LEARNING DEVICE, STATE DISPATCH DEVICE, STATE MONITORING SYSTEM AND LEARNING PROCEDURES

Abstract

Learning device (300), comprising: a global model creation unit (304) for creating a first regression model (global model) applicable to training data explainable by a plurality of explanatory variables and a first explanatory variable (x1), wherein the first explanatory variable is externally determined and is one of the plurality of explanatory variables, based on the training data and the first explanatory variable; a variable selection unit (360) for selecting a second explanatory variable (x2) from the plurality of explanatory variables and for selecting a second explanatory variable from the training data, with which target data, which are considered to vary based on the first regression model created by the global model creation unit, are separable from the training data;and a local model creation unit (354) for creating a second regression model (local model) applicable to the training data and the first explanatory variable, using training data after the target data have been separated based on the second explanatory variable selected by the variable selection unit and the first explanatory variable.

Description

TECHNICAL AREA

[0001] The present disclosure relates to a learning device, a state derivation device, a state monitoring system, and a learning method. BACKGROUND ON THE STATE OF THE TECHNOLOGY

[0002] In the manufacturing industry, anomaly detection in equipment (hereinafter also referred to as the "target device"), such as factories and rotary machines, is performed using a machine learning model. Here, a target device anomaly is understood to mean, for example, a deterioration of the target device. Generally, it is more difficult to capture abnormal data from the target device than normal data. Therefore, when training the model, a learning device often performs unsupervised learning, using only normal data collected from the target device as training data. In this case, a derivation device, which infers the state of the target device, uses the learned model to calculate an anomaly indicating how much the target device's state deviates from its normal state.At this point, a threshold value for determining an anomaly is set in the detection device, and the target device is determined to be anomalous if the calculated anomaly exceeds the threshold value. Regarding such anomaly detection method, for example, non-patent literature 1 and non-patent literature 2 describe a method for anomaly detection using a linear regression model and a Gaussian process regression model, respectively. REFERENCE LIST NON-PATENT LITERATURE Non-Patent Literature 1: Tsuyoshi Ide, “Introduction to Anomaly Detection Using Machine Learning,” Corona Publishing Co., Ltd., 2019 Non-Patent Literature 2: Tsuyoshi Ide, “Abnormality Detection and Change Detection,” Corona Publishing Co., Ltd., 2018 SUMMARY OF THE INVENTIONAL PROBLEM

[0003] Non-patent literature 1 and non-patent literature 2 describe a technique for detecting anomalies in a case where there are no deviations from the normal data learned by a learning device. On the other hand, it is less likely that a target device, such as a plant or rotating machine, will continue to operate under constant operating conditions (e.g., a specific operating pattern and motion pattern), and they are often operated under varying conditions. In this case, the normal data collected by the target device can vary depending on the different operating conditions. Note that the operating conditions of the target device are determined by a multitude of control information (parameters), ranging from several dozen to several hundred, such as a current value or a voltage value, or the power required for the target device to operate.

[0004] As described above, in a case where the anomaly detection technique described in Non-Patent 1 and Non-Patent 2 is applied to a case where deviations occur in normal data, it is conceivable that a desired regression model is learned by a learning device (computer) to which the anomaly detection technique is applied. In this case, the normal data in the learning device (hereinafter also referred to as the "conventional device") are grouped such that all patterns are covered, and the inference device performs anomaly detection based on the regression model learned as a result. Here, from a physical point of view, the operating conditions of the target device suitable for anomaly detection are often limited.For example, when detecting anomalies in a rotating machine, the normal values ​​recorded at the time the machine is switched on may contain the influence of electromagnetic interference due to the current. Therefore, in this case, it is desirable for the conventional device to perform the analysis (creation and evaluation of a regression model) by restricting it to normal data recorded at the time the machine is switched off. However, the conventional device has the disadvantage that it is currently difficult to perform the analysis described above, thus increasing the number of steps required to train the regression model.

[0005] The present disclosure was made to solve the above-mentioned problems, and one objective of the present disclosure is to provide a learning device which, compared to the prior art, can reduce the man-hours required for learning when a regression model for detecting an anomaly of a target device is trained using training data with deviations. SOLUTION TO THE PROBLEM

[0006] A learning device according to the present disclosure comprises: a global model creation unit for creating, based on training data that can be explained by a plurality of explanatory variables and a first explanatory variable, which is an externally defined explanatory variable and is one of the plurality of explanatory variables, a first regression model applicable to the training data and the first explanatory variable; a variable selection unit for selecting a second explanatory variable from the plurality of explanatory variables and for selecting from the training data; a second explanatory variable with which target data, which are considered to be variable based on the first regression model created by the global model creation unit, can be separated from the training data;and a local model building unit to create a second regression model applicable to the training data and the first explanatory variable using training data after the target data have been separated based on the second explanatory variable selected by the variable selection unit and the first explanatory variable. ADVANTAGEOUS EFFECTS OF THE INVENTION

[0007] The present disclosure was made to solve the problems mentioned above, and one objective of the present disclosure is to provide a learning device which, compared to the prior art, can reduce the man-hours required for learning when a regression model for detecting an anomaly of a target device is trained using training data with deviations. BRIEF DESCRIPTION OF THE DRAWINGS Fig. Figure 1 is a representation showing a configuration example for a condition monitoring system according to a first embodiment. Fig. Figure 2 is a representation showing a configuration example of a learning device according to the first embodiment. Fig. Figure 3 is a representation showing an example of vibration data and control information data in the first embodiment. Fig. Figure 4 is a representation illustrating a configuration example of a global learning unit (a global model creation unit and a model evaluation unit) in the first embodiment. Fig. Figure 5 is a representation showing an example of an image of a global model in the first embodiment. Fig. Figure 6 is a representation showing an example of an image of a global model in the first embodiment. Fig. 7A, Fig. 7B and Fig. Figure 7C shows a concrete example of classification processing by a filtering processing unit in the first embodiment. Fig. Figure 8 is a representation illustrating a configuration example of a local learning unit (area selection unit, local model creation unit and model evaluation unit) according to the first embodiment. Fig. Figure 9 is a representation to describe a specific example of processing by a distribution calculation unit in the first embodiment. Fig. Figure 10 is a representation showing an example of a representation of the probability distribution generated by a second variable selection processing unit in the first embodiment. Fig. Figure 11 is a representation illustrating an example of a distribution diagram of vibration data generated by a zone evaluation unit according to the first embodiment. Fig. Figure 12 is a representation showing an example of an image illustrating a calculation result of a prediction error generated by an image output unit in the first embodiment. Fig. Figure 13 is a flowchart describing an operational example of the learning device according to the first embodiment. Fig. 14A and Fig. Figure 14B shows an example of a hardware configuration of the learning device according to the first embodiment. Fig. Figure 15 is a representation showing a configuration example of a state derivation device according to the first embodiment. Fig. Figure 16 is a representation illustrating a configuration example of an evaluation unit in the state derivation device according to the first embodiment. Fig. Figure 17 is a representation showing an example of an image illustrating a comparison result produced by the image output unit in the first embodiment. Fig. Figure 18 is a flowchart describing an operating example of the state monitoring device according to the first embodiment. Fig. 19A and Fig. Figure 19B shows an example of a hardware configuration of the state monitoring device according to the first embodiment. Fig. Figure 20A is a representation to describe the difficulty in selecting a suitable regression model in a conventional device, and Fig. Figure 20B is a representation describing the difficulty of evaluating deviations in training data (normal data) in the conventional device. DESCRIPTION OF THE EXECUTION FORMS

[0008] In the following, embodiments of the present disclosure are described in detail with reference to the drawings. First embodiment

[0009] Fig. Figure 1 is a representation showing a configuration example for a condition monitoring system 1000 according to a first embodiment. As, for example, in Fig. As shown in Figure 1, the condition monitoring system 1000 comprises a recording unit 100, a training data recording unit 200, a learning device 300 and a condition derivation device 600.

[0010] The recording unit 100 comprises a recording medium such as a hard disk drive (HDD) and a solid-state drive (SSD). The recording unit 100 records data that specifies a learning model created by the learning device 300.

[0011] The training data recording unit 200 includes a recording medium such as a hard disk drive (HDD) and a solid-state drive (SSD). The training data recording unit 200 records training data that is used by the learning device 300 to create a learned learning model.

[0012] Each learning device 300 and each state monitoring device 600 is configured to be connected to the recording unit 100. Furthermore, the learning device 300 is configured to be connected to the training data recording unit 200.

[0013] The learning device 300 creates a learned model for detecting device anomalies (target device), such as in a plant or rotating machine, using machine learning and the training data recorded in the training data recording unit 200. The learning device 300 then instructs the recording unit 100 to record data that specifies the learned model.

[0014] The state derivation device 600 uses the learned model specified by the data recorded in the recording unit 100 to derive the state of the target device and thereby detect an anomaly (e.g., deterioration) of the target device.

[0015] Furthermore, the condition monitoring system includes 1000 components, such as those found in... Fig. Figure 1 shows a first external evaluation device 400, a second external evaluation device 500 and a third external evaluation device 700.

[0016] The first external evaluation device 400 and the second external evaluation device 500 are configured to be connected to the learning device 300. The first external evaluation device 400 and the second external evaluation device 500 are devices that serve as an interface for the learning device 300, for example, by transmitting instructions from a user to the learning device 300 or by presenting processing content from the learning device 300 to the user.

[0017] The third external evaluation device 700 is configured to be connected to the state monitoring device 600. The third external evaluation device 700 acts as an interface for the state monitoring device 600, for example by transmitting instructions from the user to the state monitoring device 600 and presenting processing content from the state monitoring device 600 to the user.

[0018] For the sake of simplicity, the following description will first describe the details of the learning device 300 and then the details of the state derivation device 600. <Lernvorrichtung 300>

[0019] Fig. Figure 2 is a representation showing a configuration example of the learning device 300 according to the first embodiment. As in Fig. As shown in Figure 2, the learning device 300 comprises a global learning unit 301, a local learning unit 350 and an intermediate recording unit 390.

[0020] The learning device 300 creates a learned model for detecting an anomaly of the target device by performing two-stage machine learning, consisting of learning by the global learning unit 301 and learning by the local learning unit 350, using a target variable described by a variety of arbitrary explanatory variables (hereinafter also referred to as the "explanatory variable group") as training data.

[0021] Specifically, the global learning unit 301 first acquires an initial explanatory variable, selected from the explanatory variable group by a user with knowledge and skills regarding the target device. Furthermore, the global learning unit 301 performs machine learning using the target variable described by the acquired initial explanatory variable as training data and creates a learned model applicable between the training data and the initial explanatory variable. Additionally, the global learning unit 301 receives an evaluation of the learned model from the user via the first external evaluation device 400. Thus, the global learning unit 301 creates a global learning model that has achieved physical validity. The global learning unit 301 then instructs the intermediate recording unit 390 to record data indicating the created learned model.

[0022] Next, the local learning unit 350 uses the learned model, specified by the data recorded by the global learning unit 301 in the buffer unit 390, to classify the training data described above into "data considered to be varying (with large variation)" and "data considered not to be varying (with small variation)." Furthermore, the local learning unit 350 selects a second explanatory variable from the group of explanatory variables described above that is capable of accurately distinguishing "data considered to be varying (with large variation)" from the training data described above and that differs from the first user-selected explanatory variable.Note that “separable” or “separable” here does not mean that “data considered to be variable (with a large variation)” can be completely separated from the training data, but rather that the latter can be roughly separated from the former.

[0023] Subsequently, the local learning unit 350 performs machine learning using the training data after the "data considered to vary (with large variation)" has been separated based on the desired second explanatory variable and the first explanatory variable described above. This creates the local learning model applicable between the training data after the separation and the first explanatory variable. Here, the locally learned model refers to a model learned using training data obtained after separating "data considered to vary (with large variation)" from a physical perspective.

[0024] Here, training data used by the learning device 300 for learning is recorded in the training data recording unit 200. For example, in Fig. As shown in Figure 2, the training data recording unit 200 comprises a vibration database 210 and a control information database 220.

[0025] The vibration sensor database 210 records vibration data. This vibration data includes, for example, data indicating the change in the vibration amplitude value over time, as shown in the graph above. Fig. Figure 3 illustrates this. Note that the vibration data can be data indicating a change over time in the vibration amplitude value. In this case, the vibration amplitude value need only be, for example, the RMS value of the vibration amplitude. Note that the following description presents an example of a case where the vibration data represents the RMS value of the vibration amplitude.

[0026] The control information database 220 stores control information data that represents an explanatory variable. This control information data indicates changes over time in the control information, as shown, for example, in the second to fourth illustrations from the top. Fig. Figure 3 illustrates this. Here, the control information is a parameter that determines an operating condition of the target device, for example, a rotational speed if the target device is a rotating machine, a current value of the drive power of the rotating machine, acceleration / deceleration, or the like.

[0027] Note that each element of the control information data recorded in control information database 220 and the vibration data recorded in vibration database 210 are temporally synchronized. Furthermore, in this case, the vibration data corresponds to an objective variable, and each element of the control information data corresponds to an explanatory variable for explaining the vibration data.

[0028] Note that the following description presents a case where the vibration data of the target variables and the individual control information data correspond to the explanatory variables. However, this is only an example, and the target variable and the explanatory variables can also be different data than those mentioned above. Furthermore, in the following description, the group of explanatory variables is also referred to as the group of control information. <Globale Lerneinheit 301>

[0029] For example, in Fig. As shown in Figure 2, the global learning unit 301 comprises a data extraction unit 302, an explanatory variable procurement unit 303, a global model creation unit 304, and a model evaluation unit 305. (Explanatory variable - Procurement unit 303)

[0030] First, the user selects arbitrary control information from a group of explanatory variables (control information group) to describe the vibration data and inputs the selected control information into the first external evaluation device 400. For the sake of simplicity, it is assumed here that the user selects "rotational speed" as control information. The explanatory variable acquisition unit 303 acquires the control information entered by the user into the first external evaluation device 400 from the first external evaluation device 400 as the first explanatory variable x1. Furthermore, the acquisition unit 303 outputs the acquired data specifying the first explanatory variable x1 as variable descriptor D13 to the data extraction unit 302. (Data Extraction Unit 302)

[0031] Data extraction unit 302 retrieves variable descriptor D13 from explanatory variable acquisition unit 303. Data extraction unit 302 retrieves the control information data corresponding to the captured variable descriptor D13 from control information database 220 in training data recording unit 200. Since variable descriptor D13 specifies "rotational speed" here, data extraction unit 302 retrieves the control information data that is stored in Fig. 3 are in the second position from the top, from the control information database 220.

[0032] Furthermore, the data extraction unit 302 retrieves vibration data from the vibration database 210 in the training data recording unit 200. Subsequently, the data extraction unit 302 outputs the acquired control information data and vibration data as training data D12 to the global model creation unit 304. (Global Modeling Unit 304)

[0033] The global model creation unit 304 learns a regression model applicable to the vibration data and the first explanatory variable x1, based on the vibration data, which can be explained by the multitude of explanatory variables, and the first explanatory variable x1, which is an externally defined explanatory variable and is one of the multitude of explanatory variables. For example, in Fig. As shown in Figure 4, the global model creation unit 304 comprises a model creation unit 311 and a model update unit 312.

[0034] The model creation unit 311 retrieves the training data D12 from the data extraction unit 302. Model creation unit 311 constructs a regression model by performing unsupervised learning on the acquired training data D12. At this point, model creation unit 311 performs unsupervised learning, using the control information data (rotational speed) contained in the training data D12 as the explanatory variable and the vibration data (RMS value of the vibration amplitude value) contained in the training data D12 as the target variable. Note that in this case, a well-known learning method such as linear regression, polynomial regression, or Gaussian process regression is sufficient. Note that in the following description, the regression model created here is also referred to as the "local model."

[0035] Note that the global model is a model that takes the first explanatory variable x1 (control information data) as input and outputs the target variable (vibration data). However, the global model only needs to reproduce a rough regression tendency between the control information data and the vibration data. Therefore, when constructing the global model, the model builder 311 does not necessarily need to use all of the control information and vibration data contained in the training data D12. For example, the model builder 311 can construct the global model using the control information and vibration data corresponding to a user-defined time range.

[0036] The model creation unit 311 outputs data (hereinafter also referred to as "global model data") specifying the constructed global model and the control information data, as well as the vibration data used to learn the global model, as data D18 to the model update unit 312.

[0037] It should be noted that the Model Creation Unit 311 can create a variety of global model patterns. In this case, the Model Creation Unit 311 outputs the global model data for each pattern, the control information data, and the vibration data used for learning as data D18 to the Model Update Unit 312.

[0038] The model update unit 312 retrieves the training data D18 from the model creation unit 311. If the model evaluation unit 355, described later, outputs data D60 from the local training unit 350, the model update unit 312 retrieves the data D60 from the model evaluation unit 355 and updates (reconstructs) the global model according to a user instruction. The processing update in this case is described later.

[0039] When the global model is updated, the model update unit 312 outputs data obtained by combining data specifying the updated global model with the control information data and the vibration data used at the time of the update as data D14 to the model evaluation unit 305. If the model evaluation unit 355 does not output data D60 and the global model is not updated, the model update unit 312 outputs data D18 unchanged as data D14 to the model evaluation unit 305. (Model evaluation unit 305)

[0040] The model evaluation unit 305 receives an evaluation of the global model, which is specified by the data contained in data D14 from external sources (e.g., from the user). For example, in Fig. As shown in Figure 4, the model evaluation unit 305 comprises an image output unit 313 and a model determination unit 314.

[0041] The image output unit 313 retrieves the data D14 from the model update unit 312. Based on the global model data contained in the acquired data D14, the image output unit 313 maps the global model specified by the data and generates data (hereinafter also referred to as "global model image data") that provides an image of the global model. The image output unit 313 outputs the generated global model image data as data D15 to the first external evaluation device 400.

[0042] An example of a representation of the global model is in Fig. 5 shown. For example, in Fig. 5 the reference number 501 a curve (predicted line) that specifies a regression equation obtained by the global model, and the reference number 502 denotes a limit of a confidence interval (for example, a curve 501 ± 5%) that is specified for the curve (predicted line) that specifies the regression equation.

[0043] Note that if the data D14 contains multiple patterns of global model data, the image output unit 313 generates global model image data for each pattern based on each global model data, for example, as in Fig. 6 is shown, and the generated global model image data is output as data D15 to the first external evaluation device 400.

[0044] The first external evaluation device 400 acquires the data D15 from the image output unit 313. Based on the acquired data D15, the first external evaluation device 400 displays one or more images of the overall model on a display unit (not shown), such as a monitor. When an image of the global model is displayed on the display unit, the user examines the image, determines whether the global model is considered physically correct, and, if so, enters a determination result to the first external evaluation device 400 indicating that the global model is considered correct. If multiple images of the global model are displayed on the display unit, the user examines each image, selects a global model considered physically correct, and enters a selection result into the first external evaluation device 400.

[0045] Furthermore, at this point, the user defines a search width S, a range in which a variation in the vibration data is considered relatively small or in which properties of the target device are seen as reflected in the vibration data, from the time domain of the time series of the control information data used for learning (here, the rotational speed), and inputs the range into the first external evaluation device 400. Here, the search width S is a variable that is used when the area selection unit 352 of the local learning unit 350, which will be described later, searches for a zone with low fluctuations in the vibration data.

[0046] The first external evaluation device 400 outputs data obtained by adding data indicating the determination result or the user's selection result and data indicating the search width S entered by the user, as data D16 to the model determination unit 314.

[0047] Note that if there is no global model considered physically correct, the user only needs to perform one of the following two operations, for example. For instance, the user rejects the global model created at this point using the first external evaluation device 400 and inputs control information into the first external evaluation device 400 that differs from the originally input control variable (here, rotational speed). Subsequently, the different control information can be captured by the explanation variable acquisition unit 303 as the new first explanation variable x1, and then the global model creation unit 304 only needs to be instructed to reconstruct the global model through processing similar to that described above.

[0048] Alternatively, the user leaves the first explanatory variable x1 unchanged, excludes data that, based on the image of the global model using the first external evaluation device 400, is considered to deviate from the vibration data, and then instructs the global model creation unit 304 to reconstruct the global model. The user simply needs to repeat one of the above operations until a global model is created that is considered correct from a physical point of view.

[0049] The model determination unit 314 retrieves the training data D16 from the first external evaluation device 400. Based on the acquired data D16, the model determination unit 314 causes the intermediate recording unit 390 to record as data D17 either data indicating a global model that the user has determined to be correct from a physical perspective, or data indicating a global model that the user has selected as the correct model from a physical perspective. The model indicated by data D17 corresponds to the global training model described above.

[0050] Note that at this point, the model unit 314 sets the data specifying the search width S in the data D16 as the range descriptor, incorporates the range descriptor and an identifier (e.g., a name) of the control information data used to learn the global model into the data D17, and stores the data in the intermediate recording unit 390. (Intermediate recording unit 390)

[0051] The intermediate recording unit 390 records the data D17. That is, the intermediate recording unit 390 records data (global model data) that specifies a global model corresponding to a global learned model, an identifier of the control information data, and a range descriptor. <Lokale Lerneinheit 350>

[0052] For example, in Fig. As shown in Figure 2, the local learning unit 350 comprises a second variable selection unit 360, a local model creation unit 354, and a model evaluation unit 355. Furthermore, the second variable selection unit 360 includes, for example, a filtering-processing unit 351, a range selection unit 352, and a second variable selection-processing unit 353. (Filtering-Processing Unit 351)

[0053] The filtering processing unit 351 acquires the data D17 (global model data, the identifier of the control information data and the area descriptor) recorded in the intermediate recording unit 390 as global model descriptor D51.

[0054] Furthermore, the filtering processing unit 351 accesses the training data recording unit 200 and records the vibration data recorded in the vibration database 210 and data corresponding to the identifier of the control information data contained in data D17 among the control information recorded in the control information database 220 as data D52.

[0055] Subsequently, based on the acquired global model descriptor D51 and the data D52, the filtering processing unit 351 classifies (filters) the vibration data contained in the data D52 into data that are considered to vary and data that are considered to be non-varying, and identifies both classified data.

[0056] For example, the filter processing unit 351 determines the degree of variation of the vibration data based on the global model and assigns the label "Data A" to data considered to vary between vibration data sets, and the label "Data B" to data considered not to vary between vibration data sets. The filter processing unit 351 then outputs the data obtained by combining the vibration data sets with the label and the global model descriptor D51 described above as Data D53 to the range selection unit 352.

[0057] A concrete example of classification processing by the filtering processing unit 351 is in Fig. 7 is shown. For example, it shows Fig. 7A shows a distribution diagram of vibration data (RMS value) in a case where the first explanatory variable x1 (rotational velocity) is taken on the horizontal axis and the vibration data on the vertical axis. Furthermore, it shows Fig. 7B is an image of the global model specified by the global model data contained in data D17, which is recorded in the buffer unit 390. Furthermore, Fig. 7C a distribution diagram of the vibration data after classification processing by the filtering processing unit 351.

[0058] For example, the filtration processing unit 351 overlays the Fig. 7A and Fig. 7B together, determined among those in Fig. The vibration data shown in 7A are those vibration data that are outside the range for a curve (predicted line) 701 in the global model of Fig. Data within the specified confidence interval (7B) are considered to be varying and are labelled as "Data A" (bottom representation of Fig. 7C). In addition, the filtering-processing unit 351 considers the vibration data that fall within the confidence interval between the in Fig. The vibration data shown in Figure 7A are considered to be unchanging and are labelled as "Data B" (top representation in Figure 7A). Fig. 7C).

[0059] Note that, for the sake of clarity, in the following description, data considered variable by the Filtering Processing Unit 351 is simply referred to as "Data A," and data considered non-variable is simply referred to as "Data B." This data is also collectively referred to as "labeled data." (Area selection unit 352)

[0060] For example, in Fig. As shown in Figure 8, the area selection unit 352 comprises a distribution calculation unit 361 and a distribution difference comparison unit 362.

[0061] The distribution calculation unit 361 retrieves the data D53 from the filtering processing unit 351. The distribution calculation unit 361 analyzes the distribution of data A and data B based on the tagged data contained in the acquired data D53. In particular, for example, the distribution calculation unit 361 calculates, as in Fig. Figure 9 shows a probability distribution pA of data A and a probability distribution pB of data B for the first explanatory variable x1 and outputs data specifying the calculated probability distribution and data D53 to the distribution difference comparison unit 362.

[0062] Note that the in Fig. The “search width” shown in section 9 indicates the search width described above, which is entered by the user via the first external evaluation device 400. In the example of Fig. In step 9, the search width is set so that the rotation speed is between 500 and 1000. This means that the user has found that the fluctuations in the vibration data are relatively small when the rotation speed is between 500 and 1000.

[0063] The distribution difference comparison unit 362 acquires data specifying the probability distributions pA and pB and the data D53 from the distribution calculation unit 361. The distribution difference comparison unit outputs a difference between the probability distribution pA and the probability distribution pB based on the acquired data specifying the probability distributions pA and pB. At this point, the comparison unit 362 selects, from the search width specified by the range descriptor contained in the data D53, a region where the probability distribution pB is greater than the probability distribution pA, which is a region on the time series of the control information data (rotational speed) with the largest difference.

[0064] Specifically, the distribution difference comparison unit 362 calculates each difference pB-pA based on the probability distributions pA and pB obtained by the distribution calculation unit 361. At this point, assuming the search width is S, the distribution difference comparison unit 362 selects a range Ω in which pB|Ω - pA|Ω (where Ω = [a, a + S] and a is any value of the first explanatory variable x1) is the maximum of the search width S, and sets the selected range Ω as the new range descriptor. Subsequently, the distribution difference comparison unit 362 compares the data obtained by combining the new range descriptor, the labeled data (data A and data B), and the global model descriptor D51 with the second variable selection processing unit 353 as data D54.

[0065] Note that although an example has been described here where the distribution difference comparison unit 362 selects the range Ω in which pB|Ω - pA|Ω is the maximum, the distribution difference comparison unit 362 is not restricted to this and, for example, a range Ω can be selected in which pB|Ω - pA|Ω is equal to or greater than a given value, and the selected range Ω can be used as the new range descriptor. (Second Variable Selection Processing Unit 353)

[0066] The second variable selection processing unit 353 retrieves data D54 from the range selection unit 352. Subsequently, at the time of learning in the global learning unit 301, the second variable selection processing unit 353 selects an explanatory variable (control information) from the group of explanatory variables. This variable differs from the first explanatory variable x1 selected by the user and can precisely separate data A and data B. It should be noted that the explanatory variable selected here will also be referred to as "second explanatory variable x2" in the following description.

[0067] For example, the second variable selection processing unit 353 generates a representation of the probability distribution, as in Fig. 10 shown. In Fig. Figure 10 shows the horizontal axis as a specific explanatory variable that is not the first explanatory variable x1, and indicates a possible explanatory variable for the second explanatory variable x2. Furthermore, the vertical axis indicates the frequency of occurrence of data A and data B that are contained within the area Ω described above, which is contained within data D54.

[0068] At this time, the second variable selection processing unit 353 searches for an explanatory variable using the following procedure, while successively changing the explanatory variables that are candidates for the second explanatory variable x2, and selects a found explanatory variable as the second explanatory variable x2.

[0069] (1) In a representation of the probability distribution, as in Fig. As shown in Figure 10, the second variable selection processing unit 353, for example, if the length of the entire horizontal axis is "100", defines a region on the horizontal axis in which the ratio of data A to Y is equal to or greater than "100 - ε" % (ε being a small positive integer) among all data A (first region). For example, the second variable selection processing unit 353 sets ε = 5 and defines a region on the horizontal axis such that 95% or more of all data A is contained within Y.

[0070] (2) Next, the second variable selection processing unit 353 defines a range excluding range Y as range X (second range) on the horizontal axis of the probability distribution plot and searches for an explanatory variable where the proportion of data B to all data contained in range X is equal to or greater than a predefined value (for example, 80%). The second variable selection processing unit 353 then selects the retrieved explanatory variable as second explanatory variable x2.Note that if the second variable selection processing unit 353 searches for a plurality of explanatory variables where the ratio of data B is equal to or greater than a predefined value among all data contained in the range X, for example, the second variable selection processing unit 353 selects an explanatory variable where the ratio of data B is the largest in the range X, as the second explanatory variable x2.

[0071] Note that if the search fails, the second variable selection processing unit 353 repeats steps (1) and (2) above, successively changing the explanatory variables that qualify as the second explanatory variable x2. Subsequently, the second variable selection processing unit 353 outputs data obtained by combining the second explanatory variable x2 selected by the above procedure with the data D54 above as data D56 to the local model building unit 354.

[0072] The second explanatory variable x2 is a variable that differs from the first explanatory variable x1, which was set by the user at the time of learning in global learning unit 301, and is a variable with a high probability of accurately (precisely) separating data A and data B by combining it with the first explanatory variable x1. (Local Modeling Unit 354)

[0073] For example, in Fig. As shown in Figure 8, the local model creation unit 354 comprises a zone evaluation unit 363 and a model creation unit 364. (Zone assessment unit 363)

[0074] Zone evaluation unit 363 receives data D56 from the second variable selection processing unit 353. Zone evaluation unit 363 generates a distribution diagram of the vibration data, as shown in the example below. Fig. 11 shown, using the second explanatory variable x2 and the first explanatory variable x1 (here the rotational speed) which are contained in the acquired data D56.

[0075] For example, in Fig. As shown in Figure 11, this distribution diagram is a representation in which the first explanatory variable x1 (rotational velocity) is plotted on the horizontal axis and the second explanatory variable x2 on the vertical axis, and the vibration data are displayed in a zone (hereinafter also referred to as the "combination zone") determined by a combination of both variables. It is a representation that depicts a zone in which both data A and data B occur within the combination zone. Note that in Fig. 11. Data A is marked by a grey dot and data B by a black dot.

[0076] As described above, the zone assessment unit 363 can, by displaying data A and data B in the combination zone, define a zone with a relatively large amount of data A and an area with a relatively small amount of data A (in Fig. 11 (identified by reference symbols U1 to U3) are clearly displayed in the combination zone. The zone evaluation unit 363 outputs data specifying the generated distribution diagram as data D62 to the second external evaluation device 500.

[0077] The second external evaluation device 500 acquires the data D62 from the zone evaluation unit 363. Based on the acquired data D62, the second external evaluation device 500 displays an image of the distribution diagram on a display unit (not shown), for example, a display.

[0078] The user refers to the distribution diagram displayed on the display unit, selects a zone with a relatively small amount of data A from among the combination zones, for example zones U1 to U3 in Fig. 11, and inputs the selected zone into the second external evaluation device 500. At this point, the user can select only one zone, for example, zone U1, or a plurality of zones, for example, zones U1 to U3. The second external evaluation device 500 outputs the data specifying the input range as zone range data D63 to the model creation unit 364.

[0079] The model creation unit 364 acquires the zone area data D63 from the second external evaluation device 500. In addition, the model creation unit 364 also acquires the data D56 from the second variable selection processing unit 353. Subsequently, based on the acquired zone area data D63 and the data D56, the model creation unit 364, for example, defines the zones U1 to U3 in Fig. The program defines 11 vibration data points, performs unsupervised learning using the defined vibration data and the corresponding control information data as training data, and creates a regression model. Note that in the following description, the regression model created here is also referred to as the "local model." The local model is a model that receives the first explanatory variable x1 (control information data) as input and outputs the target variable (vibration data).

[0080] Note that if the user selects multiple zones, the model creation unit 364 creates the local model for each of the selected zones. Furthermore, at this point, the model creation unit 364 uses a learning model similar to the one used by the global model creation unit 304 for the global learning unit 301. The two learning models are not necessarily identical.

[0081] The model creation unit 364 outputs data obtained by combining the data (hereinafter also referred to as "local model data") that specify the created local model and the vibration data (data A and data B) used for training, as data D57 to the model evaluation unit 355. (Model evaluation unit 355)

[0082] The model evaluation unit 355 receives an external evaluation (for example, from the user) with respect to the local model, which is specified by the local model data contained in data D57. For example, in Fig. As shown in Figure 8, the model evaluation unit 355 comprises a prediction error calculation unit 365, an image output unit 366 and a model determination unit 367.

[0083] The prediction error calculation unit 365 retrieves the data D57 from the model creation unit 364. The prediction error calculation unit 365 calculates a prediction error of the local model based on the acquired data D57.

[0084] For example, the prediction error calculation unit 365 inputs the first explanation variable x1 (rotational velocity) into the local model, which is specified by the local model data contained in data D57, and calculates how much the vibration data (RMS value) output by the local model at that time differs from the vibration data originally to be output. At this point, the prediction error calculation unit 365 calculates the prediction error with a value such as a mean absolute percentage error (MAPE). The prediction error calculation unit 365 outputs the first explanation variable x1 and the vibration data used to calculate the prediction error, as well as data indicating the calculated prediction error, to the image output unit 366.

[0085] The image output unit 366 acquires the first explanation variable x1 and the vibration data, as well as the data specifying the prediction error, from the prediction error calculation unit 365. Subsequently, based on the acquired data, the image output unit 366 generates data for each zone that specifies an image from which the prediction result is evident, for example, as shown on the right side of Fig. Figure 12 is shown. Subsequently, the image output unit 366 outputs the data obtained by combining the data specifying the generated image and the data specifying the prediction error as data D58 to the second external evaluation device 500. Note that in the illustration on the right side of Fig. 12, similar to the representation of the global model in Fig. 5, a curve (predicted line) is shown that indicates the regression equation obtained by the local model, as well as the limit of the confidence interval (e.g. curve ±5%) that has been set for the curve (predicted line) that indicates the regression equation.

[0086] The second external evaluation device 500 acquires the data D58 from the image output unit 366. The second external evaluation device 500 displays, for example, an image that is on the right side of Fig. As shown in Figure 12, the user displays the acquired data D58 on a display unit (not shown), for example, a display. Referring to the image displayed on the display unit, the user determines a local model from the available local models, which is ultimately to be output to the recording unit 100, and inputs an identifier of the determined local model into the second external evaluation device 500. The second external evaluation device 500 outputs the input identification of the local model as data D59 to the model determination unit 367.

[0087] The model determination unit 367 acquires the data D59 from the second external evaluation device 500. The model determination unit 367 causes the recording unit 100 to record data specifying the local model that will ultimately be output by the user as data D61 based on the acquired data D59.

[0088] Furthermore, at this time, the model determination unit 367 contains data (hereinafter also referred to as "control condition data") that specifies a state (hereinafter also referred to as "control condition") of the control information (explanatory variable) when the user-defined local model is created in the data D61, and causes the recording unit 100 to record the data. Here, the control condition relates, for example, to the type of the first explanatory variable x1 (such as rotational speed), the type of the second explanatory variable x2 (other than rotational speed), the range of the first explanatory variable x1 and the range of the second explanatory variable x2 in which the training data existed at the time the local model was created, and the like.

[0089] Note that if the user specifies multiple local models as output data, the model determination unit 367 causes the recording unit 100 to record multiple local model data. Furthermore, in this case, the model determination unit 367 causes the recording unit 100 to store the control state data for each of the multiple local models in conjunction with each other.

[0090] Note that if the user has referenced the image displayed on the display unit but has not found the local model that is ultimately to be output to the recording unit 100, the user enters this fact, for example, into the second external evaluation device 500. The second external evaluation device 500 outputs data indicating that no local model exists to be output to the model determination unit 367 as data D59.

[0091] After retrieving data D59, the model determination unit 367 acquires data D57 (data obtained by combining the local model data and the vibration data (data A and data B) used to learn the local model) from the prediction error calculation unit 365 and outputs the acquired data D57 as data D60 to the model update unit 312 of the global learning unit 301.

[0092] The model update unit 312 retrieves data D60 from the model determination unit 367. Upon receiving data D60, the model update unit 312 causes the display unit of the first external evaluation device 400 to display the contents of data D60. Furthermore, the model update unit 312 causes the display unit of the first external evaluation device 400 to instruct the user to perform an update process (i.e., a rebuild) of the global model.

[0093] In response to this display, the user selects a different explanatory variable than the first explanatory variable x1, which was originally selected when creating the global model, and enters the selected new explanatory variable into the first external evaluation device 400. Subsequently, the model update unit 312 updates (reconstructs) the global model using a procedure similar to the procedure of the model creation unit 311 described above.

[0094] Note that this is an example where the user reselects a different explanatory variable than the first explanatory variable x1, which was originally selected when creating the global model, and updates (reconstructs) the global model. However, the procedure for updating (reconstructing) the global model is not limited to this. For example, while retaining the first explanatory variable x1 selected at the time of global model creation, the user can change the range of the first explanatory variable x1 from its initial state, for example, by restricting its range. In this case, the model update unit 312 only needs to update (reconstruct) the global model using the training data contained within the changed range.

[0095] Furthermore, for example, the model update unit 312 can instruct the user to re-enter the search width via the first external evaluation device 400 without updating the global model. In this case, the user enters a new search width via the first external evaluation device 400, and the new search width is stored as a range descriptor in the intermediate recording unit 390. Subsequently, the second variable selection processing unit 353 selects a new second explanatory variable x2 based on the new range descriptor, and the local learning unit 350 reconstructs the local model.

[0096] Alternatively, the model update unit 312 can instruct the user to enter the following: Fig. The user can reselect zones U1 to U3 shown in the diagram via the second external evaluation device 500 without updating the global model. In this case, the user enters a new zone via the second external evaluation device 500, and data specifying the new zone is sent as zone range data to the model creation unit 364. The local model is then reconstructed by the model creation unit 364 using the new zone range data.

[0097] Next, a flowchart will be used to illustrate the process. Fig. 13 An example of the operation of the learning device 300 according to the first embodiment is described.

[0098] First, the explanation variable acquisition unit 303 acquires the control information entered by the user into the first external evaluation device 400 from the first external evaluation device 400 as the first explanation variable x1 (step ST1). The explanation variable acquisition unit 303 outputs the acquired data, which specifies the first explanation variable x1, as variable descriptor D13 to the data extraction unit 302.

[0099] Next, the data extraction unit 302 retrieves the control information data corresponding to the acquired variable descriptor D13 from the control information database 220 in the training data storage unit 200. Additionally, the data extraction unit 302 acquires vibration data as training data from the vibration database 210 in the training data recording unit 200 (step ST2).

[0100] Next, the model creation unit 311 creates a global model (step ST3) using the data acquired in step ST2.

[0101] Next, the image output unit 313 generates global model image data and outputs this data to the first external evaluation unit 400 (step ST4). Based on the acquired data, the first external evaluation unit 400 displays an image of the global model on a display unit, such as a monitor, and receives a determination or selection result from the user. The first external evaluation unit 400 then outputs data indicating the user's determination or selection result to the model determination unit 314.

[0102] Next, the model determination unit 314 collects data indicating a determination result or a user selection result and determines whether the result indicates that a global model has been selected (step ST5). If the result shows that no global model has been selected (step ST5; No), the process returns to step ST1, and the explanation variable acquisition unit 303 acquires a new first explanation variable x1 from the user via the first external evaluation device 400. Steps ST2 through ST5 are then repeated.

[0103] If, however, the above result indicates that one of the global models has been selected (step ST5; Yes), the process proceeds to step ST6, and the filtering processing unit 351 classifies (filters) the vibration data into data (data A) that are considered to be varying and data (data B) that are considered to be non-varying (step ST6).

[0104] Next, the distribution calculation unit 361 calculates the probability distribution pA of data A and the probability distribution pB of data B for the first explanatory variable x1. In addition, the distribution difference comparison unit 362 selects a range Ω in which pB|Ω - pA|Ω(Ω = [a, a + S], a is any value of the first explanatory variable x1) is maximized from the search width S (step ST7).

[0105] Next, the second variable selection processing unit 353 selects the second explanatory variable x2, which is a different explanatory variable than the first explanatory variable x1 and can accurately separate data A and data B (step ST8).

[0106] Next, the zone evaluation unit 363 generates a distribution diagram displaying vibration data for a zone defined by a combination of the first explanatory variable x1 and the second explanatory variable x2. Subsequently, the model creation unit 364 receives the user's zone selection based on the distribution diagram (step ST9).

[0107] Next, the model creation unit 364 creates a local model using the vibration data and control information data contained in the zone selected in step ST9 (step ST10).

[0108] Next, the prediction error calculation unit 365 calculates a prediction error of the local model, and the image output unit 366 generates image data indicating a prediction result and outputs the image data to the second external evaluation device 500. The second external evaluation device 500 displays an image indicating a prediction result on the display unit and receives a determination result or a selection result from the user. The second external evaluation device 500 outputs data indicating a determination result or a selection result from the user to the model determination unit 367.

[0109] Next, the model determination unit 367 captures the data indicating a determination result or a user selection result and determines whether the result indicates that a local model was selected (step ST11). If the result indicates that no local model was selected (step ST11; No), the model update unit 312 prompts the user, via the first external evaluation device 400, to select a new first explanatory variable x1. The process then proceeds to step S21, and the explanatory variable acquisition unit 303 acquires a new first explanatory variable x1 from the user via the first external evaluation device 400. Steps ST2 through ST11 are then repeated.

[0110] Note that, although this is shown in the flowchart of Fig. Figure 13 is not shown. In a case where the above result indicates that no local model was selected (step ST11; No), the model update unit 312 can instruct the user, via the first external evaluation device 400, to change the range of the first explanatory variable x1, for example, to narrow the range of the first explanatory variable x1. When the model update unit 312 instructs the user to change the range of the first explanatory variable x1, the process simply needs to return to step ST2.

[0111] If the above result also indicates that no local model was selected (step ST11; No), the model update unit 312 can instruct the user to re-enter the search width or to re-select zones U1 to U3 via the first external evaluation device 400. In a case where the model update unit 312 instructs the user to re-enter the search width, processing returns to step ST7, and in a case where the model update unit instructs the user to re-select zones U1 to U3, the process only needs to return to step ST9.

[0112] If, however, the above result indicates that one of the local models has been selected (step ST11; Yes), the process proceeds to step S32, and the model determination unit 367 instructs the recording unit 100 to record data specifying the selected local model (step ST12). Furthermore, the model determination unit 367 also stores the control condition data in the recording unit 100.

[0113] With the above configuration, the learning device 300 according to the first embodiment can reduce the number of working hours required for learning compared to the prior art when a model for detecting an anomaly of a target device is learned using data with deviations collected from the target device.

[0114] To add to this point: In a conventional device, for example, normal data (training data) is grouped with anomalies in such a way that all patterns are covered, but from a physics perspective, the operating conditions of a target device suitable for anomaly detection are often limited. With the conventional device, however, it is difficult to perform training on the normal data collected under these limited operating conditions, and consequently, the problem arises that the workload for learning the regression model increases.Furthermore, in the conventional device, if a large number of control parameters are available to determine the operating conditions, or if each control parameter has a continuous value, there is an infinite number of state-division procedures, and the computational costs are incurred. Moreover, even with some regression models, such as those found in... Fig. As depicted in 20A, which can be constructed by grouping normal data to cover all patterns, it is difficult to select a suitable regression model from these regression models. Furthermore, the conventional device assumes that the evaluation of the deviation of the normal data depends on which regression model is selected. Therefore, for example, in the conventional device, as in Fig. As shown in Figure 20B, it is also difficult to assess whether the normal data present in the square frame can really be considered as data with deviations.

[0115] In this context, in the learning device 300 according to the first embodiment as described above, the global model creation unit 304 first creates a global learned model (global model) that has achieved physical validity, then the second variable selection unit 360 selects the second explanatory variable x2, which can separate the target data, considered to vary, from the training data, and the local model creation unit 354 forms a regression model (local model) applicable between the training data and the first explanatory variable x1, using the training data after the target data have been separated on the basis of the second explanatory variable x2 and the first explanatory variable x1.As described above, in the Learning Device 300, the selection of the second explanatory variable x2, which can separate the target data considered to differ from the training data, increases the possibility of finding a bounded operating condition for the target device and reduces the workload and computational costs of learning compared to conventional devices. Furthermore, because the target data considered to be variable can be separated from the training data in the Learning Device 300, it is possible to create a regression model with high inferential accuracy and, in addition, to facilitate the selection of a suitable regression model and the evaluation of deviations in the training data, which is difficult with conventional devices.

[0116] Next, we will use the following as an example: Fig. 14 An example of the hardware configuration of the learning device 300 according to the first embodiment is described. The functions of the global learning unit 301 and the local learning unit 350 in the learning device 300 are implemented by a processing circuit. The processing circuit can be dedicated hardware, as shown in Fig. 14A, or a central processing unit (CPU, which may also be called a central processing device, processing device, computing unit, microprocessor, microcomputer, processor or digital signal processor (DSP)) 72, which executes a program stored in a working memory 73, as in Fig. 14B shown.

[0117] In a case where the processing circuit is dedicated hardware, processing circuit 71 corresponds, for example, to a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or a combination thereof. The functions of the respective units of global learning unit 301 and local learning unit 350 can be implemented by processing circuit 71, or the functions of the respective units can be implemented jointly by processing circuit 71.

[0118] If the processing circuit is the CPU 72, the functions of the global learning unit 301 and the local learning unit 350 are implemented by software, firmware, or a combination of software and firmware. The software and firmware are described as programs and stored in the main memory 73. The processing circuit implements the functions of the respective units by reading and executing the programs stored in the main memory 73. That is, the learning device 300 includes a main memory for storing a program that executes each in Fig. The step shown in 13 leads to this, for example, when it is executed by the processing circuit. Furthermore, it can also be said that these programs cause a computer to execute the procedures and methods of the global learning unit 301 and the local learning unit 350. Here, the working memory 73 corresponds, for example, to a non-volatile or volatile semiconductor memory such as random-access memory (RAM), read-only memory (ROM), flash memory, erasable programmable ROM (EPROM) or electrically erasable EPROM (EEPROM), a magnetic disk, a flexible disk, an optical disk, a compact disc, a mini-disc, or a digital versatile disc (DVD).

[0119] Note that some of the functions of the global learning unit 301 and the local learning unit 350 can be implemented by dedicated hardware, and others by software or firmware. For example, the functions of the global learning unit 301 can be implemented by a processing circuit as dedicated hardware, and the functions of the local learning unit 350 can be implemented by the processing circuit by reading and executing a program stored in memory 73.

[0120] As described above, the processing circuit can implement the aforementioned functions through hardware, software, firmware, or a combination thereof. <Zustandsableitvorrichtung 600>

[0121] Next, the state monitoring device 600 according to the first embodiment will be described. Fig. Figure 15 is a diagram showing a configuration example of the state monitoring device 600 according to the first embodiment. As, for example, in Fig. As shown in Figure 15, the state detection device 600 comprises a procurement unit 601, a data selection unit 602, an evaluation unit 603 and a feedback information generation unit 604.

[0122] The state monitoring device 600 detects an anomaly of the target device by deriving the state of the target device from the local model specified by the data (local model data) recorded by the learning device 300 in the recording unit 100. Note that in the following description, the anomaly of the target device is assumed to be a deterioration of the target device. (Procurement unit 601)

[0123] The acquisition unit 601 acquires vibration data A1 from the vibration sensor 50 attached to the target device. The vibration data A1 is data indicating a change over time in the vibration amplitude value of the target device, as detected by the vibration sensor 50 attached to the target device. Note that the vibration data A1 can be data indicating a change over time in the feature set of the vibration amplitude value. In this case, the feature value of the vibration amplitude need only be, for example, the RMS value of the vibration amplitude. Note that the following description provides an example of a case in which the vibration data represents the RMS value of the vibration amplitude.

[0124] Furthermore, procurement unit 601 acquires parts of the control information data B1 to Bn from a control information recording device 60. Here, as in the second to fourth illustrations from above in the above-described Fig. As shown in Figure 3, the control information B1 to Bn is data that specifies the temporal change of the control information and is synchronized with the vibration data. Furthermore, n is the number of control information points. Here, the control information is a parameter for determining an operating condition of the target device and, for example, a rotational speed, a current value of the drive power of a rotating machine, or the like, if the target device is the rotating machine. Note that the control information recording device 60 is a special device for recording the control information data B1 to Bn.

[0125] The procurement unit 601 outputs the data obtained by acquiring the recorded vibration data A1 and the control information data B1 to Bn as data D1 to the data selection unit 602. (Data Selection Unit 602)

[0126] The data selection unit 602 accesses the recording unit 100 and retrieves portions of the local model data MA1 to MAn and portions of the control condition data MB1 to MBn from the recording unit 100. Here, n is the number of local models, and the local model data and the control condition data correspond one-to-one. Note that for the sake of simplicity, n = 1 is assumed here.

[0127] The data selection unit 602 extracts from the vibration data and control information data contained in the data D1 described above vibration data and control information data that satisfy a control condition specified by the acquired control condition data MB1, and outputs data in which the extracted data, the local model data MA1 and the control condition data MB1 are collected, as data D2 to the evaluation unit 603. (Assessment unit 603)

[0128] For example, the assessment unit comprises 603, as in Fig. Figure 16 shows a deterioration rate calculation unit 631, a parameter adjustment unit 632 and an image output unit 633. (Deterioration degree calculation unit 631)

[0129] The deterioration rate calculation unit 631 retrieves the data D2 from the range selection unit 602. The deterioration rate calculation unit 631 analyzes the acquired data D2 to calculate the deterioration rate of the target device. Specifically, for example, the deterioration rate calculation unit 631 inputs an arbitrary value (e.g., rotational speed = 500) from the control information data B1 contained in the data D2 into the local model, which is specified by the local model data MA1 contained in the data D2. Based on the input value, the local model outputs vibration data (e.g., RMS value = 1.5) according to the input value.

[0130] The Deterioration Rate Computing Unit 631 then compares the vibration data output by the local model with the vibration data corresponding to arbitrary values ​​contained in the data D2 and calculates the difference between them. The Deterioration Rate Computing Unit 631 then calculates the deterioration of the target device by comparing the calculated error to a predefined threshold. Note that the Deterioration Rate Computing Unit 631 can calculate the deterioration using, for example, a mean absolute percentage error (MAPE), T2 hoteling, or similar methods. The Deterioration Rate Computing Unit 631 outputs data specifying the calculated deterioration as a state descriptor to the Image Output Unit 633. (Image output unit 633)

[0131] The image output unit 633 retrieves the condition descriptor from the deterioration calculation unit 631. The image output unit 633 retrieves the data D2 from the data selection unit 602. Subsequently, the image output unit 633 generates data using the acquired data, for example, a comparison image as shown in Fig. 17 shown. Fig. Figure 17 shows an image on the left illustrating a distribution of vibration data (training data) used to create the local model, and an image on the right illustrating a distribution of vibration data obtained by actually inputting the control information data B1, acquired before the target device, into the local model. This image allows the user to easily see how much the distribution of vibration data obtained by inputting the control information data B1, actually acquired from the target device, into the local model differs from the distribution of vibration data (training data) used to create the local model. The image output unit 633 outputs data indicating the generated comparison image to the third external evaluation device 700.

[0132] The third external evaluation device 700 acquires data that displays the comparison image from the image output unit 633. Based on the acquired data, the third external evaluation device 700 displays a comparison image, as shown in Fig. 17 is shown on a display unit (not shown). The user checks the comparison image displayed on the display unit and, if necessary, makes a parameter adjustment using the third external evaluation device 700.

[0133] For example, as in Fig. Figure 17 illustrates that a minor deviation between the two distributions can occur due to a random cause. In this case, the degree of wear can change due to a minor deviation caused by a random cause unrelated to the deterioration. For example, in the example of Fig. 17, although the targeting device is not actually that badly degraded, calculates a “degradation rate of 18%”. The user therefore makes a parameter adjustment to compensate for this slight difference.

[0134] For example, the user adjusts the position of a predicted line 1701, which is located on the left side of Fig. The local model shown in Figure 17 was obtained, and the position of line 1702 indicates the boundary of the confidence interval set for the predicted line 1701. In this case, the user, for example, visually checks the difference between the left and right distribution plots in Figure 1702. Fig. 17 and adjusts the position of each line or calculates the difference between the average values ​​of the left and right vibration data and adjusts the position of each line.

[0135] Alternatively, the user can adjust the distance between the predicted line 1701, which is drawn through the line on the left side of Fig. The local model shown in Figure 17 was obtained, and the line 1702, which indicates the boundary of the confidence interval set for the predicted line 1701, is adjusted. In this case as well, the user adjusts the interval, for example, by changing the ratio of the deviations of the vibration data in the left and right distribution plots of Fig. 17 visually inspected or magnified a standard deviation value of the left and right vibration data.

[0136] The third external evaluation device 700 outputs data specifying the user-entered fitting unit as fitting descriptor D4 to the parameter fitting unit 632. Fitting descriptor D4 includes regression model data to be fitted, control condition data, data required for model fitting (in particular, a correction value for a regression coefficient), and the like. Specifically, data required for model fitting are also referred to as parameter fitters. Furthermore, fitting descriptor D4 also includes a user-entered determiner that decides whether or not to output fitting descriptor D4 to the feedback information generation unit 604.For example, if the determining device is 1, this means that the adaptation descriptor D4 is output to the feedback information generation unit 604, and if the determining device is 0, this means that the adaptation descriptor D4 is not output to the feedback information generation unit 604.

[0137] Note that the parameter adjuster in the fitting descriptor D4 is entered by the user, for example, in a case where the user determines the parameter adjuster by visually observing the left and right distribution diagrams in Fig. 17 adjusts, but in other cases (for example, in a case where a difference between the average values ​​of the left and right vibration data is calculated to adjust the position of each line), the parameter adjustment unit 632 can, for example, automatically calculate the parameter adjuster, so that the parameter adjuster does not necessarily have to be entered by the user. (Parameter adjustment unit 632)

[0138] The parameter adjustment unit 632 retrieves the adjustment descriptor D4 from the third external evaluation device 700. The parameter adjustment unit 632 outputs the acquired adjustment descriptor D4 to the deterioration rate calculation unit 631 and instructs the deterioration rate calculation unit 631 to adjust the local model based on the adjustment descriptor D4. In response to this instruction, the deterioration rate calculation unit 631 adjusts the local model and recalculates the deterioration rate using the adjusted local model, following the procedure described above. Furthermore, the deterioration rate calculation unit 631 outputs data indicating the recalculated deterioration rate as a state descriptor to the image output unit 633.The image output unit 633, the third external evaluation device 700 and the parameter adjustment unit 632 then repeat the processing described above.

[0139] Note that when the parameter adjustment unit 632 finishes acquiring the adjustment descriptor D4 from the third external evaluation device 700 in the above iteration, the parameter adjustment unit instructs the image output unit 633 to display a final calculation result of the degree of deterioration on the display unit of the third external evaluation device 700.

[0140] Furthermore, the parameter adjustment unit 632 checks the content of the determination device contained in the acquired adjustment descriptor D4. If the content of the determination device indicates that the adjustment descriptor D4 is output to the feedback information generation unit 604, the parameter adjustment unit 632 outputs the adjustment descriptor D4 as data D5 to the feedback information generation unit 604. Conversely, the parameter adjustment unit 632 does not output the adjustment descriptor D4 to the feedback information generation unit 604 if the content of the determination device indicates that the adjustment descriptor D4 should not be output to the feedback information generation unit 604. (Feedback Information Generation Unit 604)

[0141] The feedback information generation unit 604 retrieves data D5 from the parameter fitting unit 632. Based on the acquired data D5, feedback information generation unit 604 generates feedback information D6 and instructs recording unit 100 to record the generated feedback information D6. Feedback information D6 includes regression model data to be fitted, control condition data, data required for model fitting (particularly a correction value for a regression coefficient), and the like, almost analogous to the fitting descriptor D4. Note that feedback information generation unit 604 instructs recording unit 100 to record feedback information D6 as distinct from the local model data MA1 and control condition data MB1 already recorded in recording unit 100.

[0142] The user can then incorporate the feedback information D6 recorded in recording unit 100 into the local model data MA1 and the control condition data MB1. This updates the local model data MA1 and the control condition data MB1 based on the feedback information D6, thereby reducing the possibility of a capture error caused by a difference (for example, a difference due to the random cause described above) that is unknown at the time of creating the second regression model between the data actually captured by the target device and the training data used in creating the second regression model.

[0143] Next, a flowchart will be used to illustrate the process. Fig. 18 describes an example of the operation of the state monitoring device 600 according to the first embodiment.

[0144] First, the acquisition unit 601 receives vibration data from the vibration sensor 50, which is attached to the target device. In addition, the acquisition unit 601 also acquires control information data from the control information recording device 60 (step ST21).

[0145] Next, the data selection unit 602 retrieves the local model data and the control condition data from the recording unit 100 and extracts from the vibration data and control condition data retrieved in step ST21 the vibration data and control condition data that satisfy the control conditions specified by the retrieved control condition data (step ST22).

[0146] Next, the deterioration calculation unit 631 calculates the deterioration level of the target device based on the data extracted in step ST22 (step ST23).

[0147] The image output unit then generates 633 data points that display a comparison image, such as in Fig. 17 is shown by using the calculation result in step ST23 (step ST24). The image output unit 633 outputs data specifying the generated comparison image to the third external evaluation device 700.

[0148] Next, the parameter fitting unit 632 determines whether the fitting descriptor D4 was acquired from the third external evaluation device 700 (step ST25). If the parameter fitting unit 632 determines that the fitting descriptor D4 was acquired from the third external evaluation device 700 (step ST25; Yes), the parameter fitting unit 632 outputs the fitting descriptor D4 to the deterioration rate calculation unit 631 and instructs the deterioration rate calculation unit 631 to fit the local model based on the fitting descriptor D4. The deterioration rate calculation unit 631 fits the local model based on the fitting descriptor D4 (step ST26). Then the process returns to step ST23.

[0149] However, if the parameter adjustment unit 632 determines that the adjustment descriptor D4 was not acquired by the third external evaluation device 700 (step ST25; No), the process continues with step ST26.

[0150] In step ST26, the parameter adjustment unit 632 determines whether the adjustment descriptor D4 has been acquired at least once by the third external evaluation device 700 (step ST26). If the parameter adjustment unit 632 determines that the adjustment descriptor D4 has not been acquired at least once by the third external evaluation device 700 (step ST26; No), the process proceeds to step ST29.

[0151] If, however, the parameter adjustment unit 632 determines that the adjustment descriptor D4 has been detected at least once by the third external evaluation device 700 (step ST26; Yes), the contents of the determination device contained in the last detected adjustment descriptor D4 are checked. If the contents of the determination device then indicate that the adjustment descriptor D4 is output to the feedback information generation unit 604, the last detected adjustment descriptor D4 is output as data D5 to the feedback information generation unit 604.

[0152] The feedback information generation unit 604 generates the feedback information D6 based on the data D5 acquired by the parameter adjustment unit 632 (step ST27) and causes the recording unit 100 to record the generated feedback information D6 (step ST28). Processing then continues to step ST29.

[0153] In step ST29, the image output unit 633 generates data that indicates a final calculation result of the deterioration level of the target device and outputs the generated data to the third external evaluation device 700 to display the final calculation result of the deterioration level of the target device on the display unit (step ST29).

[0154] Next, we will use the following as an example: Fig. 19 An example of the hardware configuration of the state monitoring device 600 according to the first embodiment is described. The functions of the procurement unit 601, the data selection unit 602, the evaluation unit 603, and the feedback information generation unit 604 in the state monitoring device 600 are implemented by a processing circuit. The processing circuit can be dedicated hardware, as shown in Fig. 19A, or a central processing unit (CPU, which may also be called a central processing device, processing unit, arithmetic device, microprocessor, microcomputer, processor or digital signal processor (DSP)) 82, which executes a program stored in a working memory 83, as in Fig. 19B shown.

[0155] In a case where the processing unit is dedicated hardware, a processing circuit 81 corresponds, for example, to a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or a combination thereof. The functions of the acquisition unit 601, the data selection unit 602, the evaluation unit 603, and the feedback information generation unit 604 can be implemented by the processing circuit 81, or the functions of the respective units can be implemented jointly by the processing circuit 81.

[0156] If the processing circuit is the CPU 82, the functions of the procurement unit 601, the data selection unit 602, the evaluation unit 603, and the feedback information generation unit 604 are implemented by software, firmware, or a combination of software and firmware. The software and firmware are described as programs and stored in the main memory 83. The processing circuit implements the functions of the respective units by reading and executing the programs stored in the main memory 83. The state-reading device 600 includes a main memory for storing a program that executes each in Fig.The step shown in Figure 18 leads to this, for example, when it is executed by the processing circuit. Furthermore, it can also be said that these programs cause a computer to execute the procedures and processes of the procurement unit 601, the data selection unit 602, the evaluation unit 603, and the feedback information generation unit 604. Here, the working memory 83 corresponds, for example, to a non-volatile or volatile semiconductor memory such as random-access memory (RAM), read-only memory (ROM), flash memory, erasable programmable ROM (EPROM), or electrically erasable EPROM (EEPROM), a magnetic disk, a flexible disk, an optical disk, a compact disc, a mini-disc, or a digital versatile disc (DVD).

[0157] Note that some of the functions of the Procurement Unit 601, the Data Selection Unit 602, the Evaluation Unit 603, and the Feedback Information Generation Unit 604 may be implemented by dedicated hardware, and some may be implemented by software or firmware. For example, the functions of the Procurement Unit 601 may be implemented by a processing circuit as dedicated hardware, and the functions of the Data Selection Unit 602, the Evaluation Unit 603, and the Feedback Information Generation Unit 604 may be implemented by the processing circuit that reads and executes programs stored in memory 83.

[0158] As described above, the processing circuit can implement the aforementioned functions through hardware, software, firmware, or a combination thereof.

[0159] As described above, the learning device 300 according to the first embodiment comprises: the global model creation unit 304 for creating, based on training data explainable by a plurality of explanatory variables and a first explanatory variable x1, which is an externally defined explanatory variable and is one of the plurality of explanatory variables, the first regression model (global model) applicable to the training data and the first explanatory variable x1; the second variable selection unit 360 for selecting a second explanatory variable x2 from the plurality of explanatory variables and for selecting the second explanatory variable x2 from the training data, with the target data, which are considered to vary based on the first regression model created by the global model creation unit 304, being separable from the training data;and the local model creation unit 354 to construct a second regression model (local model) applicable to the training data and the first explanatory variable x1, using training data after the target data have been separated based on the second explanatory variable x2, which was selected by the second variable selection unit 360, and the first explanatory variable x1. Thus, according to the first embodiment, the learning device 300 can reduce the number of working hours required for learning compared to the prior art when a model for detecting an anomaly of a target device is learned using data with deviations detected by the target device.

[0160] Furthermore, the second variable selection unit 360 also includes: the filtering processing unit 351 to classify the training data into target data (Data A), which are considered to vary, and non-target data (Data B), which are considered to not vary, based on the first regression model created by the global model creation unit 304; the range selection unit 352 to select a predefined range from the ranges that can be occupied by the first explanatory variable x1 based on the target data and the non-target data classified by the filtering processing unit 351; and the second variable selection processing unit 353 to select the second explanatory variable x2 using training data contained within the predefined range selected by the range selection unit 352.Thus, according to the first embodiment, the learning device 300 can select the second explanatory variable x2 in a suitable manner based on the first regression model and the training data.

[0161] Furthermore, the filtering processing unit 351 defines as target training data that lies outside a predefined confidence interval centered on a predicted line and defined for the predicted line obtained based on the first regression model created by the global model creation unit 304, and defines as non-target training data that lies within the predefined confidence interval centered on the predicted line. Thus, according to the first embodiment, the learning device 300 can easily classify the training data into target data (data A), which are considered to vary, and non-target data (data B), which are considered to not vary.

[0162] Furthermore, the range selection unit 352 comprises: the distribution calculation unit 361, to calculate for each of the target data and non-target data a probability distribution indicating how frequently the target data and non-target data classified by the filtering processing unit 351 occur with respect to the first explanatory variable x1; and the distribution difference comparison unit 362, to calculate a difference between the probability distribution of the target data calculated by the distribution calculation unit 361 and the probability distribution of the non-target data calculated by the distribution calculation unit 361, and to select a range of the first explanatory variable x1 in which the calculated difference is equal to or greater than a predefined value, as the predefined range.Thus, according to the first embodiment, the learning device 300 can easily select the specified range of the first explanatory variable x1, which is used to select the second explanatory variable x2.

[0163] Furthermore, the distribution difference comparison unit 362 selects the predetermined range from the externally received search width S, where the search width S specifies a range in which the proportion of the presence of non-target data in the range of the first explanatory variable x1 is assumed to be relatively high. Thus, according to the first embodiment, the learning device 300 can appropriately select the predetermined range of the first explanatory variable x1 based on the externally received search width S.

[0164] Furthermore, the second variable selection processing unit 353 generates a probability distribution that indicates how frequently the training data contained in the predefined range selected by the range selection unit 352 occur with respect to a specific explanatory variable, and if a range of the first explanatory variable x1, in which a ratio of the target data to the number of training data in the generated probability distribution is equal to or greater than a predefined value, is designated as the first range Y, and a range of the first explanatory variable x1 excluding the first range Y is designated as the second range X, it selects an explanatory variable x2 in which a ratio of the non-target data to the training data contained in the second range X is equal to or greater than a predefined value.Thus, according to the first embodiment, the learning device 300 can select the second explanatory variable x2 appropriately and efficiently.

[0165] Furthermore, the local model creation unit 354 comprises: the zone evaluation unit 363 for generating data that displays an image indicating a region where the target data occurred and a zone where the non-target data occurred, in a zone determined by a combination of the second explanatory variable x2, selected by the second variable selection unit 360, and the first explanatory variable x1; and the model creation unit 364 for receiving a zone from outside, based on the image displayed by the data generated by the zone evaluation unit 363, in a zone determined by a combination of the first explanatory variable x1 and the second explanatory variable x2, and for creating the second regression model using training data contained in the received zone.Thus, according to the first embodiment, the learning device 300 can create the second regression model, which reflects the intention of the outside world (for example, the user).

[0166] Furthermore, the learning device 300 comprises: the model evaluation unit 305 for receiving an external evaluation for a first regression model created by the global model creation unit 304; and the model evaluation unit 355 for receiving an external evaluation for a second regression model created by the local model creation unit 354. Thus, according to the first embodiment, the learning device 300 can receive an external evaluation (for example, from the user) for both the first and second regression models.

[0167] Furthermore, the global model creation unit 304 includes the model update unit 312 to generate, if the evaluation received from the model evaluation unit 355 indicates that no desired second regression model exists, a first regression model applicable to the training data and a new first explanatory variable x1 based on the training data. This new first explanatory variable x1 is externally determined and is one of several explanatory variables. Thus, if the desired second regression model cannot be created, the learning device 300, according to the first embodiment, can reconstruct the first regression model.

[0168] Furthermore, according to the first embodiment, the state derivation device 600 derives a state of a target device using a second regression model created by the local model creation unit 354 of the learning device 300, and using data corresponding to training data and data corresponding to a first explanatory variable x1 acquired from the target device, wherein the learning device 300 comprises: the global model creation unit 304 for creating, based on the training data explainable by several explanatory variables and the first explanatory variable x1, which is an externally defined explanatory variable and is one of the several explanatory variables, the first regression model (global model) applicable to the training data and the first explanatory variable x1;The second variable selection unit 360 is used to select a second explanatory variable x2 from the multitude of explanatory variables and to select the second explanatory variable x2 from the training data, with the target data, which are considered to vary based on the first regression model created by the global model creation unit 304, being separable from the training data; and the local model creation unit 354 is used to construct a second regression model (local model) using training data, after the target data have been separated based on the second explanatory variable x2 selected by the second variable selection unit 360 and the first explanatory variable x1, which is applicable to the training data and the first explanatory variable x1. Thus, according to the first embodiment, the state derivation device 600 can accurately derive the state of the target device.

[0169] Furthermore, the state monitoring device 600 includes the feedback information generation unit 604 to correct a regression coefficient in the second regression model based on a correction value received externally. Thus, according to the first embodiment, the state monitoring device 600 can reduce the possibility of a detection error caused by a difference, unknown at the time of design, between the data actually detected by the target device and the training data used in creating the second regression model.

[0170] Furthermore, the condition monitoring system 1000 according to the first embodiment comprises: the learning device 300, which includes: the global model creation unit 304 for creating a first regression model (global model) applicable to the training data and the first explanatory variable x1, based on training data explainable by a plurality of explanatory variables and a first explanatory variable x1, which is an externally defined explanatory variable and is one of the plurality of explanatory variables; the second variable selection unit 360 for selecting a second explanatory variable x2 from the plurality of explanatory variables and for selecting a second explanatory variable x2 from the training data, with the target data, which are considered to vary based on the first regression model created by the global model creation unit 304, being separable from the training data;and the local model creation unit 354 to create a second regression model (local model) applicable between the training data and the first explanatory variable x1, using training data after the target data have been separated on the basis of the second explanatory variable x2 selected by the second variable selection unit 360;and the state derivation device 600 to derive a state of a target device using the second regression model created by the local model creation unit 354 and data corresponding to the training data and data corresponding to the first explanatory variable x1 acquired from the target device. Thus, if a model for detecting an anomaly of a target device is learned from data collected from the target device and exhibiting fluctuations, the state monitoring system 1000 according to the first embodiment can reduce the number of man-hours required for learning compared to the prior art and accurately derive the state of the target device from the model.

[0171] Finally, a preferred application example for the learning device 300 and the state monitoring device 600 according to the first embodiment is described. The learning device 300 according to the first embodiment is, for example, suitable for use in a monitoring system for an electric motor mounted on a rail vehicle. The electric motor mounted on the rail vehicle contains a wealth of control information, such as braking information, rotational speed information, current information, and voltage information, along with vibration data.When creating a system that monitors vibration data reflecting the electric motor's condition based on control information, the global model creation unit 304 first creates a model (global model) using the rotational speed as an explanatory variable to leverage user knowledge (for example, it has been established that vibration and rotational speed are closely related and that high-frequency vibrations are easily observed at low speeds). To improve the model's accuracy, the model (local model) is then created by the user defining the rotational speed range and narrowing down the conditions using other control information, which allows for the exclusion of data that deviates from the global model.In this way, the monitoring system makes it possible to use the user's knowledge for model creation, to reduce the processes of model creation and evaluation by using the explanatory variables and conditions deemed unnecessary for deterioration detection, and to construct the model efficiently.

[0172] Furthermore, the state monitoring device 600 according to the first embodiment, similar to the learning device 300, is also suitable for use, for example, in a monitoring system for an electric motor mounted on a rail vehicle. For example, the monitoring device 600 according to the first embodiment includes an alarm device, and an alarm is issued to the user of the monitoring system when the state monitoring device 600, based on vibration data acquired by the vibration sensor 50 mounted on the target device, determines that the target device is not the same object as the object defined as the monitoring target. In this way, the state monitoring device 600 according to the first embodiment is applicable to a monitoring system.

[0173] Furthermore, the condition monitoring system 1000 according to the first embodiment is suitable, similar to the learning device 300 and the condition derivation device 600, for example, for use in a monitoring system of an electric motor attached to a rail vehicle.

[0174] It should be noted that in the present disclosure, any component of the embodiment can be modified or omitted. For example, the above description describes the case where the training data, as the target variable, is vibration data and the explanatory variable describing the target variable is control information data. However, the training data and the explanatory variable as the target variable are not limited to the above example, and any type of data can be used as long as the explanatory variable explains the target variable.

[0175] Furthermore, the above description provided an example where the recording unit 100 is provided separately from the learning device 300 and the state monitoring device 600. However, the recording unit 100 is not limited to this and can, for example, be integrated into the learning device 300 or the state monitoring device 600.

[0176] Alternatively, the recording unit 100 can be integrated into one of the first external evaluation device 400, the second external evaluation device 500, and the third external evaluation device 700.

[0177] Furthermore, the above description provided an example in which the first external evaluation device 400, the second external evaluation device 500, and the third external evaluation device 700 are provided separately. However, the individual devices are not limited to this, and the functions of the respective devices can be integrated into any device, or the functions of any two devices can be integrated into one device. REFERENCE MARK LIST

[0178] 50: Vibration sensor, 60: Control information recording device, 71: Processing circuit, 72: CPU, 73: RAM, 81: Processing circuit, 82: CPU, 83: RAM, 100: Recording unit, 200: Training data recording unit, 300: Learning device, 301: Global learning unit, 302: Data extraction unit, 303: Explanatory variable acquisition unit, 304: Global model creation unit, 305: Model evaluation unit (first model evaluation unit), 311: Model creation unit, 312: Model update unit, 313: Image output unit, 314: Model determination unit, 350: Local learning unit, 351: Filtering processing unit, 352: Range selection unit, 353: Second Variable selection processing unit (Variable Selection Processing Unit), 354: Local model creation unit, 355: Model evaluation unit (Second Model Evaluation Unit), 360: Second variable selection unit (Variable Selection Unit), 361: Distribution calculation unit,362: Distribution Difference Comparison Unit, 363: Zone Evaluation Unit, 364: Model Creation Unit, 365: Prediction Error Calculation Unit, 366: Image Output Unit, 367: Model Determination Unit, 390: Intermediate Recording Unit, 400: First External Evaluation Device, 500: Second External Evaluation Device, 501: Predicted Line, 502: Line Indicating the Boundary of the Confidence Interval, 600: State Derivation Device, 601: Procurement Unit, 602: Data Selection Unit, 603: Evaluation Unit, 604: Feedback Information Generation Unit, 631: Deterioration Rate Calculation Unit, 632: Parameter Adjustment Unit, 633: Image Output Unit, 700: Third External Evaluation Device, 1000: Condition Monitoring System, 1701: predicted line, 1702: line indicating the limit of the confidence interval, A1: vibration data, B1: control information data, 210: vibration database, 220: control information database, U1: zone, U2: zone, U3: zone, QUOTES INCLUDED IN THE DESCRIPTION

[0000] This list of documents cited by the applicant was automatically generated and is included solely for the reader's convenience. The list is not part of the German patent or utility model application. The DPMA accepts no liability for any errors or omissions. Cited non-patent literature

[0000] Tsuyoshi Ide, “Introduction to Anomaly Detection Using Machine Learning”, Corona Publishing Co., Ltd., 2019

[0002] Tsuyoshi Ide, “Abnormality Detection and Change Detection”, Corona Publishing Co., Ltd., 2018

[0002]

Claims

[1] Learning device, encompassing: a global model creation unit for creating, based on training data that can be explained by a multitude of explanatory variables, and a first explanatory variable that is an externally defined explanatory variable and is one of the multitude of explanatory variables, a first regression model that is applicable to the training data and the first explanatory variable; a variable selection unit for selecting a second explanatory variable from the multitude of explanatory variables and for selecting a second explanatory variable from the training data, with which target data, which are considered to vary based on the first regression model created by the global model creation unit, are separable from the training data; and a global model creation unit to create a second regression model applicable to the training data and the first explanatory variable, using training data after the target data have been separated based on the second explanatory variable selected by the variable selection unit and the first explanatory variable. [2] Learning device according to claim 1, wherein the variable selection unit comprises: a filtering processing unit to classify the training data into target data, which are considered to vary, and non-target data, which are considered to not vary, based on the initial regression model created by the global model creation unit, a range selection unit for selecting a predefined range from ranges that can be assumed by the first explanatory variable based on the target data and the non-target data classified by the filtering processing unit, and A variable selection unit for selecting the second explanatory variable using training data contained within the predefined range selected by the range selection unit. [3] Learning device according to claim 2, wherein the filtering processing unit defines as target data training data that lies outside a predefined confidence interval centered on a predicted line and defined for the predicted line obtained based on the first regression model created by the global model creation unit, and defines as non-target data training data that lies within the predefined confidence interval centered on the predicted line. [4] Learning device according to claim 2 or 3, wherein the area selection unit comprises: a distribution calculation unit to calculate, for each of the target data and non-target data, a probability distribution indicating how frequently the target data and non-target data classified by the filtering processing unit occur with respect to the first explanatory variable, and a distribution difference comparison unit to calculate a difference between the probability distribution of the target data calculated by the distribution calculation unit and the probability distribution of the non-target data calculated by the distribution calculation unit, and to select a range of the first explanatory variable in which the calculated difference is equal to or greater than a predefined value than the predefined range. [5] Learning device according to claim 4, wherein the distribution difference comparison unit selects the specified range from an externally received search width, where the search width specifies a range in which the proportion of the presence of the non-target data in the range of the first explanatory variable is assumed to be relatively high. [6] Learning device according to any one of claims 2 to 5, wherein the variable selection processing unit generates a probability distribution that indicates how frequently the training data contained in the predefined range selected by the range selection unit occur with respect to a specific explanatory variable, and if a range of the first explanatory variable in which a ratio of the target data to a number of training data in the generated probability distribution is equal to or greater than a predefined value is designated as the first range, and a range of the first explanatory variable excluding the first range is designated as the second range, selects an explanatory variable where the ratio of the non-target data to the training data contained in the second area is equal to or greater than a predefined value, as the second explanatory variable. [7] Learning device according to any one of claims 1 to 6, wherein the local model creation unit a zone evaluation unit for generating data that displays an image indicating a zone in which the target data, considered to be varying, occurred, and a zone in which non-target data, considered to be not varying, occurred, in a zone determined by a combination of the second explanatory variable selected by the variable selection unit and the first explanatory variable, and a model creation unit for receiving a zone, determined from the outside based on an image displayed by the data generated by the zone evaluation unit, in a zone determined by a combination of the first explanatory variable and the second explanatory variable, and for creating the second regression model using training data contained in the received zone. [8] Learning device according to any one of claims 1 to 7, further comprising: a first model evaluation unit that receives an evaluation from an external source for the first regression model created by the global model creation unit; and a second model evaluation unit that receives an external evaluation for the second regression model created by the local model creation unit. [9] Learning device according to claim 8, wherein the global model creation unit comprises: a model update unit to reconstruct, if the evaluation received from the second model evaluation unit indicates that a desired second regression model is not available, a first regression model applicable to the training data and a new first explanatory variable based on the training data and the new first explanatory variable, which is a new first explanatory variable determined from the outside and is one of several explanatory variables. [10] State derivation device for deriving a state of a target device from a second regression model created by a local model creation unit of a learning device and data corresponding to training data and data corresponding to a first explanatory variable acquired from the target device, wherein the learning device comprises: a global model creation unit, for creating, based on training data that can be explained by a multitude of explanatory variables, and a first explanatory variable that is an externally defined explanatory variable and is one of the multitude of explanatory variables, a first regression model that is applicable to the training data and the first explanatory variable, a variable selection unit for selecting a second explanatory variable from the multitude of explanatory variables and for selecting a second explanatory variable from the training data, with which target data, which are considered to vary based on the first regression model created by the global model creation unit, are separable from the training data; and the local model creation unit to construct a second regression model applicable to the training data and the first explanatory variable using training data after the target data has been separated based on the second explanatory variable selected by the variable selection unit and the first explanatory variable. [11] State derivation device according to claim 10, which further comprises a feedback information generation unit to correct a regression coefficient in the second regression model on the basis of a correction value for correcting the regression coefficient in the second regression model, wherein the correction value is received from outside. [12] Condition monitoring system comprising: a learning device, comprehensive: a global model creation unit for creating, based on training data that can be explained by a multitude of explanatory variables, and a first explanatory variable that is an externally defined explanatory variable and is one of the multitude of explanatory variables, a first regression model that is applicable to the training data and the first explanatory variable, a variable selection unit for selecting a second explanatory variable from the multitude of explanatory variables and for selecting a second explanatory variable from the training data, with which target data, which are considered to vary based on the first regression model created by the global model creation unit, are separable from the training data; and a local model building unit for creating a second regression model applicable between the training data and the first explanatory variable, using training data after the target data have been separated based on the second explanatory variable selected by the variable selection unit and the first explanatory variable; and a state derivation device for deriving a state of a target device using a second regression model created by the local model creation unit and data corresponding to the training data, as well as data corresponding to the first explanatory variable acquired from the target device. [13] Learning procedures performed by a learning device, comprising: a global model creation unit for creating, based on training data that can be explained by a multitude of explanatory variables, and a first explanatory variable that is an externally defined explanatory variable and is one of the multitude of explanatory variables, a first regression model that is applicable to the training data and the first explanatory variable; Selecting a second explanatory variable from the multitude of explanatory variables and selecting a second explanatory variable from the training data by a variable selection unit, with the target data, which are considered to vary based on the first regression model created by the global model creation unit, being separable from the training data; and Creating a second regression model applicable between the training data and the first explanatory variable by a local model building unit using training data after separating the target variables based on the second explanatory variable selected by the variable selection unit and the first explanatory variable.

Images