Support method, support device, and support system for machine learning application system

Abstract

To provide a support method of a machine learning application system capable of shortening the time from the setting of the machine learning application system to the start of operation and suppressing the deterioration of operation quality.SOLUTION: The support method of the machine learning-applied system includes an activation step S10 of receiving an activation instruction to the automatic setting wizard 102, a target identification step 72a of receiving selection of a process to be automatically executed by the computer from (1) a data-acquisition-and-setting process 72b for acquiring measurement-data 53a, (2) a data-accumulation process 53c, (3) a data-set-creation process 72c, (4) a learning process 53c for learning the machine-learning model 73a, and (5) a past evaluation process for evaluating the machine-learning model, and an entry screen generation step of dynamically generating an entry screen for collectively entering parameters necessary for automatically executing the selected process by the computer. S11 S12 71b.SELECTED DRAWING: Figure 5

Description

[Technical Field]

[0001] The present disclosure relates to an assistance method, an assistance device, and an assistance system that use a computer to assist the operation of a machine learning application system that performs applied processing using a machine learning model based on measurement data obtained from equipment. [Background technology]

[0002] BACKGROUND ART Conventionally, a facility anomaly detection system is known as a machine learning application system that performs applied processing using a machine learning model as AI (Artificial Intelligence) based on measurement data obtained from a facility (see, for example, Patent Document 1).

[0003] The equipment anomaly detection system of Patent Document 1 (referred to as a "trend calculation system" in Patent Document 1) is composed of sensors, edge devices, and cloud computers, and is configured to monitor and operate the equipment through a process in which the user repeatedly changes settings directly. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Publication No. 2024-25095 Summary of the Invention [Problem to be solved by the invention]

[0005] However, with the technology of Patent Document 1, each process requires time for data collection and time for learning and evaluating machine learning models, resulting in waiting times ranging from several hours to several months to complete each process. Furthermore, there are also waiting times between processes, so the overall process typically takes approximately 10 to 90 days. Furthermore, the complex operability increases the chance of configuration errors in each process, and operational quality may be reduced due to factors such as neglecting adjustments to achieve optimal operational conditions. Furthermore, even after operation has begun, reconfiguration may be required due to changes in equipment conditions or seasonal fluctuations, potentially further reducing operational quality.

[0006] Therefore, the present disclosure aims to provide a support method, support device, and support system for a machine learning application system that can shorten the time from setting up the machine learning application system to starting operation and suppress a decline in operational quality. [Means for solving the problem]

[0007] In order to achieve the above object, a method for supporting a machine learning application system according to one embodiment of the present disclosure is a method for supporting, by a computer, the operation of a machine learning application system that performs application processing using a machine learning model based on measurement data obtained from equipment, the method including: a startup step for receiving a startup instruction for an automatic setting wizard that provides the support; and, after the startup instruction is received in the startup step, a target identification step for accepting a selection of a scenario, which is a selection of a process to be automatically executed by a computer, from among (1) a first process for performing settings to acquire the measurement data, (2) a second process for accumulating the measurement data, (3) a third process for creating a dataset for training the machine learning model from the accumulated measurement data, (4) a fourth process for training the machine learning model using the dataset, and (5) a fifth process for evaluating the trained machine learning model; and an input screen generation step for dynamically generating, based on the selection in the target identification step, an input screen that is a user interface for inputting all parameters required for automatically executing the process selected in the target identification step by a computer, and accepting input for the generated input screen.

[0008] In order to achieve the above object, an assistance device for a machine learning application system according to one embodiment of the present disclosure is an apparatus that uses a computer to assist the operation of a machine learning application system that performs application processing using a machine learning model based on measurement data obtained from equipment, and includes: a launch unit that accepts a launch instruction for an automatic setting wizard that provides the assistance; a target identification unit that, after the launch instruction is accepted by the launch unit, accepts a selection of a scenario, which is a selection of a process to be automatically executed by a computer, from among (1) a first process that performs settings to acquire the measurement data, (2) a second process that accumulates the measurement data, (3) a third process that creates a dataset for training the machine learning model from the accumulated measurement data, (4) a fourth process that trains the machine learning model using the dataset, and (5) a fifth process that evaluates the trained machine learning model; and an input screen generation unit that dynamically generates an input screen, which is a user interface for inputting all parameters required for automatically executing the process selected by the target identification unit on a computer, based on the selection by the target identification unit, and accepts input for the generated input screen.

[0009] In order to achieve the above-mentioned object, a support system for a machine learning application system according to one embodiment of the present disclosure is a system that uses a computer to support the operation of a machine learning application system that performs application processing using a machine learning model based on measurement data obtained from equipment, and includes the above-mentioned support device and a terminal device that accesses the support device via a communication network and, in accordance with instructions from a user, selects a scenario for the target identification unit provided in the support device and inputs data into the input screen generated by the input screen generation unit provided in the support device.

[0010] The present disclosure can be realized not only as an assistance method, assistance device, and assistance system for a machine learning application system, but also as a program that causes a computer to execute the steps included in the assistance method for a machine learning application system, or as a non-transitory recording medium such as a computer-readable DVD on which the program is recorded. [Effects of the Invention]

[0011] The present disclosure provides a method, device, and system for supporting a machine learning application system that can shorten the time from setting up the machine learning application system to starting operation and prevent a decline in operational quality. [Brief explanation of the drawings]

[0012] [Figure 1] FIG. 1 is a block diagram showing the configuration of a facility anomaly detection system, which is an example of a machine learning application system that is a target of a support system according to an embodiment. [Figure 2] FIG. 2 is a block diagram showing the configuration of the server shown in FIG. [Figure 3] FIG. 3 is a diagram showing a process flow from installation to operation of the equipment anomaly detection system shown in FIG. [Figure 4] FIG. 4 is a block diagram showing the configuration of the support system according to the embodiment. [Figure 5] FIG. 5 is a flowchart showing an operation procedure of the support device of the support system according to the embodiment. [Figure 6] FIG. 6 is a diagram illustrating the operation pattern of the equipment to be monitored. [Figure 7A] FIG. 7A is a diagram showing an example of a user interface in the activation step by the activation unit in FIG. [Figure 7B] FIG. 7B is a diagram showing an example of a user interface in the target identification step by the target identification unit in FIG. [Figure 7C]FIG. 7C is a diagram showing an example of a user interface of a batch input screen generated in the input screen generating step by the input screen generating unit in FIG. DETAILED DESCRIPTION OF THE INVENTION

[0013] Hereinafter, embodiments of the present disclosure will be described in detail with reference to the drawings. Note that each embodiment described below represents a specific example of the present disclosure. The numerical values, specific configurations of the equipment, components of the device, the arrangement and connection of the components, steps, the order of steps, etc. shown in the following embodiments are merely examples and are not intended to limit the present disclosure. Furthermore, each drawing is not necessarily an exact illustration. In each drawing, substantially identical components are assigned the same reference numerals, and duplicate explanations are omitted or simplified.

[0014] First, before describing the assistance method, assistance device, and assistance system according to the embodiment, an outline of a machine learning application system that is the target of such assistance will be described with reference to FIGS. 1 to 3. FIG.

[0015] 1 is a block diagram showing the configuration of an equipment anomaly detection system 10, which is an example of a machine learning application system that is a target of a support system according to an embodiment. The equipment anomaly detection system 10 is an example of a machine learning application system that performs application processing (here, anomaly detection of the equipment 40) using a machine learning model 53c based on measurement data 53a obtained from the equipment 40, and is composed of the equipment 40 and a terminal device 30 installed in a factory 20, and a server 50 that constitutes a cloud computer connected to them via a router 61 and a communication network 60.

[0016] The equipment 40 is composed of a motor 41, a motor driver 42 that drives the motor 41, a current sensor 43 that non-contactly measures the current flowing through a drive cable that connects the motor 41 and the motor driver 42, an edge slave device 44 that generates measurement data 53a that indicates the current measured by the current sensor 43, and an edge master device 45 that transmits the measurement data 53a generated by the edge slave device 44 to the terminal device 30 via a router 61. Note that the edge master device 45 may transmit the measurement data 53a directly to the server 50 via the router 61 and the communication network 60.

[0017] The terminal device 30 is a terminal device that transmits measurement data 53a sent from the edge parent device 45 via the router 61 to the server 50 via the router 61 and the communication network 60, and allows the user to interact with the server 50 via a web browser.

[0018] Fig. 2 is a block diagram showing the configuration of the server 50 shown in Fig. 1. The server 50 includes a communication unit 51 which is a communication interface connected to the communication network 60, an anomaly detection processing unit 52 which provides an anomaly detection service for the equipment 40, and a storage unit 53 which stores measurement data 53a and the like.

[0019] The abnormality detection processing unit 52 is a functional processing unit realized by the processor of the server 50 executing a program, and has a data acquisition and setting unit 52a, a data accumulation unit 52b, a data set creation unit 52c, a learning unit 52d, a past evaluation unit 52e, and a diagnosis unit 52f as processing units based on interaction with the user via the terminal device 30.

[0020] The data acquisition setting unit 52a assists the user in setting conditions related to waveform cutting, which are extraction conditions for the measurement data 53a acquired from the equipment 40. The conditions related to waveform cutting include, for example, a minimum amplitude that specifies a portion of the current waveform obtained by the current sensor 43 to be cut out to be used as the measurement data 53a.

[0021] The data accumulation unit 52b acquires the measurement data 53a obtained in the equipment 40 via the communication unit 51 and accumulates the data in the storage unit 53 for learning of the machine learning model 53c.

[0022] The dataset creation unit 52c performs various filtering operations on the measurement data 53a stored in the memory unit 53, including filtering of normal periods specified by the user, to create a dataset 53b for learning the machine learning model 53c and store the dataset 53b in the memory unit 53.

[0023] The learning unit 52d creates a trained machine learning model 53c by performing learning on the machine learning model 53c using the data set 53b stored in the storage unit 53. The learning unit 52d creates the machine learning model 53c that can recognize normal data as normal by learning the machine learning model 53c using the data set 53b created based on the measurement data 53a obtained when the equipment 40 is operating normally (i.e., using only normal data) based on, for example, a one-class SVM (Support Vector Machine) or the like, and stores the machine learning model 53c in the storage unit 53.

[0024] The past evaluation unit 52e evaluates the created machine learning model 53c using the past measurement data 53a stored in the storage unit 53. If the user determines that the evaluation result does not satisfy the criteria, the user performs measurements again, reviews the waveform extraction conditions, reviews various filtering for creating a data set, reviews various conditions during learning, and so on, and then re-learns the machine learning model 53c.

[0025] The diagnosis unit 52f is a processing unit that supports the operation of the equipment anomaly detection system 10, and detects an abnormality or a sign of an abnormality in the equipment 40 based on the measurement data 53a obtained from the equipment 40 in real time using the machine learning model 53c that has been determined to satisfy the criteria, and presents the result to the terminal device 30. When input data corresponding to the measurement data 53a is given to the machine learning model 53c, the machine learning model 53c outputs a degree of change that indicates the degree of deviation from a state in which the equipment 40 is operating normally, and the diagnosis unit 52f presents the degree of change to the user as information indicating the degree of abnormality of the equipment 40 or a sign of an abnormality, or executes an "email alert" that sends an email when the degree of change exceeds a preset threshold.

[0026] Fig. 3 is a diagram showing the process flow from installation to operation of the equipment anomaly detection system 10 shown in Fig. 1. From the bottom, the diagram shows the process flow, the flow of the phases to which each process belongs, and an overview diagram explaining each phase. Note that the diagram also explains the processes that are the target of support by the support system according to the embodiment.

[0027] As shown in this figure, the process from installation to operation of the equipment anomaly detection system 10 can be roughly divided into an installation phase 71, a learning phase 72, and an operation phase 73.

[0028] The installation phase 71 is composed of an equipment installation process 71a in which the user installs the current sensor 43, edge slave unit 44, and edge master unit 45 in the factory 20, and a data acquisition setting process 71b in which the user sets extraction conditions for the measurement data 53a by interacting with the data acquisition setting unit 52a of the server 50. The overview diagram of the installation phase 71 illustrates a setting example 75 in which "a waveform is transmitted only when the target part moves" as an example of setting the extraction conditions for the measurement data 53a in the data acquisition setting process 71b.

[0029] The learning phase 72 is composed of a data accumulation process 72a in which the user activates the data accumulation unit 52b of the server 50 to accumulate measurement data 53a obtained at the equipment 40 in the memory unit 53 of the server 50, a dataset creation process 72b in which the user interacts with the dataset creation unit 52c of the server 50 to create a dataset 53b for training the machine learning model 53c, and a learning process 72c in which the user interacts with the learning unit 52d of the server 50 to train the machine learning model 53c. The overview diagram of the learning phase 72 illustrates an example 76a of setting a normal period in the dataset creation process 72b, an example 76b of specifying a dataset to be used for training in the learning process 72c, and an example 76c of setting learning conditions in the learning process 72c.

[0030] The operation phase 73 is composed of a past evaluation process 73a in which a user interacts with the past evaluation unit 52e of the server 50 to evaluate the machine learning model 53c, and an operation process 73b in which the user activates the diagnosis unit 52f of the server 50 to detect abnormalities and their precursors in the equipment 40. The overview diagram of the operation phase 73 shows examples 77a and 77b of presentations to the user in the operation process 73b, and an example 77c of an "email alert" setting in the operation process 73b.

[0031] Of all the processes shown in Figure 3, from the equipment installation process 71a to the operation process 73b, each process from the data acquisition and setting process 71b to the operation process 73b is the process that is the target of the support system of this embodiment, which will be described next.

[0032] 4 is a block diagram showing the configuration of a support system 110 according to an embodiment. The support system 110 is a system that uses a computer to support the operation of a machine learning application system that performs application processing using a machine learning model based on measurement data 53a obtained from equipment 40, and in this embodiment, supports the operation of an equipment anomaly detection system 10. To this end, the support system 110 includes a support device 100 that is connected to the equipment anomaly detection system 10 that is the support target via a communication network 60 or the like. Note that the components of the support system 110 may also include the components of the equipment anomaly detection system 10 (such as the terminal device 30 and the equipment 40).

[0033] The support device 100 is a device that provides a user interface that simplifies and automates each process from the data acquisition setting process 71b to the operation process 73b shown in Fig. 3, and includes a communication unit 101 that is a communication interface connected to the communication network 60, an automatic setting wizard 102 that provides the user interface, and a storage unit 103. The support device 100 may be configured on the same computer as the server 50 of the equipment anomaly detection system 10, or may be configured on a different computer.

[0034] The automatic setting wizard 102 is a functional processing unit realized by the processor of the support device 100 executing a program, and has a launching unit 102a, a target identification unit 102b, an input screen generation unit 102c, and an automatic execution unit 102d as processing units based on dialogue with the user via the terminal device 30.

[0035] The launching unit 102 a receives a launch instruction for the automatic setting wizard 102 .

[0036] After the startup instruction is received by the startup unit 102a, the target identification unit 102b receives a selection of a scenario, which is a selection of processes to be automatically executed by a computer, from among (1) a data acquisition and setting process 71b, which is a first process for setting up to acquire the measurement data 53a, (2) a data accumulation process 72a, which is a second process for accumulating the measurement data 53a, (3) a dataset creation process 72b, which is a third process for creating a dataset for training the machine learning model 53c from the accumulated measurement data 53a, (4) a learning process 72c, which is a fourth process for training the machine learning model 53c using the dataset, and (5) a past evaluation process 73a, which is a fifth process for evaluating the trained machine learning model 53c. Furthermore, the target identification unit 102b receives a selection of an operation pattern to be monitored from among multiple operation patterns in which the equipment 40 can operate (i.e., multiple operations to be measured).

[0037] Based on the selection made by the target identification unit 102b, the input screen generation unit 102c dynamically generates an input screen, which is a user interface for inputting all parameters required for automatically executing the process selected by the target identification unit 102b on a computer, and accepts input for the generated input screen. In other words, the input screen generation unit 102c generates an input screen required for automatically executing the process selected by the target identification unit 102b on a computer, with the operation pattern accepted by the target identification unit 102b as the target.

[0038] Furthermore, when the data acquisition setting process 71b is selected in the scenario selection by the target identification unit 102b, the input screen generation unit 102c generates, as the input screen, an input screen including an input field for inputting a set deadline, which is the maximum time required to identify measurement data 53a that satisfies predetermined conditions as measurement data 53a to be accumulated; when the dataset creation process 72b is selected in the scenario selection by the target identification unit 102b, the input screen generation unit 102c generates, as the input screen, an input screen including an input field for inputting the type of machine learning model 53c; when the dataset creation process 72b is selected in the scenario selection by the target identification unit 102b, the input screen generation unit 102c generates, as the input screen, an input screen including an input field for inputting the number of times the machine learning model 53c is trained; and when the dataset creation process 72b is selected in the scenario selection by the target identification unit 102b, the input screen generation unit 102c generates, as the input screen, an input screen including an input field for inputting an operation pattern to be trained by the machine learning model 53c.

[0039] The automatic execution unit 102d automatically executes the processes that are the targets of the scenario in sequence by a computer in accordance with the input received on the input screen generated by the input screen generation unit 102c. At this time, when the automatic execution unit 102d executes the past evaluation process 73a, it evaluates the machine learning model 53c using the measurement data 53a obtained in the past.

[0040] FIG. 5 is a flowchart showing the operation procedure of the support device 100 of the support system 110 according to the embodiment.

[0041] In the support device 100, first, the activation unit 102a receives an activation instruction for the automatic setting wizard 102 through a dialogue with the user via the terminal device 30 (activation step S10).

[0042] Next, the target identification unit 102b, through dialogue with the user via the terminal device 30, accepts a scenario selection, which is a selection of processes to be automatically executed by the computer from among the data acquisition setting process 71b, the data accumulation process 72a, the data set creation process 72b, the learning process 72c, and the past evaluation process 73a, and a selection of one of the multiple operating patterns in which the equipment 40 can operate to be the target of monitoring (target identification step S11).

[0043] FIG. 6 is a diagram illustrating the operation patterns of the equipment 40 to be monitored. The horizontal axis indicates the elapsed time since the equipment 40 started operating, and the vertical axis indicates the operating frequency of the equipment 40. To enable selection of an operation pattern, the support device 100 performs frequency analysis using the measurement data 53a acquired from the current sensor 43 to obtain an intensity distribution of the operation frequency and time. In this intensity distribution, when the equipment 40 operates intermittently, the operation frequency increases at the start ("A" in FIG. 6), stabilizes for a certain period of time ("B" in FIG. 6), and then decreases at the end ("C" in FIG. 6). In the case of steady operation, the operation frequency is always stable ("B" in FIG. 6). Through this analysis, the following types of operation patterns of the operation frequency at the start and end of equipment operation are prepared.

[0044] (1) Acceleration / deceleration operation (ABC or AC in FIG. 6; first operation pattern): Includes the operation of the equipment 40 from start to finish. (2) Constant speed operation (B in FIG. 6; second operation pattern): This includes a section in which the equipment 40 is steadily operating. This is normal constant speed operation that prioritizes time resolution. (3) Acceleration operation (A or AB in FIG. 6; third operation pattern): Includes only the start of operation of the equipment 40. (4) Deceleration operation (C or BC in FIG. 6; fourth operation pattern): Includes only the end of operation of the equipment 40. (5) Constant-speed (high-resolution) operation (B in FIG. 6; fifth operation pattern): Includes a section in which the equipment 40 is operating steadily. However, compared to the normal constant-speed operation described above, which prioritizes time resolution, this is constant-speed operation that prioritizes frequency resolution and allows the calculation of the fundamental frequency to select a current waveform.

[0045] The motion pattern selected under the user interface by the target specifying unit 102b of the assistance device 100 is used for recursive adjustment in the data acquisition setting process 71b.

[0046] Returning to FIG. 5 again, next, the input screen generation unit 102c dynamically generates an input screen, which is a user interface for inputting all the parameters required for automatically executing the process selected by the target identification unit 102b on a computer, based on the selection made by the target identification unit 102b, and accepts input from the user on the generated input screen (input screen generation step S12).

[0047] Finally, the automatic execution unit 102d automatically executes the processes that are the targets of the scenario in sequence by the computer in accordance with the input received on the input screen generated by the input screen generation unit 102c (automatic execution step S13). At this time, when the automatic execution unit 102d executes the past evaluation process 73a, it evaluates the machine learning model 53c using the measurement data 53a obtained in the past.

[0048] 7A to 7C are diagrams showing specific examples of user interfaces (that is, examples of screen displays on the terminal device 30) in each step of FIG.

[0049] 7A is a diagram showing an example user interface 120 (left diagram) in activation step S10 by activation unit 102a in Fig. 5, and a data flow 130 (right diagram) showing the process and data generation supported by user interface example 120. In data flow 130, ellipses indicate processes to be supported, and document symbols indicate input data to the process or output data from the process.

[0050] As shown in FIG. 7A, when the button "Automatic Configuration Wizard" 120a in the example user interface 120 is clicked in the startup step S10 by the startup unit 102a, the automatic configuration wizard 102 is started by the startup unit 102a ("Start Automatic Configuration Wizard" in the data flow 130), and the example user interface 121 shown in FIG. 7B is presented ("GUI" in the data flow 130).

[0051] In the user interface example 120 of FIG. 7A, the button "Automatic Setting Wizard" 120a is displayed as one item on the input screen of the "Operation Sensor List" for selecting the type of current sensor 43, etc., to be used in the equipment anomaly detection system 10, but the present invention is not limited to this input screen, and the button "Automatic Setting Wizard" 120a may be displayed as one item on another type of input screen or as an independent input screen.

[0052] 7B is a diagram showing a user interface example 121 (left diagram) in the target identification step S11 by the target identification unit 102b in Fig. 5, and a data flow 130 (right diagram) showing the process and data generation supported by the user interface example 121. Here, examples of "setting scenario" 121a ("select operation pattern" in the data flow 130) for selecting a process to be automatically executed, and "operation pattern" 121b ("select scenario" in the data flow 130) for selecting from a plurality of operation patterns in which the equipment 40 can operate are shown.

[0053] The "setting scenario" 121a shows an example of selecting from a first scenario 121a1 consisting of all processes including the data acquisition setting process 71b, the data accumulation process 72a, the data set creation process 72b, the learning process 72c, and the past evaluation process 73a; a second scenario 121a2 consisting of the data accumulation process 72a, the data set creation process 72b, and the learning process 72c (i.e., the process from data accumulation to machine learning); and a third scenario 121a3 consisting of only the learning process 72c (i.e., only the machine learning process).

[0054] Furthermore, "operation pattern" 121b shows an example in which one of "acceleration" 121b1 in which the operating frequency of equipment 40 increases, "deceleration" 121b2 in which the operating frequency decreases, "acceleration / deceleration" 121b3 which includes acceleration and deceleration, "constant speed" 121b4 in which the operating frequency does not change substantially, and "constant speed (high resolution)" 121b5 in which frequency resolution is prioritized over constant speed is selected. Here, "the operating frequency does not change substantially" means that the operating frequency does not change substantially, and refers to a state in which the operating frequency fluctuates within a range of ±20% (more particularly ±10%, and even more particularly ±5%) from its average value, for example.

[0055] In addition, in the "setting scenario" 121a in the user interface example 121 of Figure 7B, a method is adopted in which a selection is made from three representative scenarios (i.e., the first scenario 121a1, the second scenario 121a2, and the third scenario 121a3), but this method is not limited to this, and a method in which a desired process is arbitrarily selected from all five types of processes may also be used.

[0056] 7B, the "operation pattern" 121b is selected from five types of operation patterns, but the present invention is not limited to this, and the support device 100 may automatically identify an operation pattern to be monitored. That is, in the target identification step S11, the support device 100 may extract important feature quantities such as fundamental frequency and amplitude from the measurement data 53a obtained by measuring the equipment anomaly detection system 10, classify (i.e., cluster) the measurement data 53a based on the extracted feature quantities, and automatically select an appropriate operation pattern including at least one of acceleration, constant speed, constant speed (high resolution), and deceleration based on the classification result.

[0057] When input on the user interface example 121 shown in FIG. 7B is completed, the user interface example 122 shown in FIG. 7C is presented (the "setting screen" generated by the "generate setting screen" in the data flow 130).

[0058] Figure 7C is a diagram showing an example user interface 122 (left diagram) of a bulk input screen generated in input screen generation step S12 by the input screen generation unit 102c in Figure 5, and a data flow 130 (right diagram) showing the process and data generation supported by the example user interface 122.

[0059] The user interface example 122 in FIG. 7C displays an "operation pattern" 122a1 as a setting item for the "waveform extraction setting" 122a, and displays an input item for a "set deadline [time]" 122a2. Here, the "waveform extraction setting" 122a is one of the input items required for the data acquisition setting process 71b, and is one of the predetermined conditions that the measurement data 53a to be accumulated must satisfy. It is a condition setting that specifies a portion of the current waveform obtained by the current sensor 43 to be extracted for use as the measurement data 53a. The "operation pattern" 122a1 is an item for inputting an operation pattern to be accumulated as the measurement data 53a. The "set deadline [time]" 122a2 is an item for inputting the maximum time required to identify the measurement data 53a that satisfies the predetermined conditions as the measurement data 53a to be accumulated.

[0060] 7C further displays an input item for "accumulation period" 122b1 as a setting item for "data accumulation" 122b. "Data accumulation" 122b is an input item required for data accumulation process 72a, and is an item for setting conditions for data accumulation process 72a. "Accumulation period" 122b1 is one of the conditions for data accumulation process 72a, and is an item for inputting the period for accumulating measurement data 53a.

[0061] Furthermore, the user interface example 122 in FIG. 7C further displays, as setting items related to "AI learning and evaluation" 122c, an input item for "AI frame" 122c1, an input item for "number of times learning" 122c2, and an input item for "waveform selection" 122c3. "AI learning and evaluation" 122c is an input item required for the learning process 72c. "AI frame" 122c1 is an input item for inputting the type of machine learning model 53c. "Number of times learning" 122c2 is an input item for inputting the number of times the machine learning model 53c is trained. "Waveform selection" 122c3 is an input item for inputting the behavior pattern to be learned by the machine learning model 53c.

[0062] After completing the input in the user interface example 122 shown in Fig. 7C ("Setting screen" and "Settings" in the data flow 130), when a "Start setting" button 123 displayed in the lower right of the user interface example 122 shown in Fig. 7C is clicked, the automatic execution unit 102d automatically executes the processes that are the targets of the scenario in sequence according to the input settings ("Start setting" in the data flow 130). At this time, if the past evaluation process 73a is selected as the target of the scenario, the machine learning model 53c is evaluated using the measurement data 53a obtained in the past.

[0063] As described above, the method for supporting a machine learning application system according to this embodiment is a method for supporting, by a computer, the operation of an equipment anomaly detection system 10, which is an example of a machine learning application system that uses measurement data 53a obtained from equipment 40 as input data and performs applied processing using a machine learning model 53c, and includes a startup step S10 for receiving a startup instruction for an automatic setting wizard 102 that provides support, and after the startup instruction is received in the startup step S10, (1) a data acquisition setting process 71b for making settings for acquiring the measurement data 53a, (2) a data accumulation process 72a for accumulating the measurement data 53a, and (3) a process for training the machine learning model 53c from the accumulated measurement data 53a. (4) a dataset creation process 72b that creates a dataset for learning a machine learning model 53c using the dataset; and (5) a past evaluation process 73a that evaluates the learned machine learning model 53c. From these, a target identification step S11 that accepts a selection of a scenario, which is a selection of a process to be automatically executed by a computer, and an input screen generation step S12 that dynamically generates an input screen, which is a user interface for inputting all at once the parameters required for automatically executing the process selected in the target identification step S11 by a computer, based on the selection in the target identification step S11, and accepts input for the generated input screen.

[0064] As a result, based on the selection of a scenario, which is the selection of processes to be automatically executed by a computer, an input screen, which is a user interface for inputting all the parameters required for automatically executing the selected processes on a computer, is dynamically generated.This allows for more intuitive configuration than conventional techniques in which parameters are input for each individual process, and reduces operational variation through standardization of system configuration and operation.As a result, a method for supporting machine learning application systems is realized that shortens the time from setting up the machine learning application system to starting operation and can suppress a decline in operational quality.

[0065] Here, the support method may further include an automatic execution step S13 in which the processes targeted by the scenario are automatically executed in sequence by a computer in accordance with the input received in the input screen generation step S12. As a result, when the execution of one process is completed, the next process is automatically started, thereby shortening the time from setting up the machine learning application system to starting operation.

[0066] In addition, in this embodiment, the applied processing performed by the machine learning application system is processing for detecting an abnormality in the equipment 40. This realizes a method for supporting the operation of the equipment anomaly detection system 10 by a computer.

[0067] Furthermore, the equipment 40 may operate in a plurality of operation patterns, and in the target identification step S11, a selection from the plurality of operation patterns may be accepted, and in the input screen generation step S12, an input screen required for automatically executing the process selected in the target identification step S11 on a computer may be generated as the input screen, with the operation pattern selected in the target identification step S11 as the target. This allows the equipment 40 to be efficiently analyzed by the machine learning application system by limiting the operations of the equipment 40 to be supported to a type of interest.

[0068] Furthermore, in the input screen generation step S12, when the data acquisition setting process 71b is selected in the target identification step S11, an input screen may be generated that includes an input field for inputting a set deadline, which is the maximum time required to identify the measurement data 53a that satisfies the predetermined conditions as the measurement data 53a to be accumulated. This enables efficient analysis of the equipment 40 by the machine learning application system by targeting only the measurement data 53a that satisfies the desired conditions.

[0069] Furthermore, in the input screen generation step S12, when the dataset creation process 72b is selected in the target identification step S11, an input screen including an input field for inputting the type of machine learning model 53c may be generated as the input screen, thereby allowing the machine learning application system to execute application processing using the desired type of machine learning model 53c.

[0070] Furthermore, in the input screen generation step S12, when the dataset creation process 72b is selected in the target identification step S11, an input screen including an input field for inputting the number of times the machine learning model 53c is trained may be generated as the input screen. This standardizes the training process for the machine learning model 53c, and improves the accuracy and reliability of the application process by the machine learning application system.

[0071] Furthermore, if the equipment 40 operates in a plurality of operation patterns, and the data set creation process 72b is selected in the target identification step S11, the input screen generation step S12 may generate an input screen including input fields for inputting the operation pattern to be learned by the machine learning model 53c. This standardizes the learning process for the machine learning model 53c, and improves the accuracy and reliability of the application process by the machine learning application system.

[0072] Furthermore, when the past evaluation process 73a is executed in the automatic execution step S13, the machine learning model 53c may be evaluated using previously obtained measurement data 53a. This verifies the performance of the machine learning model 53c, confirms the accuracy and range of application of the machine learning model 53c, and enables re-learning or parameter adjustment as necessary.

[0073] Furthermore, the scenario selection may include a selection from among a first scenario consisting of the data acquisition setting process 71b, the data accumulation process 72a, the data set creation process 72b, the learning process 72c, and the past evaluation process 73a (i.e., all processes), a second scenario consisting of the data accumulation process 72a, the data set creation process 72b, and the learning process 72c (i.e., the processes from data accumulation to machine learning), and a third scenario consisting of only the learning process 72c (i.e., only the machine learning process). This allows the user to select the most suitable scenario from three representative types depending on their own needs and the current state of the facility 40.

[0074] The plurality of operation patterns may include acceleration where the operation frequency of the equipment 40 increases, constant speed where the operation frequency remains substantially unchanged, and deceleration where the operation frequency decreases, and the target identification step S11 may accept a selection of either a first operation pattern including acceleration and deceleration, or a second operation pattern including constant speed. This allows applied processing such as monitoring by the machine learning application system to be performed on either the first operation pattern corresponding to intermittent operation in the equipment 40 or the second operation pattern corresponding to steady operation in the equipment 40.

[0075] Furthermore, in the target identification step S11, in addition to the first and second operation patterns, a selection of either a third operation pattern including only acceleration or a fourth operation pattern including only deceleration may be accepted. This allows for further selection from among the third operation pattern corresponding to the start of the equipment 40 or the fourth operation pattern corresponding to the stop of the equipment 40, thereby enabling applied processing such as monitoring by a machine learning application system to be performed on dynamic state changes of the equipment 40.

[0076] Furthermore, in the target identification step S11, in addition to the first and second operation patterns, a selection of a fifth operation pattern including a constant speed that prioritizes frequency resolution over constant speed may be accepted. By adding the fifth operation pattern including a constant speed that prioritizes frequency resolution in the facility 40 to the options, it is expected that the classification accuracy of application processes such as monitoring by a machine learning application system will be improved.

[0077] The multiple operation patterns may include acceleration, where the operation frequency of the equipment 40 increases, constant speed, where the operation frequency remains almost unchanged, and deceleration, where the operation frequency decreases, and in the target identification step S11, feature amounts may be extracted from the measurement data 53a obtained by measuring the equipment anomaly detection system 10, the measurement data 53a may be classified based on the extracted feature amounts, and an operation pattern including at least one of acceleration, constant speed, and deceleration may be selected based on the classification results. This enables automatic selection of the operation pattern, simplifies the complicated operation of the machine learning application system, and also simplifies resetting due to changes in the condition of the equipment 40, seasonal fluctuations, etc., and suppresses deterioration in operational quality.

[0078] The present disclosure may be realized not only as a support method, but also as a program that causes a computer to execute steps included in the support method for the equipment anomaly detection system 10 described above.

[0079] Furthermore, the support device for the machine learning application system according to this embodiment is a support device 100 for the equipment anomaly detection system 10, which uses measurement data 53a obtained from equipment 40 as input data and performs application processing using a machine learning model 53c, by using a computer. The support device 100 includes a startup unit 102a that receives a startup instruction for an automatic setting wizard 102 that provides support, and after the startup instruction is received by the startup unit 102a, (1) a data acquisition setting process 71b that performs settings for acquiring the measurement data 53a, (2) a data accumulation process 72a that accumulates the measurement data 53a, and (3) a process for training the machine learning model 53c from the accumulated measurement data 53a. The system includes: (1) a dataset creation process 72b that creates a dataset for the process; (2) a learning process 72c that uses the dataset to train a machine learning model 53c; and (3) a past evaluation process 73a that evaluates the learned machine learning model 53c. The system also includes a target identification unit 102b that accepts a selection of a scenario, which is a selection of a process to be automatically executed by a computer, from among the following: (4) a dataset creation process 72b that creates a dataset for the process; (5) a learning process 72c that uses the dataset to train a machine learning model 53c; and (6) a past evaluation process 73a that evaluates the learned machine learning model 53c. The system also includes an input screen generation unit 102c that dynamically generates an input screen, which is a user interface for inputting all the parameters required for the process selected by the target identification unit 102b to be automatically executed by a computer, based on the selection by the target identification unit 102b, and accepts input for the generated input screen.

[0080] Furthermore, the support system for the machine learning application system according to this embodiment is a support system that uses a computer to support the operation of the equipment anomaly detection system 10 that uses measurement data 53a obtained from equipment 40 as input data and performs application processing using a machine learning model 53c, and includes the support device 100 and a terminal device 30 that accesses the support device 100 via a communication network 60 and, in accordance with instructions from a user, selects a scenario for the target identification unit 102b included in the support device 100 and inputs data into the input screen generated by the input screen generation unit 102c included in the support device 100.

[0081] As a result, based on the selection of a scenario, which is the selection of processes to be automatically executed by a computer, an input screen, which is a user interface for inputting all the parameters necessary for automatically executing the selected processes on a computer, is dynamically generated.This means that, compared to conventional technologies in which parameters are input for each individual process, the time from setting up a machine learning application system to starting operation is shortened, and a method for supporting machine learning application systems is realized that can suppress a decline in operational quality.

[0082] While the method, device, and system for supporting a machine learning application system according to the present disclosure have been described above based on the embodiments, the present disclosure is not limited to these embodiments. As long as they do not deviate from the gist of the present disclosure, various modifications conceivable by those skilled in the art to the present embodiments and other forms constructed by combining some of the components of the embodiments are also included within the scope of the present disclosure.

[0083] For example, in the above embodiment, the machine learning application system to be supported was the equipment anomaly detection system 10, but this is not limited to this and any machine learning application system that performs applied processing using a machine learning model based on measurement data obtained from the equipment may be used.For example, it may be a machine learning application system that uses a machine learning model based on measurement data obtained from the equipment to predict the lifespan of the equipment, predict the health of the equipment, or predict when a failure will occur.

[0084] Furthermore, in the above embodiment, the measurement data is a current waveform obtained by a non-contact current sensor, but this is not limiting and the measurement data may be, for example, a waveform of power consumed by the facility.

[0085] Furthermore, in the above embodiment, the server 50 of the equipment anomaly detection system 10 is a cloud computer, but it may also be an on-premise server installed in the factory 20. Similarly, the support device 100 that supports the equipment anomaly detection system 10 may also be installed in the factory 20, or may be realized by the same computer as the server 50 of the equipment anomaly detection system 10. [Industrial Applicability]

[0086] The present disclosure can be used as a support device that uses a computer to support the operation of a machine learning application system that performs applied processing using a machine learning model based on measurement data obtained from equipment, for example, as a cloud computer that supports everything from the setup of an equipment anomaly detection system to the start of operation. [Explanation of symbols]

[0087] 10 Equipment abnormality detection system 20 Factories 30 Terminal Equipment 40 Equipment 41 Motor 42 Motor driver 43 Current Sensor 44 Edge Sub-unit 45 Edge parent unit 50 servers 51 Communications Department 52 Abnormality detection processing unit 52a Data acquisition setting section 52b Data storage section 52c Dataset Creation Department 52d Learning Department 52e Past Evaluation Department 52f Diagnostic Department 53 Storage section 53a Measurement data 53b dataset 53c Machine Learning Model 60 Communication Network 61 Router 71 Installation Phase 71a Equipment Installation Process 71b Data Acquisition Setup Process 72 Learning Phase 72a Data accumulation process 72b Dataset Creation Process 72c Learning Process 73 Operational Phase 73a Past Evaluation Process 73b Operational Process 100 Support device (support device for machine learning application systems) 101 Communications Department 102 Auto Configuration Wizard 102a Starting section 102b Target Identification Section 102c Input screen generation unit 102d Automatic execution section 103 Storage section 110 Support System (Support System for Machine Learning Application Systems) 120~122 User interface example 120a Button "Automatic Configuration Wizard" 121a "Configuration Scenario" 121a1 Scenario 1 121a2 Second Scenario 121a3 Third Scenario 121b "Movement Pattern" 121b1 "Acceleration" 121b2 "Deceleration" 121b3 "Acceleration / Deceleration" 121b4 "Constant velocity" 121b5 "Constant velocity (high resolution)" 122a "Waveform Extraction Settings" 122a1 "Movement Pattern" 122a2 “Setting deadline [time]” 122b "Data Accumulation" 122b1 "Accumulation Period" 122c "AI Learning and Evaluation" 122c1 "AI Frame" 122c2 "Number of learning sessions" 122c3 "Waveform Selection"

Claims

1. A method for supporting the operation of a machine learning application system that performs application processing using a machine learning model based on measurement data obtained from equipment, using a computer, a startup step of receiving a startup instruction for an automatic setting wizard that provides the support; After the startup instruction is accepted in the startup step, a target identification step is performed to accept a selection of a scenario, which is a selection of a process to be automatically executed by a computer, from among (1) a first process for setting up to acquire the measurement data, (2) a second process for accumulating the measurement data, (3) a third process for creating a dataset for training the machine learning model from the accumulated measurement data, (4) a fourth process for training the machine learning model using the dataset, and (5) a fifth process for evaluating the trained machine learning model. an input screen generating step of dynamically generating an input screen, which is a user interface for inputting all at once parameters required for automatically executing the process selected in the target identification step by a computer, based on the selection in the target identification step, and accepting input on the generated input screen; Methods for supporting machine learning application systems.

2. further comprising an automatic execution step of automatically executing the processes that are the targets of the scenario in sequence by a computer in accordance with the input received in the input screen generation step. The method for supporting a machine learning application system according to claim 1.

3. The applied processing is a processing for detecting an abnormality in the equipment. The method for supporting a machine learning application system according to claim 1.

4. The equipment operates in a plurality of operation patterns; In the target identification step, a selection from the plurality of movement patterns is accepted, In the input screen generating step, an input screen required for automatically executing the process selected in the target specifying step by a computer is generated as the input screen for the operation pattern selected in the target specifying step. The method for supporting a machine learning application system according to any one of claims 1 to 3.

5. In the input screen generating step, when the first process is selected in the target specifying step, an input screen including an input item for inputting a set deadline, which is the maximum time required to specify measurement data that satisfies a predetermined condition as measurement data to be stored, is generated as the input screen. The method for supporting a machine learning application system according to any one of claims 1 to 3.

6. In the input screen generation step, when the third process is selected in the target identification step, an input screen including an input item for inputting the type of the machine learning model is generated as the input screen. The method for supporting a machine learning application system according to any one of claims 1 to 3.

7. In the input screen generation step, when the third process is selected in the target identification step, an input screen including an input field for inputting the number of times the machine learning model is trained is generated as the input screen. The method for supporting a machine learning application system according to any one of claims 1 to 3.

8. The equipment operates in a plurality of operation patterns; In the input screen generation step, when the third process is selected in the target identification step, an input screen including input items for inputting a behavior pattern to be learned by the machine learning model is generated as the input screen. The method for supporting a machine learning application system according to any one of claims 1 to 3.

9. When the fifth process is executed in the automatic execution step, the machine learning model is evaluated using the measurement data obtained in the past. The method for supporting a machine learning application system according to claim 2.

10. The selection of the scenario includes selection from among a first scenario consisting of the first process, the second process, the third process, the fourth process, and the fifth process, a second scenario consisting of the second process, the third process, and the fourth process, and a third scenario consisting of only the fourth process. The method for supporting a machine learning application system according to any one of claims 1 to 3.

11. The plurality of operation patterns include acceleration in which the operation frequency of the equipment increases, constant speed in which the operation frequency remains substantially unchanged, and deceleration in which the operation frequency decreases, In the target identification step, a selection of either a first movement pattern including the acceleration and the deceleration or a second movement pattern including the constant velocity is accepted. The method for supporting a machine learning application system according to claim 4.

12. In the target identification step, in addition to the first movement pattern and the second movement pattern, a selection of either a third movement pattern including only the acceleration or a fourth movement pattern including only the deceleration is accepted. The method for supporting a machine learning application system according to claim 11.

13. In the target identification step, in addition to the first movement pattern and the second movement pattern, a fifth movement pattern including a constant speed that prioritizes frequency resolution over the constant speed is selected. The method for supporting a machine learning application system according to claim 11.

14. The plurality of operation patterns include acceleration in which the operation frequency of the equipment increases, constant speed in which the operation frequency remains substantially unchanged, and deceleration in which the operation frequency decreases, In the target identification step, features are extracted from measurement data obtained by measuring the machine learning application system, the measurement data is classified based on the extracted features, and a motion pattern including at least one of the acceleration, the constant velocity, and the deceleration is selected based on the classification result. The method for supporting a machine learning application system according to claim 4.

15. A program that causes a computer to execute the steps included in the method for supporting a machine learning application system according to claim 1.

16. An apparatus that uses a computer to support the operation of a machine learning application system that performs application processing using a machine learning model based on measurement data obtained from equipment, a launch unit that receives a launch instruction for an automatic setting wizard that provides the support; a target identification unit that, after the startup instruction is accepted by the startup unit, accepts a selection of a scenario, which is a selection of a process to be automatically executed by a computer, from among (1) a first process that performs settings for acquiring the measurement data, (2) a second process that accumulates the measurement data, (3) a third process that creates a dataset for training the machine learning model from the accumulated measurement data, (4) a fourth process that trains the machine learning model using the dataset, and (5) a fifth process that evaluates the trained machine learning model; an input screen generation unit that dynamically generates an input screen, which is a user interface for inputting parameters required for automatically executing the process selected by the target identification unit on a computer, based on the selection by the target identification unit, and accepts input on the generated input screen; Support device for machine learning application systems.

17. A system that uses a computer to support the operation of a machine learning application system that performs application processing using a machine learning model based on measurement data obtained from equipment, a support device according to claim 16; a terminal device that accesses the assistance device via a communication network, and selects a scenario for the target identification unit included in the assistance device in accordance with an instruction from a user, and inputs information into the input screen generated by the input screen generation unit included in the assistance device; Support system for machine learning application systems.

Images