Learning device, state inference device, state monitoring system, and learning method

Abstract

A learning device (300) is provided with: a global model construction unit (304) that constructs a first regression model (global model) suitable for learning data and a first explanatory variable on the basis of the learning data that can be explained by a plurality of explanatory variables and the first explanatory variable (x1) that is one of the plurality of explanatory variables and that is specified from the outside; a variable selection unit that selects, from among the plurality of explanatory variables, a second explanatory variable (x2) by means of which target data, from among the learning data, deemed to have a deviation on the basis of the first regression model constructed by the global model construction unit can be separated from the learning data; and a local model construction unit (354) that constructs a second regression model (local model) suitable for the learning data and the first explanatory variable by using the learning data obtained by separating the target data on the basis of the second explanatory variable and the first explanatory variable selected by the variable selection unit.

Description

Technical Field

[0001] The present disclosure relates to a learning device, a state inference device, a state monitoring system, and a learning method. Background Art

[0002] In the manufacturing industry, a learned model obtained through machine learning is used to detect anomalies in equipment such as plants and rotating machinery (hereinafter referred to as "target equipment"). Here, an anomaly in the target equipment refers to, for example, degradation of the target equipment. Collecting abnormal data from the target equipment is generally more difficult than collecting normal data from the target equipment. Therefore, when learning a learned model, in most cases, a learning device uses only the normal data collected from the target equipment as learning data for unsupervised learning, allowing the model to learn. In this case, an inference device that infers the state of the target equipment uses the learned model to calculate an abnormality level, which indicates the degree to which the state of the target equipment deviates from normal. In this case, a threshold value for determining an abnormality is set for the abnormality level, and if the calculated abnormality level exceeds the threshold, the target equipment is determined to be abnormal. Related to this abnormality detection technology, for example, Non-Patent Documents 1 and 2 describe anomaly detection technologies based on linear regression models and Gaussian process regression models.

[0003] Prior art literature

[0004] Non-patent literature

[0005] Non-patent literature 1: Tsuyoshi Ide, "Introduction to Machine Learning: Abnormal Knowledge", Kornosha, 2019 Non-patent Literature 2: Tsuyoshi Ide, "Abnormal Knowledge and Abnormal Knowledge", Kornosha, 2018 Summary of the Invention

[0006] Problems to be solved by the invention

[0007] The aforementioned non-patent documents 1 and 2 describe anomaly detection techniques for preventing deviations in normal data learned by a learning device. On the other hand, target equipment such as complete plants and rotating machinery rarely operate continuously under certain operating conditions (e.g., certain operation modes and motion modes) and often operate under a variety of operating conditions. In such cases, the normal data collected from the target equipment may sometimes deviate due to differences in these operating conditions. Furthermore, the operating conditions of the target equipment are determined by a large amount of control information (parameters) ranging from tens to hundreds, such as the current and voltage values ​​of the power required for the operation of the target equipment.

[0008] Thus, when applying the anomaly detection techniques described in Non-Patent Documents 1 and 2 to a situation where normal data exhibits deviations, it is considered possible to train a desired regression model using a learning device (computer) employing this anomaly detection technique. In this case, the learning device (hereinafter also referred to as the "conventional device") clusters normal data in a manner that encompasses all patterns, and as a result, an inference device uses the learned regression model to perform anomaly detection. From a physical perspective, the operating conditions of the equipment targeted for anomaly detection are often limited. For example, in the case of anomaly detection in rotating machinery, normal data collected from the rotating machinery when power is supplied may include the influence of electromagnetic noise caused by the current. Therefore, in this case, conventional devices are expected to limit their analysis (construction and evaluation of the regression model) to normal data collected from the rotating machinery when power is not supplied. However, conventional devices currently have difficulty performing such an analysis, resulting in an increase in the number of hours required to learn the regression model.

[0009] The present disclosure is made to solve the above-mentioned problems, and its purpose is to provide a learning device that can reduce the time required for learning compared to conventional methods when learning a regression model for detecting abnormalities in target equipment using biased learning data.

[0010] Means used to solve problems

[0011] The learning device disclosed in the present invention comprises: a global model construction unit, which constructs a first regression model suitable for the learning data and the first explanatory variable based on learning data that can be explained by multiple explanatory variables and a first explanatory variable specified from the outside as one of the multiple explanatory variables; a variable selection unit, which selects a second explanatory variable from the multiple explanatory variables, and through which object data in the learning data that is considered to have deviations based on the first regression model constructed by the global model construction unit can be separated from the learning data; and a local model construction unit, which uses the learning data after the object data is separated based on the second explanatory variable and the first explanatory variable selected by the variable selection unit to construct a second regression model suitable for the learning data and the first explanatory variable.

[0012] Effects of the Invention

[0013] According to the present disclosure, when a model for detecting abnormality in a target device is trained using data with deviations collected from the target device, the man-hours required for the training can be reduced compared to conventional methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 This is a diagram showing a configuration example of a state monitoring system according to the first embodiment.

[0015] Figure 2 This is a diagram showing a configuration example of a learning device according to the first embodiment.

[0016] Figure 3 This is a diagram showing an example of vibration data and control information data in the first embodiment.

[0017] Figure 4 This is a diagram showing a configuration example of a global learning unit (global model building unit and model evaluation unit) in the first embodiment.

[0018] Figure 5 This is a diagram showing an image example of the global model in the first embodiment.

[0019] Figure 6 This is a diagram showing an image example of the global model in the first embodiment.

[0020] Figure 7A 、 Figure 7B 、 Figure 7C This is a diagram illustrating a specific example of the classification process performed by the filtering processing unit in the first embodiment.

[0021] Figure 8 This is a diagram showing a configuration example of a local learning unit (a range selection unit, a local model construction unit, and a model evaluation unit) in the first embodiment.

[0022] Figure 9 This is a diagram illustrating a specific example of processing performed by the distribution calculation unit in the first embodiment.

[0023] Figure 10 This is a diagram showing an example of a probability distribution diagram generated by the second variable selection processing unit in the first embodiment.

[0024] Figure 11 This is a diagram showing an example of a distribution map of vibration data generated by the region evaluation unit in the first embodiment.

[0025] Figure 12 This is a diagram showing an example of an image generated by the image output unit in the first embodiment and indicating a calculation result of a prediction error.

[0026] Figure 13 This is a flowchart for explaining an example of the operation of the learning device according to the first embodiment.

[0027] Figure 14A 、 Figure 14B This is a diagram showing an example of the hardware configuration of the learning device according to the first embodiment.

[0028] Figure 15 This is a diagram showing a configuration example of a state inference device according to the first embodiment.

[0029] Figure 16 This is a diagram showing a configuration example of an evaluation unit in the state inference device according to the first embodiment.

[0030] Figure 17 This is a diagram showing an example of an image representing a comparison result generated by the image output unit in the first embodiment.

[0031] Figure 18 This is a flowchart for explaining an example of the operation of the state inference device according to the first embodiment.

[0032] Figure 19A 、 Figure 19B This is a diagram showing an example of the hardware configuration of the state inference device according to the first embodiment.

[0033] Figure 20A This is a diagram for explaining the difficulty of selecting an appropriate regression model in conventional devices. Figure 20B This is a diagram for explaining the difficulty of evaluating the deviation of learning data (normal data) in a conventional device. DETAILED DESCRIPTION

[0034] Hereinafter, embodiments of the present disclosure will be described in detail with reference to the accompanying drawings.

[0035] Implementation method 1.

[0036] Figure 1 1 is a diagram showing a configuration example of a state monitoring system 1000 according to Embodiment 1. Figure 1 As shown, the system includes a recording unit 100 , a learning data recording unit 200 , a learning device 300 , and a state inference device 600 .

[0037] The recording unit 100 is composed of a recording medium such as a HDD (Hard Disk Drive) or an SDD (Solid State Drive), and records data representing a learned model constructed by the learning device 300 .

[0038] The learning data recording unit 200 is composed of a recording medium such as an HDD (Hard Disk Drive) or an SDD (Solid State Drive), and records learning data used by the learning device 300 to construct a learned model.

[0039] The learning device 300 and the state inference device 600 are each configured to be connectable to the recording unit 100 . Furthermore, the learning device 300 is configured to be connectable to the learning data recording unit 200 .

[0040] The learning device 300 constructs a learned model for detecting abnormalities in equipment (target equipment) such as plants and rotating machinery through machine learning using the learning data recorded in the learning data recording unit 200. The learning device 300 records data representing the constructed learned model in the recording unit 100.

[0041] The state inference device 600 infers the state of the target device using the learned model indicated by the data recorded in the recording unit 100 by the learning device 300, thereby detecting an abnormality (eg, deterioration) of the target device.

[0042] In addition, the state monitoring system 1000 is, for example, Figure 1 As shown, the configuration includes a first external evaluation device 400 , a second external evaluation device 500 , and a third external evaluation device 700 .

[0043] The first external evaluation device 400 and the second external evaluation device 500 are configured to be connectable to the learning device 300. The first external evaluation device 400 and the second external evaluation device 500 serve as interfaces with the learning device 300, such as transmitting user instructions to the learning device 300 and presenting the user with the details of the processing performed by the learning device 300.

[0044] The third external evaluation device 700 is connectable to the state inference device 600. The third external evaluation device 700 serves as an interface to the state inference device 600, such as transmitting user instructions to the state inference device 600 or presenting the user with the details of the processing performed by the state inference device 600.

[0045] In the following description, for convenience of explanation, the details of the learning device 300 will be described first, and then the details of the state inference device 600 will be described.

[0046] <Learning device 300>

[0047] Figure 2 1 is a diagram showing a configuration example of a learning device 300 according to Embodiment 1. The learning device 300 is, for example, Figure 2 As shown, the system includes a global learning unit 301 , a local learning unit 350 , and an intermediate recording unit 390 .

[0048] The learning device 300 uses the target variable described by any multiple explanatory variables (hereinafter also referred to as "explanatory variable group") as learning data, and performs machine learning in two stages: learning based on the global learning unit 301 and learning based on the local learning unit 350, thereby constructing a completed learning model for detecting abnormalities of the target device.

[0049] Specifically, first, the global learning unit 301 obtains a first explanatory variable selected from the explanatory variable group through a user with skills and knowledge related to the target device. Furthermore, the global learning unit 301 uses the target variable explained by the obtained first explanatory variable as learning data for machine learning, and constructs a learned model that is suitable for the learning data and the first explanatory variable. Furthermore, the global learning unit 301 obtains an evaluation of the learned model from the user via the first external evaluation device 400. Thus, the global learning unit 301 constructs a global learned model that has obtained appropriateness from a physical point of view. The global learning unit 301 records data representing the constructed learned model in the intermediate recording unit 390.

[0050] Next, the local learning unit 350 uses the learned model indicated by the data recorded in the intermediate recording unit 390 by the global learning unit 301 to classify the aforementioned learning data into "data deemed to have deviations (large deviations)" and "data deemed to have no deviations (small deviations)." Furthermore, the local learning unit 350 searches the aforementioned explanatory variable group for a second explanatory variable, different from the first explanatory variable selected by the user, that can accurately separate the "data deemed to have deviations (large deviations)" from the aforementioned learning data. The phrase "separable" or "separable" does not mean that the "data deemed to have deviations (large deviations)" can be completely separated from the learning data, but rather that the "data deemed to have deviations (large deviations)" can be roughly separated from the "data deemed to have deviations (large deviations)."

[0051] The local learning unit 350 then performs machine learning using the learning data from which "data deemed to have significant deviations" has been separated based on the searched second explanatory variable and the aforementioned first explanatory variable. This allows the local learning model 350 to be constructed, which is appropriate for the local relationship between the separated learning data and the first explanatory variable. Here, the local learning model refers to a model obtained by learning using the learning data from which "data deemed to have significant deviations" has been separated from a physical perspective.

[0052] Here, the learning data used by the learning device 300 for learning is recorded in the learning data recording unit 200. The learning data recording unit 200 is, for example, Figure 2 As shown, a vibration DB 210 and a control information DB 220 are provided.

[0053] The vibration DB 210 records vibration data. For example, the vibration data Figure 3As shown in the top graph, this is data representing the temporal variation of vibration amplitude values. Alternatively, vibration data can be data representing the temporal variation of a characteristic quantity of the vibration amplitude values. In this case, the characteristic quantity of the vibration amplitude values ​​can be, for example, the RMS value of the vibration amplitude values. The following description uses the case where the vibration data is the RMS value of the vibration amplitude values ​​as an example.

[0054] The control information DB 220 records the control information data as the description variable. The control information data is, for example, Figure 3 As shown in the second through fourth graphs from the top, this data represents the temporal changes in control information. Control information refers to parameters that determine the operating conditions of the target device, such as the rotational speed of rotating machinery, the current value of the drive power for the rotating machinery, and acceleration / decel (acceleration / deceleration).

[0055] In addition, each control information data recorded in the control information DB 220 is synchronized in time with the vibration data recorded in the vibration DB 210. In this case, the vibration data corresponds to the target variable, and each control information data corresponds to the explanatory variable for explaining the vibration data.

[0056] In the following description, the vibration data corresponds to the target variable and the control information data corresponds to the explanatory variables. However, this is merely an example; the target variable and explanatory variables may be other data types. Furthermore, in the following description, the explanatory variable group is also referred to as the control information group.

[0057] <Global Learning Unit 301>

[0058] The global learning unit 301 is, for example, Figure 2 As shown, the system includes a data extraction unit 302 , an explanatory variable acquisition unit 303 , a global model construction unit 304 , and a model evaluation unit 305 .

[0059] (Explanatory Variable Acquisition Unit 303)

[0060] First, the user selects arbitrary control information from the explanatory variable group (control information group) that describes the vibration data and inputs the selected control information into the first external evaluation device 400. Here, for easier understanding, it is assumed that the user has selected "rotation speed" as the control information. The explanatory variable acquisition unit 303 acquires the control information input by the user from the first external evaluation device 400 as the first explanatory variable x1. Furthermore, the explanatory variable acquisition unit 303 outputs data representing the acquired first explanatory variable x1 to the data extraction unit 302 as a variable descriptor D13.

[0061] (Data Extraction Unit 302)

[0062] The data extraction unit 302 obtains the variable descriptor D13 from the explanation variable acquisition unit 303. The data extraction unit 302 obtains the control information data corresponding to the obtained variable descriptor D13 from the control information DB 220 in the learning data recording unit 200. Here, since the variable descriptor D13 indicates "rotation speed", the data extraction unit 302 obtains the control information data corresponding to the obtained variable descriptor D13 from the control information DB 220. Figure 3 The control information data is shown second from the top.

[0063] Furthermore, the data extraction unit 302 acquires vibration data from the vibration DB 210 in the learning data recording unit 200. The data extraction unit 302 then outputs the acquired control information data and vibration data to the global model construction unit 304 as learning data D12.

[0064] (Global Model Building Unit 304)

[0065] The global model building unit 304 learns a regression model suitable for the vibration data and the first explanatory variable x1 based on the vibration data that can be explained by a plurality of explanatory variables and the first explanatory variable x1, wherein the first explanatory variable x1 is an explanatory variable specified from the outside and is one of the plurality of explanatory variables. Figure 4 As shown, the system includes a model building unit 311 and a model updating unit 312 .

[0066] The model construction unit 311 obtains the learning data D12 from the data extraction unit 302. The model construction unit 311 uses the obtained learning data D12 to perform learning based on unsupervised learning, thereby constructing a regression model. At this time, the model construction unit 311 uses the control information data (rotation speed) contained in the learning data D12 as an explanatory variable and the vibration data (RMS value of the vibration amplitude value) contained in the learning data D12 as a target variable to perform learning based on unsupervised learning. In addition, as a learning method in this case, known learning methods such as linear regression, polynomial regression, and Gaussian process regression can be used. In addition, in the following description, the regression model constructed here is also referred to as a "global model."

[0067] The global model 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 be able to reproduce the rough regression trend between the control information data and the vibration data. Therefore, when constructing the global model, the model construction unit 311 does not necessarily need to use all the control information data and vibration data included in the learning data D12. For example, the model construction unit 311 can also construct the global model using control information data and vibration data corresponding to an arbitrary time range specified by the user.

[0068] The model construction unit 311 outputs data representing the constructed global model (hereinafter also referred to as “global model data”) and control information data and vibration data used for learning the global model to the model updating unit 312 as data D18 .

[0069] Alternatively, the model construction unit 311 may construct a global model for a plurality of modes. In this case, the model construction unit 311 outputs global model data for each mode, as well as control information data and vibration data used for learning, to the model updating unit 312 as data D18.

[0070] The model updating unit 312 obtains data D18 from the model building unit 311. Furthermore, when the model evaluation unit 355 of the local learning unit 350 (described later) outputs data D60, the model updating unit 312 obtains this data D60 from the model evaluation unit 355 and updates (rebuilds) the global model in accordance with user instructions. The update process in this case will be described later.

[0071] When the global model is updated, model updating unit 312 outputs data representing the updated global model, along with the control information data and vibration data used in the update, as data D14 to model evaluation unit 305. Furthermore, when model evaluation unit 355 does not output data D60 and the global model is not updated, model updating unit 312 outputs data D18 as data D14 to model evaluation unit 305.

[0072] (Model Evaluation Unit 305)

[0073] The model evaluation unit 305 receives external (eg user) evaluations of the global model represented by the data included in the data D14. Figure 4 As shown, the configuration includes an image output unit 313 and a model determination unit 314 .

[0074] Image output unit 313 obtains data D14 from model updating unit 312. Based on the global model data included in the obtained data D14, image output unit 313 visualizes the global model represented by the data, thereby generating data representing an image of the global model (hereinafter also referred to as "global model image data"). Image output unit 313 outputs the generated global model image data as data D15 to first external evaluation device 400.

[0075] Figure 5 Examples of images showing global models. For example, Figure 5 In FIG. 5 , reference numeral 501 denotes a curve (prediction line) representing a regression equation obtained by the global model, and reference numeral 502 denotes a boundary of a confidence interval (eg, curve 501±5%) set for the curve (prediction line) representing the regression equation.

[0076] In addition, when the data D14 includes global model data of a plurality of patterns, the image output unit 313 generates a plurality of patterns based on the global model data, for example, Figure 6 In this manner, global model image data for each mode is generated, and each generated global model image data is output to the first external evaluation device 400 as data D15 .

[0077] The first external evaluation device 400 obtains data D15 from the image output unit 313. Based on the obtained data D15, the first external evaluation device 400 displays one or more global model images on a display unit (not shown), such as a monitor. If a single global model image is displayed on the display unit, the user examines the image and determines whether the global model is correct from a physical perspective. If the global model is correct, the user inputs the determination result to the first external evaluation device 400. Alternatively, if multiple global model images are displayed on the display unit, the user examines each image, selects the global model deemed correct from a physical perspective, and inputs the selection result to the first external evaluation device 400.

[0078] Furthermore, at this time, the user specifies a range within the time range of the time series of the control information data (here, the rotational speed) used for the above-mentioned learning, in which the vibration data is expected to have relatively low deviations, or a range in which the vibration data is expected to reflect the characteristics of the target device, as search width S, and inputs this to first external evaluation device 400. Here, search width S is a variable used when searching for an area with low vibration data deviations in range selection unit 352 of local learning unit 350, which will be described later.

[0079] The first external evaluation device 400 outputs data obtained by combining the data indicating the user's determination result or selection result and the data indicating the search width S input by the user to the model determination unit 314 as data D16 .

[0080] If a physically correct global model does not exist, the user can perform either of the following two actions. For example, the user can use the first external evaluation device 400 to discard the global model constructed at that point in time and input control information different from the initially input control variable (here, the rotational speed) into the first external evaluation device 400. The explanatory variable acquisition unit 303 then acquires this different control information as the new first explanatory variable x1. Subsequently, the global model construction unit 304 can reconstruct the global model through the same process as described above.

[0081] Alternatively, the user can maintain the first explanatory variable x1 constant and, using the first external evaluation device 400, exclude data deemed to be deviated from the vibration data based on the global model image, and then instruct the global model construction unit 304 to rebuild the global model. The user can repeat any of these steps until a global model deemed correct from a physical point of view is constructed.

[0082] Model determination unit 314 obtains data D16 from first external evaluation device 400. Based on this obtained data D16, model determination unit 314 records data representing the global model determined by the user to be correct from a physical perspective, or data representing the global model selected by the user as the correct model from a physical perspective, as data D17 in intermediate recording unit 390. The model indicated by data D17 corresponds to the aforementioned global learned model.

[0083] In addition, at this time, the model determination unit 314 uses the data representing the above-mentioned search width S contained in the data D16 as a range descriptor, and includes the range descriptor and the identifier (such as the name) of the control information data used for learning the above-mentioned global model in the data D17 and records it in the intermediate recording unit 390.

[0084] (Intermediate Recording Unit 390)

[0085] The intermediate recording unit 390 records data D17. Specifically, the intermediate recording unit 390 records data indicating a global model corresponding to a global learned model (global model data), an identifier of the control information data, and a range descriptor.

[0086] <Local Learning Unit 350>

[0087] The local learning unit 350 is, for example, Figure 2 As shown, the second variable selection unit 360 includes a local model construction unit 354 and a model evaluation unit 355. The second variable selection unit 360 includes a filtering unit 351, a range selection unit 352 and a second variable selection processing unit 353, for example.

[0088] (Filtering Processing Unit 351)

[0089] The filtering processing unit 351 acquires the data D17 (global model data, identifier of control information data, and range descriptor) recorded in the intermediate recording unit 390 as a global model descriptor D51.

[0090] Furthermore, the filtering processing unit 351 refers to the learning data recording unit 200 and acquires, as data D52 , data that matches the identifier of the control information data included in the data D17 from among the vibration data recorded in the vibration DB 210 and the control information data recorded in the control information DB 220 .

[0091] Then, based on the acquired global model descriptor D51 and data D52 , the filtering unit 351 classifies (filters) the vibration data included in the data D52 into data deemed to have deviations and data deemed not to have deviations, and labels both classified data.

[0092] For example, filtering unit 351 determines the degree of deviation in the vibration data based on the global model, labels the vibration data deemed to have deviations with the label "DataA," and labels the vibration data deemed to have no deviations with the label "DataB." Filtering unit 351 then combines the labeled vibration data with the global model descriptor D51, and outputs the resulting data as data D53 to range selection unit 352.

[0093] FIG7 shows a specific example of the classification process performed by the filter processing unit 351. For example, Figure 7A The distribution diagram of vibration data (RMS value) is shown in which the horizontal axis is the first explanatory variable x1 (rotation speed) and the vertical axis is the vibration data. Figure 7B The image of the global model indicated by the global model data included in the data D17 recorded in the intermediate recording unit 390 is shown. Figure 7C This is a distribution diagram of the vibration data after the classification process performed by the filtering processing unit 351 .

[0094] For example, the filtering processing unit 351 Figure 7A and Figure 7B Overlap, will Figure 7A The vibration data shown is located for Figure 7B The vibration data outside the confidence interval set by the curve (prediction line) 701 in the global model of the present invention are regarded as data with deviations, and the data are labeled "DataA" ( Figure 7C In addition, the filtering processing unit 351 Figure 7AAmong the vibration data shown, the vibration data that are inside the above-mentioned confidence interval are regarded as data without deviation, and these data are labeled as "Data B" ( Figure 7C (see the upper side of the figure).

[0095] In the following description, for easier understanding, data deemed to have discrepancies by the filtering unit 351 will be referred to as "Data A," and data deemed to have no discrepancies will be referred to as "Data B." Furthermore, these data will be collectively referred to as "labeled data."

[0096] (Range Selection Unit 352)

[0097] The range selection unit 352 is as follows, for example Figure 8 As shown, the configuration includes a distribution calculation unit 361 and a distribution difference comparison unit 362 .

[0098] The distribution calculation unit 361 obtains the data D53 from the filtering processing unit 351. The distribution calculation unit 361 analyzes the distribution of DataA and DataB based on the labeled data included in the obtained data D53. Specifically, the distribution calculation unit 361 analyzes the distribution of DataA and DataB. Figure 9 As shown, the probability distribution pA of DataA and the probability distribution pB of DataB for the first explanatory variable x1 are calculated, and data representing the calculated probability distributions and data D53 are output to the distribution difference comparison unit 362 .

[0099] in addition, Figure 9 The “search width” shown in FIG. 4 shows the search width input by the user via the first external evaluation device 400. Figure 9 In the example shown in FIG, the search width is set to a rotation speed between 500 and 1000. This means that if the rotation speed is between 500 and 1000, the user determines that the deviation of the vibration data is relatively small.

[0100] Distribution difference comparison unit 362 obtains data representing probability distributions pA and pB and data D53 from distribution calculation unit 361. Based on the obtained data representing probability distributions pA and pB, distribution difference comparison unit 362 calculates the difference between probability distribution pA and probability distribution pB. At this time, distribution difference comparison unit 362 selects a range in the time series of control information data (rotation speed) where the difference is maximized, from the search width indicated by the range descriptor included in data D53. This range is the range where probability distribution pB is larger than probability distribution pA.

[0101] Specifically, the distribution difference comparison unit 362 calculates the differences pB-pA based on the probability distributions pA and pB obtained from the distribution calculation unit 361. In this case, when the search width is set to S, the distribution difference comparison unit 362 selects a range Ω from the search width S that maximizes pB|Ω-pA|Ω (where Ω = [a, a+S], where a is an arbitrary value of the first explanatory variable x1) and uses the selected range Ω as a new range descriptor. The distribution difference comparison unit 362 then outputs the resulting data, which is a combination of this new range descriptor, the labeled data (DataA and DataB), and the global model descriptor D51, as data D54 to the second variable selection processing unit 353.

[0102] In addition, here, an example is described in which the distribution difference comparison unit 362 selects the range Ω that maximizes pB|Ω-pA|Ω, but the distribution difference comparison unit 362 is not limited to this. For example, it can also select a range Ω that makes pB|Ω-pA|Ω greater than a specified value, and use the selected range Ω as a new range descriptor.

[0103] (Second variable selection processing unit 353)

[0104] The second variable selection processing unit 353 obtains data D54 from the range selection unit 352. The second variable selection processing unit 353 then selects, from the explanatory variable group, explanatory variables (control information) other than the first explanatory variable x1 selected by the user during learning in the global learning unit 301, that can accurately separate Data A and Data B. In the following description, the explanatory variable selected here will be referred to as the "second explanatory variable x2."

[0105] For example, the second variable selection processing unit 353 generates Figure 10 The probability distribution diagram shown. Figure 10 In the graph, the horizontal axis represents an explanatory variable other than the first explanatory variable x1 and a candidate explanatory variable for the second explanatory variable x2. The vertical axis represents the frequency of occurrence of DataA and DataB included in the data D54 and existing in the above-mentioned range Ω.

[0106] At this time, the second variable selection processing unit 353 searches for explanatory variables according to the following procedure while sequentially changing the explanatory variables that are candidates for the second explanatory variable x2, and selects the searched explanatory variable as the second explanatory variable x2.

[0107] (1) The second variable selection processing unit 353 is, for example, Figure 10In the probability distribution diagram shown, the range of the horizontal axis where Data A accounts for a proportion of "100-ε" or more of all Data A, assuming the entire length of the horizontal axis is "100" (ε is a small positive integer), is designated as Y (first range). For example, assuming ε = 5, the second variable selection processing unit 353 designates the range of the horizontal axis that includes Data A at least 95% of all Data A as Y.

[0108] (2) Next, the second variable selection processing unit 353 sets the range excluding range Y on the horizontal axis of the probability distribution graph as range X (the second range), and searches for an explanatory variable for which the proportion of DataB in all the data included in range X is greater than or equal to a predetermined value (e.g., 80%). The second variable selection processing unit 353 then selects the searched explanatory variable as the second explanatory variable x2. Furthermore, if the second variable selection processing unit 353 searches for multiple explanatory variables for which the proportion of DataB in all the data included in range X is greater than or equal to a predetermined value, the second variable selection processing unit 353 selects the explanatory variable with the largest proportion of DataB in range X as the second explanatory variable x2, for example.

[0109] If the search fails, the second variable selection processing unit 353 repeats steps (1) and (2) above while sequentially changing the explanatory variables that are candidates for the second explanatory variable x2. The second variable selection processing unit 353 then outputs data obtained by combining the second explanatory variable x2 selected in the above steps with the data D54 as data D56 to the local model building unit 354.

[0110] The second explanatory variable x2 is a variable different from the first explanatory variable x1 specified by the user during learning in the global learning unit 301 , and is highly likely to enable accurate separation of Data A and Data B by combining with the first explanatory variable x1 .

[0111] (Local Model Construction Unit 354)

[0112] The local model building unit 354 is as follows, for example Figure 8 As shown, the system is configured to include a region evaluation unit 363 and a model construction unit 364 .

[0113] (Regional Evaluation Unit 363)

[0114] The region evaluation unit 363 obtains the data D56 from the second variable selection processing unit 353. The region evaluation unit 363 generates, for example, Figure 11 The distribution diagram of the vibration data is shown.

[0115] The distribution diagram is, for example, Figure 11 As shown, the horizontal axis is the first explanatory variable x1 (rotation speed), the vertical axis is the second explanatory variable x2, and the vibration data is displayed in the area determined by the combination of both variables (hereinafter also referred to as the "combination area".), and the area where DataA and DataB appear in the combination area is shown. Figure 11 In FIG, DataA is shown as gray dots and DataB is shown as black dots.

[0116] Thus, the area evaluation unit 363 can clearly show the area with more DataA and the area with less DataA in the combined area by displaying DataA and DataB in the combined area (in the example of FIG. Figure 11 (Indicated by reference numerals U1 to U3 in the figure). The region evaluation unit 363 outputs data representing the generated distribution map to the second external evaluation device 500 as data D62.

[0117] The second external evaluation device 500 acquires the data D62 from the area evaluation unit 363. Based on the acquired data D62, the second external evaluation device 500 displays the image of the distribution map on a display unit (not shown) such as a monitor.

[0118] The user refers to the image of the distribution graph displayed on the display unit, and selects, for example Figure 11 The user selects a region with relatively less Data A in the combined region, such as regions U1 to U3, and inputs the selected region to the second external evaluation device 500. In this case, the user can select only one region, such as region U1, or multiple regions, such as regions U1 to U3. The second external evaluation device 500 outputs data representing the input region as region range data D63 to the model construction unit 364.

[0119] The model building unit 364 obtains the area range data D63 from the second external evaluation device 500. In addition, the model building unit 364 obtains the data D56 from the second variable selection processing unit 353. Then, the model building unit 364 determines, for example, Figure 11 The vibration data contained in regions U1 to U3 is used as learning data, and unsupervised learning is performed to construct a regression model. In the following description, the regression model constructed here is referred to as a "local model." The local model takes the first explanatory variable x1 (control information data) as input and outputs the target variable (vibration data).

[0120] Furthermore, when the user selects multiple regions, the model construction unit 364 constructs a local model for each selected region. Furthermore, in this case, the model construction unit 364 uses the same learning model as that used by the global model construction unit 304 of the global learning unit 301. However, the two learning models do not necessarily need to be the same model.

[0121] The model construction unit 364 outputs data obtained by combining data representing the constructed local model (hereinafter also referred to as “local model data”) and the vibration data (Data A and Data B) used for learning to the model evaluation unit 355 as data D57 .

[0122] (Model Evaluation Unit 355)

[0123] The model evaluation unit 355 receives external (eg user) evaluation of the local model indicated by the local model data included in the data D57. Figure 8 As shown, the system is configured to include a prediction error calculation unit 365 , an image output unit 366 , and a model determination unit 367 .

[0124] The prediction error calculation unit 365 acquires the data D57 from the model construction unit 364. The prediction error calculation unit 365 calculates the prediction error of the local model based on the acquired data D57.

[0125] For example, prediction error calculation unit 365 inputs the first explanatory variable x1 (rotational speed) into the local model indicated by the local model data included in data D57 and calculates the error between the vibration data (RMS value) output from the local model and the vibration data that should have been output. In this case, prediction error calculation unit 365 calculates the prediction error using, for example, a mean absolute percentage error (MAPE). Prediction error calculation unit 365 outputs the first explanatory variable x1 and the vibration data used in the prediction error calculation, along with data representing the calculated prediction error, to image output unit 366.

[0126] The image output unit 366 obtains the first explanatory variable x1, the vibration data, and the data indicating the prediction error from the prediction error calculation unit 365. Then, based on the obtained data, the image output unit 366 generates, for example, Figure 12 Then, the image output unit 366 outputs the data obtained by combining the generated image data and the prediction error data to the second external evaluation device 500 as data D58. Figure 12 In the image shown on the right, Figure 5The image of the global model shown similarly shows a curve (prediction line) representing a regression equation obtained by a local model and the boundaries of a confidence interval (eg, ±5% of the curve) set for the curve (prediction line) representing the regression equation.

[0127] The second external evaluation device 500 obtains data D58 from the image output unit 366. Based on the obtained data D58, the second external evaluation device 500 converts, for example, Figure 12 The image shown on the right side of the image is displayed on a display unit (not shown), such as a monitor. The user refers to the image displayed on the display unit, determines a local model to be ultimately output to the recording unit 100 from among the local models, and inputs the identifier of the determined local model to the second external evaluation device 500. The second external evaluation device 500 outputs the input local model identifier as data D59 to the model determination unit 367.

[0128] The model determination unit 367 acquires data D59 from the second external evaluation device 500. Based on the acquired data D59, the model determination unit 367 records data representing the local model that the user has determined to output as a final output in the recording unit 100 as data D61.

[0129] Furthermore, at this time, the model determination unit 367 includes data (hereinafter also referred to as "control condition data") indicating the conditions (hereinafter also referred to as "control conditions") of the control information (explanatory variables) when constructing the local model determined by the user to be output, in the data D61 and records it in the recording unit 100. Here, the control conditions include, for example, the type of the first explanatory variable x1 (such as the rotational speed), the type of the second explanatory variable x2 (other than the rotational speed), the range of the first explanatory variable x1, and the range of the second explanatory variable x2 for which learning data exists when constructing the local model.

[0130] When the user specifies multiple partial models as output targets, the model determination unit 367 records the multiple partial model data in the storage unit 100. In this case, the model determination unit 367 records the control condition data in the storage unit 100 in association with each of the multiple partial models.

[0131] Furthermore, if the user, despite referring to the image displayed on the display unit, fails to find the partial model to be ultimately output to the recording unit 100, the user inputs this information to the second external evaluation device 500. The second external evaluation device 500 outputs data indicating that there is no partial model to be output as data D59 to the model determination unit 367.

[0132] After obtaining data D59, the model determination unit 367 obtains data D57 (data obtained by merging local model data and vibration data (DataA and DataB) used for learning the local model) from the prediction error calculation unit 365, and outputs the obtained data D57 as data D60 to the model update unit 312 of the global learning unit 301.

[0133] The model updating unit 312 obtains data D60 from the model determining unit 367. Upon obtaining data D60, the model updating unit 312 causes the display unit of the first external evaluation device 400 to display the contents of data D60. Furthermore, the model updating unit 312 causes the display unit of the first external evaluation device 400 to instruct the user to update (i.e., recreate) the global model.

[0134] The user accepts this display, reselects an explanatory variable different from the first explanatory variable x1 initially selected when constructing the global model, and inputs the newly selected explanatory variable into the first external evaluation device 400. Subsequently, the model updating unit 312 updates (reconstructs) the global model using the same method as the model construction unit 311 described above.

[0135] While this example describes updating (rebuilding) the global model by reselecting an explanatory variable different from the first explanatory variable x1 initially selected when building the global model, the method for updating (rebuilding) the global model is not limited to this. For example, the user can maintain the first explanatory variable x1 initially selected when building the global model, then change the range of the first explanatory variable x1 relative to its initial state, such as by narrowing the range of the first explanatory variable x1. In this case, the model updating unit 312 can simply update (rebuild) the global model using the learning data included in the changed range.

[0136] Alternatively, the model updating unit 312 may instruct the user to re-enter the search width via the first external evaluation device 400, for example, without updating the global model. In this case, the user enters a new search width via the first external evaluation device 400, and this new search width is recorded as a range descriptor in the intermediate recording unit 390. Subsequently, based on this new range descriptor, the second variable selection processing unit 353 selects a new second explanatory variable x2, and the local learning unit 350 reconstructs the local model.

[0137] Alternatively, the model updating unit 312 may not update the global model, but may instruct the user via the second external evaluation device 500 to Figure 11In this case, the user inputs a new region via the second external evaluation device 500, and data representing the new region is sent to the model construction unit 364 as region range data. Subsequently, the model construction unit 364 reconstructs the local model using this new region range data.

[0138] Next, refer to Figure 13 An example of the operation of the learning device 300 according to the first embodiment will be described with reference to the flowchart shown in FIG.

[0139] First, the explanatory variable acquisition unit 303 acquires control information input by the user to the first external evaluation device 400 as the first explanatory variable x1 (step ST1). The explanatory variable acquisition unit 303 outputs data representing the acquired first explanatory variable x1 to the data extraction unit 302 as a variable descriptor D13.

[0140] Next, the data extraction unit 302 obtains control information data corresponding to the obtained variable descriptor D13 from the control information DB 220 in the learning data recording unit 200. The data extraction unit 302 also obtains vibration data as learning data from the vibration DB 210 in the learning data recording unit 200 (step ST2).

[0141] Next, the model construction unit 311 constructs a global model using the data acquired in step ST2 (step ST3 ).

[0142] Next, the image output unit 313 generates global model image data and outputs the generated global model image data to the first external evaluation device 400 (step ST4). Based on the acquired data, the first external evaluation device 400 displays the image of the global model on a display unit such as a monitor and receives the user's judgment or selection result. The first external evaluation device 400 outputs data indicating the user's judgment or selection result to the model determination unit 314.

[0143] Next, the model determination unit 314 obtains data indicating the user's decision or selection result and determines whether the result indicates that any global model has been selected (step ST5). If the result indicates that no global model has been selected (step ST5: No), the process returns to step ST1, and the explanatory variable acquisition unit 303 obtains a new first explanatory variable x1 from the user via the first external evaluation device 400. Steps ST2 to ST5 are then repeated.

[0144] On the other hand, when the above result indicates that an arbitrary global model is selected (step ST5; yes), the processing moves to step ST6, and the filtering processing unit 351 classifies (filters) the vibration data into data (DataA) that is considered to have deviations and data (DataB) that is considered to have no deviations (step ST6).

[0145] Next, the distribution calculation unit 361 calculates the probability distribution pA of DataA and the probability distribution pB of DataB for the first explanatory variable x1. Furthermore, the distribution difference comparison unit 362 selects a range Ω from the search width S that maximizes pB|Ω-pA|Ω (Ω=[a, a+S], where a is an arbitrary value of the first explanatory variable x1) (step ST7).

[0146] Next, the second variable selection processing unit 353 selects a second explanatory variable x2 which is an explanatory variable other than the first explanatory variable x1 and which can separate DataA and DataB with high accuracy (step ST8 ).

[0147] Next, the region evaluation unit 363 generates a distribution map showing the vibration data in the region determined by the combination of the first explanatory variable x1 and the second explanatory variable x2. Then, the model construction unit 364 receives region selection from the user based on the distribution map (step ST9).

[0148] Next, the model construction unit 364 constructs a local model using the vibration data and control information data included in the region selected in step ST9 (step ST10 ).

[0149] Next, the prediction error calculation unit 365 calculates the prediction error of the local model, and the image output unit 366 generates image data representing the prediction result and outputs it to the second external evaluation device 500. The second external evaluation device 500 displays the image representing the prediction result on the display unit and receives the user's judgment or selection result. The second external evaluation device 500 outputs data representing the user's judgment or selection result to the model determination unit 367.

[0150] Next, the model determination unit 367 obtains data indicating the user's decision or selection result and determines whether the result indicates that any partial model has been selected (step ST11). If the result indicates that no partial model has been selected (step ST11: No), the model update unit 312 instructs the user to select a new first explanatory variable x1 via the first external evaluation device 400. The process then proceeds to step S21, where the explanatory variable acquisition unit 303 obtains the new first explanatory variable x1 from the user via the first external evaluation device 400. Steps ST2 to ST11 are then repeated.

[0151] In addition, although Figure 13 Although not shown in the flowchart, if the above result indicates that no local model has been selected (step ST11: No), the model updating unit 312 may instruct the user, via the first external evaluation device 400, to change the range of the first explanatory variable x1, for example, by narrowing the range of the first explanatory variable x1. If the model updating unit 312 instructs the user to change the range of the first explanatory variable x1, the process returns to step ST2.

[0152] Similarly, if the above result indicates that no local model has been selected (step ST11; No), the model updating unit 312 may instruct the user to re-enter the search width or reselect the regions U1 to U3 via the first external evaluation device 400. If the model updating unit 312 instructs the user to re-enter the search width, the process returns to step ST7; if the model updating unit 312 instructs the user to re-select the regions U1 to U3, the process returns to step ST9.

[0153] On the other hand, if the result indicates that an arbitrary local model has been selected (step ST11; YES), the process proceeds to step S32, and the model determination unit 367 records data indicating the selected local model in the recording unit 100 (step ST12). Furthermore, the model determination unit 367 also records the control condition data in the recording unit 100.

[0154] The learning device 300 of the first embodiment is configured as described above. When using data with variations collected from target equipment to learn a model for detecting abnormalities in the target equipment, the learning device 300 can reduce the time required for learning compared to conventional methods.

[0155] To supplement this point, for example, in existing devices, for normal data with deviations (learning data), normal data are clustered in a manner that includes all patterns. However, from a physical point of view, the operating conditions of the target equipment for abnormality detection are often limited. However, in existing devices, it is difficult to use normal data collected based on the limited operating conditions for learning, and as a result, there is a problem of increased man-hours spent on learning the regression model. In addition, in existing devices, when a large amount of control information (parameters) that determines the operating conditions is involved or when each control information takes a continuous value, it is considered that there are countless ways to divide the conditions, and there is a problem of consuming computational costs. In addition, in existing devices, by clustering normal data in a manner that includes all patterns, for example, even if Figure 20AEven if several regression models are constructed as shown in FIG, it is difficult to select an appropriate regression model from them. In addition, in the conventional apparatus, the evaluation of the deviation of the assumed normal data also differs depending on which regression model is selected. Therefore, in the conventional apparatus, for example, Figure 20B As shown, it is also difficult to evaluate whether the normal data within the box can truly be said to be deviant data.

[0156] In this regard, as described above, the learning device 300 of Embodiment 1 first constructs a globally learned model (global model) that has achieved physical validity through the global model construction unit 304. Next, the second variable selection unit 360 selects a second explanatory variable x2 that can separate target data considered to have deviations from the learning data. The local model construction unit 354 uses the learning data, after separating the target data based on the second explanatory variable x2 and the first explanatory variable x1, to construct a regression model (local model) appropriate for the learning data and the first explanatory variable x1. Thus, in the learning device 300, by selecting the second explanatory variable x2 that can separate target data considered to have deviations from the learning data, the probability of finding the defined operating conditions of the target device is increased, thereby reducing the time and computational cost required for learning compared to conventional devices. Furthermore, since the learning device 300 can separate target data considered to have deviations from the learning data, it facilitates the selection of an appropriate regression model and the evaluation of deviations in the learning data, which are difficult in conventional devices. Furthermore, it is possible to construct a regression model with high inference accuracy.

[0157] Next, an example of the hardware configuration of the learning device 300 according to the first embodiment will be described with reference to FIG14. 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 may be as follows: Figure 14A As shown, this is dedicated hardware, but it can also be Figure 14B As shown, a CPU (also called a Central Processing Unit, central processing device, processing device, arithmetic device, microprocessor, microcomputer, processor, or DSP (Digital Signal Processor)) 72 executes a program stored in a memory 73 .

[0158] When the processing circuit is dedicated hardware, processing circuit 71 may correspond to, for example, a single circuit, a complex circuit, a programmable processor, a parallel programmable processor, an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or a combination thereof. The functions of each component of global learning unit 301 and local learning unit 350 may be implemented separately by processing circuit 71, or the functions of each component may be implemented collectively by processing circuit 71.

[0159] When the processing circuit is a 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 in the form of programs and stored in the memory 73. The processing circuit reads and executes the program recorded in the memory 73 to implement the functions of each unit. In other words, the learning device 300 has a memory for storing a program, and the program is executed by the processing circuit. Figure 13 These programs can also be said to be programs that cause a computer to execute the steps and methods of the global learning unit 301 and the local learning unit 350. Here, the memory 73 includes, for example, RAM (Random Access Memory), ROM (Read Only Memory), flash memory, EPROM (Erasable Programmable ROM), EEPROM (Electrically Erasable Programmable Read-Only Memory), and other non-volatile or volatile semiconductor memories, magnetic disks, floppy disks, optical disks, high-density disks, minidiscs, or DVDs (Digital Versatile Discs).

[0160] Furthermore, the functions of the global learning unit 301 and the local learning unit 350 may be partially implemented by dedicated hardware and partially implemented by software or firmware. For example, the functions of the global learning unit 301 may be implemented by a dedicated hardware processing circuit, while the functions of the local learning unit 350 may be implemented by the processing circuit reading and executing a program stored in the memory 73.

[0161] In this way, the processing circuit can implement the above functions through hardware, software, firmware, or a combination thereof.

[0162] <State Inference Device 600>

[0163] Next, the state inference device 600 according to the first embodiment will be described. Figure 15 1 is a diagram showing a configuration example of a state inference device 600 according to Embodiment 1. Figure 15 As shown, the system includes an acquisition unit 601 , a data selection unit 602 , an evaluation unit 603 , and a feedback information generation unit 604 .

[0164] The state inference device 600 infers the state of the target device using the local model represented by the data (local model data) recorded in the recording unit 100 by the learning device 300, thereby detecting an abnormality in the target device. In the following description, an abnormality in the target device refers to degradation of the target device.

[0165] (Acquisition Unit 601)

[0166] The acquisition unit 601 acquires vibration data A1 from the vibration sensor 50 attached to the target device. This vibration data A1 is data representing the temporal variation of the vibration amplitude value of the target device, obtained from the target device by the vibration sensor 50 attached to the target device. Alternatively, the vibration data A1 may be data representing the temporal variation of a characteristic quantity of the vibration amplitude value. In this case, the characteristic quantity of the vibration amplitude value may be, for example, the RMS value of the vibration amplitude value. The following description uses the case where the vibration data is the RMS value of the vibration amplitude value as an example.

[0167] Furthermore, the acquisition unit 601 acquires the control information data B1 to Bn from the control information recording device 60. Here, the control information data B1 to Bn are as described above. Figure 3 As shown in the second through fourth graphs from the top, this data represents the temporal changes in control information, synchronized with the vibration data. Furthermore, n represents the number of control information items. Control information here refers to parameters that determine the operating conditions of the target device, such as the rotational speed of a rotating machine or the current value of the drive power for the rotating machine. Furthermore, control information recording device 60 is a dedicated device for recording control information data B1 through Bn.

[0168] The acquisition unit 601 outputs data obtained by integrating the acquired vibration data A1 and the control information data B1 to Bn as data D1 to the data selection unit 602 .

[0169] (Data Selection Unit 602)

[0170] The data selection unit 602 refers to the storage unit 100 and obtains the local model data MA1-MAn and the control condition data MB1-MBn from the storage unit 100. Here, n is the number of local models, and the local model data and control condition data correspond one-to-one. For ease of explanation, n is assumed to be 1.

[0171] The data selection unit 602 extracts the vibration data and control information data that satisfy the control conditions shown in the acquired control condition data MB1 from the vibration data and control information data contained in the above-mentioned data D1, and outputs the data obtained by summarizing the extracted data and the local model data MA1 and the control condition data MB1 as data D2 to the evaluation unit 603.

[0172] (Evaluation Department 603)

[0173] The evaluation unit 603 is as follows, for example Figure 16 As shown, the system includes a degradation degree calculation unit 631 , a parameter adjustment unit 632 , and an image output unit 633 .

[0174] (Degradation Degree Calculation Unit 631)

[0175] Degradation degree calculator 631 obtains data D2 from data selection unit 602. It analyzes this data D2 and calculates the degree of degradation of the target device. Specifically, it inputs an arbitrary value (e.g., rotational speed = 500) of control information data B1 included in data D2 into the local model indicated by local model data MA1 included in data D2. Based on the input value, the local model outputs vibration data corresponding to that value (e.g., RMS value = 1.5).

[0176] The degradation degree calculation unit 631 then compares the vibration data output from the local model with the vibration data corresponding to the aforementioned arbitrary value contained in the data D2 and calculates the error between the two. The degradation degree calculation unit 631 then calculates the degradation degree of the target device by comparing the calculated error with a predetermined threshold. Alternatively, the degradation degree calculation unit 631 can calculate the degradation degree using, for example, the mean absolute error (MAPE) or T2 Hotelling. The degradation degree calculation unit 631 outputs data representing the calculated degradation degree as a state descriptor to the image output unit 633.

[0177] (Image Output Unit 633)

[0178] The image output unit 633 obtains the state descriptor from the degradation degree calculation unit 631. In addition, the image output unit 633 obtains the above-mentioned data D2 from the data selection unit 602. Then, the image output unit 633 uses the obtained data to generate, for example, Figure 17The data of the comparison images are shown. Figure 17 In the figure, the left side shows an image representing the distribution of vibration data (learning data) used when constructing the local model, and the right side shows an image representing the distribution of vibration data obtained by inputting control information data B1 actually obtained from the target device into the local model. This image allows the user to easily understand the extent to which the distribution of vibration data obtained by inputting control information data B1 actually obtained from the target device into the local model deviates from the distribution of vibration data (learning data) used when constructing the local model. Image output unit 633 outputs the generated data representing the comparison image to third external evaluation device 700.

[0179] The third external evaluation device 700 obtains data representing the comparison image from the image output unit 633. Based on the obtained data, the third external evaluation device 700 Figure 17 The comparative image is displayed on a display unit (not shown). The user confirms the comparative image displayed on the display unit and adjusts parameters using the third external evaluation device 700 as needed.

[0180] For example, Figure 17 As shown in , the distribution of the two mentioned above may have a slight difference due to accidental reasons. In this case, the degradation degree may change due to a slight difference caused by accidental reasons unrelated to the degradation. Figure 17 In the example, although the target device is not actually deteriorating that seriously, it is calculated as "deterioration degree 18%." Therefore, the user adjusts the parameters to compensate for this slight difference.

[0181] For example, user adjustments Figure 17 The position of the prediction line 1701 obtained by the local model shown on the left side of , and the position of the line 1702 indicating the boundary of the confidence interval set for the prediction line 1701. In this case, the user visually confirms the position of the prediction line 1701, for example. Figure 17 The position of each line is adjusted by calculating the difference between the distribution diagrams on the left and right sides, or the position of each line is adjusted by calculating the difference between the average values ​​of the vibration data on the left and right sides.

[0182] Alternatively, the user adjusts Figure 17 The distance between the prediction line 1701 obtained by the local model shown on the left side of FIG and the line 1702 indicating the boundary of the confidence interval set for the prediction line 1701. In this case, the user can also visually confirm the distance between the prediction line 1701 and the confidence interval set for the prediction line 1701. Figure 17 The interval is adjusted by adjusting the ratio of the deviation of the vibration data in the left and right distribution diagrams, or by multiplying the standard deviation value of the left and right vibration data.

[0183] The third external evaluation device 700 outputs data representing the adjustment content input by the user as an adjustment descriptor D4 to the parameter adjustment unit 632. The adjustment descriptor D4 includes the regression model data of the adjustment object, control condition data, data required for model adjustment (specifically, correction values ​​of the regression coefficients), etc. In particular, the data required for model adjustment is also called a parameter adjuster. In addition, the adjustment descriptor D4 also includes a determinator input by the user that determines whether to output the adjustment descriptor D4 to the feedback information generation unit 604. For example, if the determinator is 1, it indicates that the adjustment descriptor D4 is output to the feedback information generation unit 604. If the determinator is 0, it indicates that the adjustment descriptor D4 is not output to the feedback information generation unit 604.

[0184] In addition, regarding the parameter adjustment symbol in the adjustment descriptor D4, for example, when the user visually Figure 17 In the case of adjusting the distribution diagram on the left and right sides, the user input is required, but in other cases (for example, calculating the difference between the average values ​​of the left and right vibration data to adjust the position of each line), for example, the parameter adjustment unit 632 can automatically perform the calculation, so the user input is not required.

[0185] (Parameter Adjustment Unit 632)

[0186] The parameter adjustment unit 632 obtains the adjusted descriptor D4 from the third external evaluation device 700. The parameter adjustment unit 632 outputs the obtained adjusted descriptor D4 to the degradation degree calculation unit 631 and instructs the degradation degree calculation unit 631 to adjust the local model based on the adjusted descriptor D4. In response to this instruction, the degradation degree calculation unit 631 adjusts the local model and recalculates the degradation degree using the adjusted local model according to the above steps. Furthermore, the degradation degree calculation unit 631 outputs data representing the recalculated degradation degree 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 above process.

[0187] When the parameter adjustment unit 632 no longer obtains the adjustment descriptor D4 from the third external evaluation apparatus 700 in the above repetition, it instructs the image output unit 633 to display the final calculation result of the degree of degradation on the display unit of the third external evaluation apparatus 700 .

[0188] Furthermore, the parameter adjustment unit 632 checks the content of the determinant included in the obtained adjustment descriptor D4. If the content of the determinant indicates that the adjustment descriptor D4 should be output to the feedback information generator 604, the parameter adjustment unit 632 outputs the adjustment descriptor D4 as data D5 to the feedback information generator 604. On the other hand, if the content of the determinant indicates that the adjustment descriptor D4 should not be output to the feedback information generator 604, the parameter adjustment unit 632 does not output the adjustment descriptor D4 to the feedback information generator 604.

[0189] (Feedback Information Generating Unit 604)

[0190] Feedback information generation unit 604 obtains data D5 from parameter adjustment unit 632. Based on the obtained data D5, feedback information generation unit 604 generates feedback information D6 and records the generated feedback information D6 in recording unit 100. Similar to adjustment descriptor D4, feedback information D6 includes the regression model data to be adjusted, control condition data, and data required for model adjustment (specifically, correction values ​​of the regression coefficients). Furthermore, feedback information generation unit 604 records feedback information D6 in recording unit 100 as separate information from the local model data MA1 and control condition data MB1 already recorded in recording unit 100.

[0191] The user can then appropriately reflect the feedback information D6 recorded in the recording unit 100 on the local model data MA1 and the control condition data MB1. Thus, by updating the local model data MA1 and the control condition data MB1 based on the feedback information D6, the possibility of false detection caused by unknown differences between the data actually acquired from the target device and the learning data used when constructing the second regression model (e.g., differences caused by the aforementioned accidental factors) can be reduced.

[0192] Next, refer to Figure 18 An example of the operation of the state inference device 600 according to the first embodiment is described with reference to the flowchart shown.

[0193] First, the acquisition unit 601 receives vibration data from the vibration sensor 50 provided in the target device. Furthermore, the acquisition unit 601 acquires control information data from the control information recording device 60 (step ST21).

[0194] Next, the data selection unit 602 obtains the local model data and control condition data from the recording unit 100, and extracts the vibration data and control information data that satisfy the control conditions indicated by the obtained control condition data from the vibration data and control information data obtained in step ST21 (step ST22).

[0195] Next, the degradation degree calculation unit 631 calculates the degradation degree of the target device using the data extracted in step ST22 (step ST23 ).

[0196] Next, the image output unit 633 uses the calculation result in step ST23 to generate, for example, Figure 17 The image output unit 633 outputs the generated data representing the comparative image to the third external evaluation device 700 (step ST24 ).

[0197] Next, the parameter adjustment unit 632 determines whether the adjustment descriptor D4 has been obtained from the third external evaluation device 700 (step ST25). If the parameter adjustment unit 632 determines that the adjustment descriptor D4 has been obtained from the third external evaluation device 700 (step ST25: Yes), the parameter adjustment unit 632 outputs the adjustment descriptor D4 to the degradation degree calculation unit 631 and instructs the degradation degree calculation unit 631 to adjust the local model based on the adjustment descriptor D4. The degradation degree calculation unit 631 adjusts the local model based on the adjustment descriptor D4 (step ST26). The process then returns to step ST23.

[0198] On the other hand, when the parameter adjustment unit 632 determines that the adjustment descriptor D4 has not been acquired from the third external evaluation device 700 (step ST25 ; No), the process proceeds to step ST26 .

[0199] In step ST26, the parameter adjustment unit 632 determines whether the adjustment descriptor D4 has been obtained at least once from the third external evaluation device 700 (step ST26). If the parameter adjustment unit 632 determines that the adjustment descriptor D4 has not been obtained from the third external evaluation device 700 (step ST26: No), the process proceeds to step ST29.

[0200] On the other hand, if the parameter adjustment unit 632 determines that the adjustment descriptor D4 has been obtained at least once from the third external evaluation device 700 (step ST26: Yes), it checks the content of the determinant included in the most recently obtained adjustment descriptor D4. Then, if the content of the determinant indicates that the adjustment descriptor D4 should be output to the feedback information generation unit 604, the most recently obtained adjustment descriptor D4 is output to the feedback information generation unit 604 as data D5.

[0201] The feedback information generator 604 generates feedback information D6 based on the data D5 obtained from the parameter adjuster 632 (step ST27 ), and records the generated feedback information D6 in the recording unit 100 (step ST28 ). The process then proceeds to step ST29 .

[0202] In step ST29 , the image output unit 633 generates data indicating the final calculation result of the degradation degree of the target device and outputs the generated data to the third external evaluation apparatus 700 , thereby causing the display unit to display the final calculation result of the degradation degree of the target device (step ST29 ).

[0203] Next, an example of the hardware configuration of the state inference device 600 according to the first embodiment will be described with reference to FIG19. The functions of the acquisition unit 601, the data selection unit 602, the evaluation unit 603, and the feedback information generation unit 604 in the state inference device 600 are implemented by a processing circuit. The processing circuit can be as follows: Figure 19A As shown, this is dedicated hardware, but it can also be Figure 19B As shown, a CPU (also referred to as a Central Processing Unit, central processing device, processing device, arithmetic device, microprocessor, microcomputer, processor, or DSP (Digital Signal Processor)) 82 executes programs stored in a memory 83 .

[0204] When the processing circuit is dedicated hardware, the processing circuit 81 may correspond to, for example, a single circuit, a complex circuit, a programmable processor, a parallel programmable processor, an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or a combination thereof. The processing circuit 81 may separately implement the functions of the acquisition unit 601, the data selection unit 602, the evaluation unit 603, and the feedback information generation unit 604, or the processing circuit 81 may implement the functions of each unit collectively.

[0205] When the processing circuit is a CPU 82, the functions of the acquisition 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 in the form of programs and stored in the memory 83. The processing circuit reads and executes the program recorded in the memory 83 to implement the functions of each unit. That is, the state inference device 600 has a memory for storing a program, and the program is executed by the processing circuit. For example, Figure 18These programs can also be said to be programs that cause a computer to execute the steps and methods of the acquisition unit 601, the data selection unit 602, the evaluation unit 603, and the feedback information generation unit 604. Here, the memory 83 includes, for example, RAM (Random Access Memory), ROM (Read Only Memory), flash memory, EPROM (Erasable Programmable ROM), EEPROM (Electrically Erasable Programmable Read-Only Memory), and other non-volatile or volatile semiconductor memories, magnetic disks, floppy disks, optical disks, high-density disks, minidiscs, or DVDs (Digital Versatile Discs).

[0206] Furthermore, the functions of the acquisition unit 601, data selection unit 602, evaluation unit 603, and feedback information generation unit 604 may be partially implemented by dedicated hardware and partially implemented by software or firmware. For example, the acquisition unit 601 may be implemented by a dedicated hardware processing circuit, while the data selection unit 602, evaluation unit 603, and feedback information generation unit 604 may be implemented by the processing circuit reading and executing a program stored in the memory 83.

[0207] In this way, the processing circuit can implement the above functions through hardware, software, firmware, or a combination thereof.

[0208] As described above, according to embodiment 1, the learning device 300 includes: a global model construction unit 304, which constructs a first regression model (global model) suitable for the learning data and the first explanatory variable x1 based on the learning data that can be explained by multiple explanatory variables and the first explanatory variable x1 specified from the outside as one of the multiple explanatory variables; a second variable selection unit 360, which selects the second explanatory variable x2 from the multiple explanatory variables, and through the second explanatory variable x2, the object data in the learning data that is considered to have a deviation based on the first regression model constructed by the global model construction unit 304 can be separated from the learning data; and a local model construction unit 354, which uses the learning data after the object data is separated based on the second explanatory variable x2 and the first explanatory variable x1 selected by the second variable selection unit 360 to construct a second regression model (local model) suitable for the learning data and the first explanatory variable x1. Thus, when the learning device 300 according to the first embodiment uses data with variations collected from a target device to learn a model for detecting an abnormality in the target device, the number of hours required for learning can be reduced compared to conventional methods.

[0209] Furthermore, the second variable selection unit 360 includes a filtering unit 351 that classifies the learning data into target data (Data A) considered to have a bias and non-target data (Data B) considered to have no bias, based on the first regression model constructed by the global model construction unit 304; a range selection unit 352 that selects a predetermined range from the range that the first explanatory variable x1 can take, based on the target data and non-target data classified by the filtering unit 351; and a second variable selection unit 353 that selects the second explanatory variable x2 using the learning data included in the predetermined range selected by the range selection unit 352. Thus, the learning device 300 according to the first embodiment can appropriately select the second explanatory variable x2 based on the first regression model and the learning data.

[0210] Furthermore, the filtering unit 351 treats as target data the learning data that is outside a predetermined confidence interval set for the prediction line, which is centered on the prediction line, and treats as non-target data the learning data that is inside the predetermined confidence interval centered on the prediction line, where the prediction line is obtained based on the first regression model constructed by the global model construction unit 304. Thus, the learning device 300 according to the first embodiment can easily classify the learning data into target data (Data A) that is considered to have a deviation and non-target data (Data B) that is considered to have no deviation.

[0211] Furthermore, the range selection unit 352 includes a distribution calculation unit 361 that calculates a probability distribution for each of the target data and non-target data, indicating how often the target data and non-target data, as classified by the filtering unit 351, appear with respect to the first explanatory variable x1; and a distribution difference comparison unit 362 that calculates the 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 selects a range of the first explanatory variable x1 for which the calculated difference is greater than or equal to a predetermined value as the predetermined range. Thus, the learning device 300 according to the first embodiment can easily select the predetermined range of the first explanatory variable x1 used to select the second explanatory variable x2.

[0212] Furthermore, distribution difference comparison unit 362 selects a predetermined range from an externally received search width S, which represents a range within the range of the first explanatory variable x1 where the proportion of non-target data is assumed to be relatively high. Thus, learning device 300 according to Embodiment 1 can appropriately select the predetermined range for the first explanatory variable x1 based on the externally received search width S.

[0213] Furthermore, the second variable selection processing unit 353 generates a probability distribution indicating how often the learning data included in the specified range selected by the range selection unit 352 appear relative to a certain explanatory variable. In this generated probability distribution, when the range of the first explanatory variable x1 in which the ratio of target data to the number of learning data is greater than or equal to a specified value is defined as a first range Y, and the range of the first explanatory variable x1 excluding the first range Y is defined as a second range X, the explanatory variable in which the ratio of non-target data to the learning data included in the second range X is greater than or equal to the specified value is selected as the second explanatory variable x2. Thus, the learning device 300 according to the first embodiment can appropriately and efficiently select the second explanatory variable x2.

[0214] Furthermore, the local model construction unit 354 includes a region evaluation unit 363 that generates data representing an image showing regions where target data appears and regions where non-target data appears, within a region determined by the combination of the second explanatory variable x2 and the first explanatory variable x1 selected by the second variable selection unit 360; and a model construction unit 364 that receives an externally specified region within the region determined by the combination of the first explanatory variable x1 and the second explanatory variable x2, based on the image represented by the data generated by the region evaluation unit 363, and constructs a second regression model using learning data included in the received region. Thus, the learning device 300 according to the first embodiment can construct a second regression model that reflects the intention of an external user (e.g., a user).

[0215] Furthermore, the learning device 300 includes a model evaluation unit 305 that receives external evaluations of the first regression model constructed by the global model construction unit 304, and a model evaluation unit 355 that receives external evaluations of the second regression model constructed by the local model construction unit 354. Thus, the learning device 300 according to the first embodiment can obtain external (e.g., user) evaluations of the first and second regression models.

[0216] Furthermore, the global model construction unit 304 includes a model updating unit 312. When the evaluation received by the model evaluation unit 355 indicates that the desired second regression model does not exist, the model updating unit 312 reconstructs a first regression model appropriate for the learning data and a new first explanatory variable x1 specified externally as one of the plurality of explanatory variables. Thus, the learning device 300 of Embodiment 1 can reconstruct the desired second regression model from the first regression model if the desired second regression model cannot be constructed.

[0217] Furthermore, according to the first embodiment, the state inference device 600 uses the second regression model constructed by the local model construction unit 354 of the learning device 300, and the data corresponding to the learning data and the data corresponding to the first explanatory variable obtained from the target device to infer the state of the target device. The learning device 300 includes: a global model construction unit 304 that constructs a first regression model (global model) suitable for the learning data and the first explanatory variable x1 based on the learning data that can be explained by a plurality of explanatory variables and the first explanatory variable x1 that is one of the plurality of explanatory variables specified from the outside. The first model is constructed using the following components: a global model (a global model); a second variable selection unit 360 that selects a second explanatory variable x2 from a plurality of explanatory variables, using which the second explanatory variable x2 can be used to separate target data in the learning data that is considered to have deviations based on the first regression model constructed by the global model construction unit 304; and a local model construction unit 354 that uses the learning data after separating the target data based on the second explanatory variable x2 selected by the second variable selection unit 360 and the first explanatory variable x1 to construct a second regression model (local model) appropriate for the learning data and the first explanatory variable x1. Thus, the state inference device 600 of the first embodiment can infer the state of the target device with high accuracy.

[0218] Furthermore, the state inference device 600 includes a feedback information generation unit 604 that corrects the regression coefficients in the second regression model based on a correction value received from the outside for correcting the regression coefficients in the second regression model. Thus, the state inference device 600 of Embodiment 1 can reduce the possibility of false detections caused by unknown differences between data actually acquired from the target device and the learning data used when constructing the second regression model at the time of construction.

[0219] Furthermore, according to the first embodiment, the state monitoring system 1000 includes a learning device 300 and a state inference device 600. The learning device 300 includes: a global model construction unit 304 for constructing a first regression model (global model) suitable for the learning data and the first explanatory variable x1 based on the learning data that can be explained by a plurality of explanatory variables and a first explanatory variable x1 that is specified externally as one of the plurality of explanatory variables; a second variable selection unit 360 for selecting a second explanatory variable x2 from the plurality of explanatory variables, by which the second explanatory variable x2 in the learning data can be converted into the state inference device 600. 4, separates the target data deemed to be deviated from the learning data using the first regression model constructed by the local model construction unit 354; and a local model construction unit 354 constructs a second regression model (local model) suitable for the relationship between the learning data and the first explanatory variable x1 using the learning data after separating the target data based on the second explanatory variable x2 and the first explanatory variable x1 selected by the second variable selection unit 360. The state inference device 600 infers the state of the target device using the second regression model constructed by the local model construction unit 354 and data corresponding to the learning data and data corresponding to the first explanatory variable x1 acquired from the target device. Thus, the state monitoring system 1000 according to Embodiment 1 can reduce the time required for learning compared to conventional methods when using deviated data collected from the target device to learn a model for detecting abnormalities in the target device, and can also use this model to infer the state of the target device with high accuracy.

[0220] Finally, a preferred application example of the learning device 300 and the state inference device 600 of embodiment 1 is described. The learning device 300 of embodiment 1 is suitable for use in a monitoring system for an electric motor mounted on a railway vehicle, for example. In an electric motor mounted on a railway vehicle, a large amount of control information such as the motor's braking information, speed information, current information, and voltage information exists simultaneously with the vibration data. In the case of building a system that monitors vibration data reflecting the state of the electric motor based on control information, in order to utilize the user's knowledge (for example, knowledge that vibration is closely related to speed and that high-frequency vibration characteristics are likely to appear at low speeds), first, a model (global model) with speed as an explanatory variable is constructed by the global model construction unit 304. Then, in order to improve the accuracy of the model, a model (local model) is constructed by specifying the speed region by the user and filtering the conditions using other control information that can exclude data that deviates from the global model. Thus, in the monitoring system, the user's knowledge is incorporated into the construction of the model, reducing the process of model construction and evaluation using explanatory variables and conditions that are considered unnecessary for degradation detection, and enabling efficient model construction.

[0221] Furthermore, similar to the learning device 300, the state inference device 600 of the first embodiment is also suitable for use in a monitoring system for electric motors installed in railway vehicles. For example, the state inference device 600 of the first embodiment is further configured to include an alarm device. When the state inference device 600 determines, based on vibration data obtained from the vibration sensor 50 attached to the target device, that the target device is not the same as the object designated for monitoring, an alarm is output to the user of the monitoring system. The state inference device 600 of the first embodiment can be applied to monitoring systems in this manner.

[0222] Furthermore, the state monitoring system 1000 according to the first embodiment is also suitable for use in a monitoring system for an electric motor mounted on a railway vehicle, for example, similarly to the learning device 300 and the state inference device 600 .

[0223] Furthermore, the present disclosure allows for modifications to any of the structural elements of the embodiments, or for any of the structural elements of the embodiments to be omitted. For example, the above description uses an example where the learning data serving as the target variable is vibration data, and the explanatory variable describing the target variable is control information data. However, the learning data serving as the target variable and the explanatory variable are not limited to the above examples; any type of data may be used as long as the explanatory variable describes the target variable.

[0224] In the above description, the recording unit 100 is described as being provided separately from the learning device 300 and the state inference device 600. However, the recording unit 100 is not limited thereto and may be built into either the learning device 300 or the state inference device 600.

[0225] Alternatively, the recording unit 100 may be built into any one of the first external evaluation device 400 , the second external evaluation device 500 , and the third external evaluation device 700 .

[0226] In the above description, an example is described in which the first external evaluation device 400, the second external evaluation device 500, and the third external evaluation device 700 are each provided separately. However, the devices are not limited to this, and the functions of the devices may be integrated into any one device, or the functions of any two devices may be integrated into a single device.

[0227] Description of Reference Numerals

[0228] 50 vibration sensor, 60 control information recording device, 71 processing circuit, 72 CPU, 73 memory, 81 processing circuit, 82 CPU, 83 memory, 100 recording unit, 200 learning data recording unit, 300 learning device, 301 global learning unit, 302 data extraction unit, 303 explanatory variable acquisition unit, 304 global model construction unit, 305 model evaluation unit (first model evaluation unit), 311 model construction unit, 312 model updating 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 construction unit, 355 model evaluation unit (second model evaluation unit), 360 second variable selection unit (variable selection unit), 361 distribution calculation , 362 distribution difference comparison unit, 363 area evaluation unit, 364 model construction 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 prediction line, 502 line indicating the boundary of the trust interval, 600 state inference device, 601 acquisition unit, 602 data selection unit, 603 evaluation unit, 604 feedback information generation unit, 631 degradation degree calculation unit, 632 parameter adjustment unit, 633 image output unit, 700 third external evaluation device, 1000 state monitoring system, 1701 prediction line, 1702 line indicating the boundary of the trust interval, A1 vibration data, B1 control information data, 210 vibration DB, 220 control information DB, U1 area, U2 area, U3 area.

Claims

1. A learning device, wherein: The learning device comprises: a global model construction unit for constructing a first regression model suitable for the learning data and the first explanatory variable based on learning data that can be explained by a plurality of explanatory variables and a first explanatory variable that is externally specified and is one of the plurality of explanatory variables; a variable selection unit configured to select a second explanatory variable from the plurality of explanatory variables, the second explanatory variable being capable of separating target data in the learning data that is considered to have a deviation based on the first regression model constructed by the global model construction unit from the learning data; as well as The local model construction unit constructs a second regression model suitable for the learning data and the first explanatory variable using the learning data obtained by separating the target data based on the second explanatory variable selected by the variable selection unit and the first explanatory variable.

2. The learning device according to claim 1, wherein The variable selection unit includes: a filtering processing unit that classifies the learning data into the target data considered to have a bias and non-target data considered to have no bias, based on the first regression model constructed by the global model construction unit; a range selection unit that selects a predetermined range from a range that the first explanatory variable can take based on the target data and the non-target data classified by the filtering processing unit; as well as The variable selection processing unit selects the second explanatory variable using the learning data included in the predetermined range selected by the range selection unit.

3. The learning device according to claim 2, wherein: The filtering processing unit uses the learning data located outside the prescribed confidence interval set for the prediction line with the prediction line as the object data, and uses the learning data located inside the prescribed confidence interval with the prediction line as the non-object data, wherein the prediction line is obtained based on the first regression model constructed by the global model construction unit.

4. The learning device according to claim 2 or 3, characterized in that The range selection unit includes: a distribution calculation unit that calculates a probability distribution for each of the target data and the non-target data, the probability distribution indicating how often the target data and the non-target data classified by the filtering unit appear with respect to the first explanatory variable; as well as A distribution difference comparison unit calculates the difference between the probability distribution of the object data calculated by the distribution calculation unit and the probability distribution of the non-object data calculated by the distribution calculation unit, and selects a range of the first explanatory variable in which the calculated difference is greater than a specified value as the specified range.

5. The learning device according to claim 4, characterized in that The distribution difference comparing unit selects the predetermined range from a search width received from the outside, the search width indicating a range in the range of the first explanatory variable where a ratio of the non-target data being assumed to exist is relatively high.

6. The learning device according to any one of claims 2 to 5, characterized in that The variable selection processing unit generates the following probability distribution, which indicates how often the learning data included in the specified range selected by the range selection unit appears relative to a certain explanatory variable. In the generated probability distribution, when the range of the first explanatory variable in which the proportion of the object data relative to the number of the learning data is greater than the specified value is set as the first range, and the range of the first explanatory variable other than the first range is set as the second range, the explanatory variable in which the proportion of the non-object data in the learning data included in the second range is greater than the specified value is selected as the second explanatory variable.

7. The learning device according to any one of claims 1 to 6, characterized in that The local model building unit includes: a region evaluation unit that generates data representing an image showing a region where the target data considered to have a deviation appears and a region where non-target data considered to have no deviation appears, in a region determined by a combination of the second explanatory variable selected by the variable selection unit and the first explanatory variable; as well as A model building unit receives an area specified externally based on an image represented by data generated by the area evaluation unit in an area determined by a combination of the first explanatory variable and the second explanatory variable, and uses the learning data included in the received area to build the second regression model.

8. The learning device according to any one of claims 1 to 7, characterized in that The learning device comprises: a first model evaluation unit that receives external evaluation of the first regression model constructed by the global model construction unit; and The second model evaluation unit receives external evaluation of the second regression model constructed by the local model construction unit.

9. The learning device according to claim 8, characterized in that The global model construction unit includes a model updating unit that, when the evaluation received by the second model evaluation unit indicates that the desired second regression model does not exist, reconstructs a first regression model suitable for the learning data and the new first explanatory variable specified externally as one of the multiple explanatory variables.

10. A state inference device, wherein: The state inference device infers the state of the target device using the second regression model constructed by the local model construction unit of the learning device, data corresponding to the learning data and data corresponding to the first explanatory variable obtained from the target device, The learning device comprises: a global model construction unit that constructs a first regression model suitable for the learning data and the first explanatory variable based on the learning data that can be explained by a plurality of explanatory variables and the first explanatory variable that is externally specified as one of the plurality of explanatory variables; a variable selection unit that selects a second explanatory variable from the plurality of explanatory variables, the second explanatory variable being capable of separating target data in the learning data that is considered to have a deviation based on the first regression model constructed by the global model construction unit from the learning data; and The local model construction unit uses the learning data obtained by separating the target data based on the second explanatory variable selected by the variable selection unit and the first explanatory variable to construct a second regression model suitable for the learning data and the first explanatory variable.

11. The state inference device according to claim 10, characterized in that: The state inference device includes a feedback information generating unit that corrects the regression coefficient in the second regression model based on a correction value for correcting the regression coefficient in the second regression model received from the outside.

12. A condition monitoring system, wherein: The state monitoring system includes a learning device and a state inference device, The learning device comprises: a global model construction unit for constructing a first regression model suitable for the learning data and the first explanatory variable based on learning data that can be explained by a plurality of explanatory variables and a first explanatory variable that is externally specified and is one of the plurality of explanatory variables; a variable selection unit configured to select a second explanatory variable from the plurality of explanatory variables, the second explanatory variable being capable of separating target data in the learning data that is considered to have a deviation based on the first regression model constructed by the global model construction unit from the learning data; as well as a local model construction unit that uses the learning data obtained by separating the target data based on the second explanatory variable selected by the variable selection unit and the first explanatory variable to construct a second regression model suitable for the learning data and the first explanatory variable; The state inference device infers the state of the target device using the second regression model constructed by the local model construction unit, and data corresponding to the learning data and data corresponding to the first explanatory variable acquired from the target device.

13. A learning method, which is a learning method based on a learning device, wherein: The learning method has the following steps: The global model construction unit constructs a first regression model suitable for the learning data and the first explanatory variable based on learning data that can be explained by a plurality of explanatory variables and a first explanatory variable that is externally specified and is one of the plurality of explanatory variables; a variable selection unit selecting a second explanatory variable from the plurality of explanatory variables, the second explanatory variable being capable of separating target data in the learning data that is considered to have a deviation based on the first regression model constructed by the global model construction unit from the learning data; as well as The local model construction unit constructs a second regression model suitable for the learning data and the first explanatory variable using the learning data from which the target data is separated based on the second explanatory variable selected by the variable selection unit and the first explanatory variable.

Images