator W to display the generated normal status screen in a manner that can be switched between display and non-display through the operation of the operator W.

[0255] [system]

[0256] The processing sequence, control sequence, specific names, and information including various data and parameters shown in the above description and drawings can be arbitrarily changed unless otherwise specified.

[0257] In addition, the components of each device shown in the figure are functional concepts and do not need to be physically configured as shown in the figure. That is, the specific method of distributing and integrating each device is not limited to the method shown in the figure. That is, the whole or part of the devices can be functionally or physically distributed / integrated in any unit according to various loads, usage conditions, etc.

[0258] Furthermore, all or any part of each processing function executed by each device may be realized by a CPU and a program analyzed and executed by the CPU, or may be realized as hardware based on wired logic.

[0259] [hardware]

[0260] Next, a hardware configuration example of the server device 10 as an alarm detection device will be described. In addition, other devices may also be configured with the same hardware configuration. Fig.14 1 is a diagram showing an example of a hardware configuration according to an embodiment of the present invention. Fig.14 As shown, the server device 10 includes a communication device 10a, a HDD (Hard Disk Drive) 10b, a memory 10c, and a processor 10d. Fig.14 The respective components shown are connected to each other using a bus or the like.

[0261] The communication device 10a is a network interface card, etc., and performs communication with other servers. Figure 2 The functions shown are stored in programs and databases that execute actions.

[0262] The processor 10d reads and executes data from the HDD 10b or the like. Figure 2 The same processing program is developed in the memory 10c, thereby executing Figure 2The process of each function described in the above etc. executes the action. For example, this process executes the same function as each processing unit of the server device 10. Specifically, the processor 10d reads a program having the same function as the collection unit 15a, the detection unit 15b, the execution unit 15c, etc. from the HDD 10b, etc. Then, the processor 10d executes the following process, that is, executes the same processing as the collection unit 15a, the detection unit 15b, the execution unit 15c, etc.

[0263] In this way, the server device 10 performs actions as a device that executes various processing methods by reading and executing programs. In addition, the server device 10 can also read the above program from the recording medium using a medium reading device and execute the read program, thereby realizing the same functions as the above embodiment. In addition, the program mentioned in this embodiment is not limited to being executed by the server device 10. For example, when other computers or servers execute the program, or when they cooperate to execute the program, the present invention can also be applied in the same way.

[0264] The program can be distributed via a network such as the Internet. In addition, the program can be recorded on a computer-readable recording medium such as a hard disk, a floppy disk (FD), a CD-ROM, an MO (Magneto-Optical disk), or a DVD (Digital Versatile Disc), and executed by being read from the recording medium by a computer.

Claims

1. A detection device, wherein: The detection device comprises: a collecting unit that collects measurement data measured by the measuring device; a detection unit that inputs the collected measurement data into a learned model that has been learned to predict a predetermined phenomenon based on input measurement data, and acquires a detection result that is an output of the learned model; as well as An execution unit, when a detection result of the learned model is different from a judgment result of a user of the measuring device who confirms that the specified phenomenon is predicted, performs re-learning of the learned model using a label value input by the user and the collected measurement data to generate a re-learned model.

2. The detection device according to claim 1, wherein: The learned learning model is a learning model that has been learned by unsupervised learning using each measurement data as a normal value among a plurality of measurement data collected during a predetermined period as training data, The execution unit performs re-learning of the learned learning model by assigning the label value based on the determination result as teaching data to the collected measurement data through taught learning.

3. The detection device according to claim 1, wherein: When the output of the learned learning model, i.e., the detection result, is equivalent to a precursor of an abnormal state, and no abnormal state or abnormal precursor state is confirmed based on the judgment result, the execution unit uses the teacher data obtained by assigning the presence or absence of the abnormal state or the abnormal precursor state based on the judgment result as the label value to the collected measurement data to perform re-learning of the learned learning model.

4. The detection device according to claim 1, wherein: The execution unit calculates an evaluation index of the learned model based on the detection result, which is an output of the learned model, and the determination result.

5. The detection device according to claim 4, wherein: The execution unit executes re-learning of the learned learning model based on the evaluation index.

6. The detection device according to claim 1, wherein: The detection unit obtains detection results that are outputs of each of a plurality of learned models that predict the predetermined phenomenon. The execution unit calculates an evaluation index for each of the plurality of learned learning models based on the detection result and the judgment result, which are outputs of each of the plurality of learned learning models. It is determined to adopt a learned model of the plurality of learned models, the learned model having the evaluation index greater than or equal to a predetermined value.

7. The detection device according to claim 1, wherein: The execution unit determines to adopt a learned learning model selected by the user among the plurality of learned learning models.

8. The detection device according to claim 1, wherein: The execution unit calculates at least one of the correct answer rate, the fit rate, the reproducibility rate, the specificity and the F value as an evaluation index of the learned model based on the detection result output by the learned model and the judgment result of the user of the measuring device who confirmed that the specified phenomenon was predicted, The evaluation index is displayed on a terminal used by the user.

9. The detection device according to claim 1, wherein: The execution unit generates a time series graph showing the history of the collected measurement data. causing the terminal used by the user to display the time series graph, The adoption of the learning model is determined based on the displayed time series graph.

10. The detection device according to claim 1, wherein: The execution unit generates a screen showing a period of a normal state as the detection result, The terminal used by the user is caused to display the generated screen in a manner that the display or non-display can be switched by the operation of the user.

11. A detection method, wherein: The following processing is performed by a computer, namely, Collecting measurement data measured by the measuring device, The collected measurement data is input to a learning model that has been learned to predict a predetermined phenomenon based on the input of the measurement data, and the output of the learning model, that is, the detection result, is obtained. When the output of the learned learning model, i.e., the detection result, is different from the judgment result of the user of the measuring device who confirmed that the specified phenomenon was predicted, the learned learning model is relearned using the label value input by the user and the collected measurement data to generate a relearned model.

12. A computer-readable recording medium having a detection program recorded thereon, wherein: The detection program causes the computer to execute the following processing, namely, Collecting measurement data measured by the measuring device, The collected measurement data is input to a learning model that has been learned to predict a predetermined phenomenon based on the input of the measurement data, and the output of the learning model, that is, the detection result, is obtained. When the output of the learned learning model, i.e., the detection result, is different from the judgment result of the user of the measuring device who confirmed that the specified phenomenon was predicted, the learned learning model is relearned using the label value input by the user and the collected measurement data to generate a relearned model.

Images