Label creation device, diagnostic model creation device, degradation diagnostic device, information processing system, and label creation method

Abstract

The present invention includes: a training data acquisition unit (11) that acquires training data in which feature amounts and external factor data are associated with each other; an external factor area division unit (12) that divides a range of values of external factor data in the axial direction of the external factor data into a constant band and a transition band; and an external factor label data creation unit (13) that assigns, to a range of values of the external factor data divided into the constant band with respect to the axis of the external factor data, an external factor label indicating that the relationship between the external factor data in the range and the corresponding feature amounts is constant, and assigns, to a range of values of the external factor data divided into the transition band with respect to the axis of the external factor data, an external factor label indicating that the relationship between the external factor data in the range and the corresponding feature amounts frequently changes, and creates external factor label data in which the ranges of values of the external factor data and the external factor labels are associated with each other.

Description

Label creation device, diagnostic model creation device, degradation diagnosis device, information processing system, and label creation method

[0001] The present disclosure relates to a label creation device, a diagnostic model creation device, a degradation diagnosis device, an information processing system, and a label creation method that support the construction of a diagnostic model used in the degradation diagnosis of equipment.

[0002] Conventionally, there has been known a technique for diagnosing deterioration of equipment using a model (hereinafter referred to as a "diagnostic model") that has been created based on feature quantities extracted from data acquired from the equipment. As a technique for diagnosing the operating state of equipment using a diagnostic model, for example, Patent Document 1 discloses a diagnostic system that manages equipment with similar operating conditions as a group under a common equipment feature pattern number, and diagnoses the equipment state reflecting the features of the equipment using a diagnostic pattern that matches the equipment feature pattern number.

[0003] JP 2016-91378 A

[0004] The feature quantities used in equipment degradation diagnosis vary not only depending on the degradation state of the equipment but also due to the operating environment. Therefore, the diagnostic model used in equipment degradation diagnosis must be constructed as a model capable of diagnosing a normal state or a degraded state, taking into account the influence of the operating environment. However, the indicators of the operating environment are likely to be continuous values. At the boundary where the diagnostic model switches depending on the operating environment, accurate diagnostic results may not be obtained regardless of whether a diagnostic model is used before or after the switch. Conventional technologies have had a problem in that it is difficult to construct a diagnostic model that takes this into account. As a result, the accuracy of the diagnostic results of equipment degradation diagnosis performed using a diagnostic model based on feature quantities may be reduced. Note that the indicators of the equipment's operating conditions are unlikely to be continuous values. In other words, the accuracy of the diagnostic results is unlikely to decrease at the boundary where the diagnostic model switches depending on the equipment's operating conditions. The technology disclosed in Patent Document 1, which diagnoses the equipment state using a diagnostic pattern corresponding to an equipment feature pattern, does not consider the influence of the operating environment on the diagnostic results, and the above-mentioned problem remains unresolved.

[0005] The present disclosure has been made to solve the above-mentioned problems, and aims to provide a label creation device that can provide data that will help build a diagnostic model that prevents a decrease in the accuracy of equipment deterioration diagnosis due to the operating environment.

[0006] The label creation device according to the present disclosure includes: a learning data acquisition unit that acquires learning data in which feature quantities that can be used to create a diagnostic model for diagnosing deterioration of a target device that is the subject of deterioration diagnosis correspond to external factor data related to the operating environment; an external factor region division unit that, based on the learning data acquired by the learning data acquisition unit, divides a value range of external factor data in an axial direction of the external factor data into a constant band, which is a range of values ​​of external factor data in which the relationship with the corresponding feature quantities is estimated to be constant, and a transition band, which is a range of values ​​of external factor data in which the relationship with the corresponding feature quantities is estimated to change frequently; and an external factor label data creation unit that assigns, on the axis of the external factor data, an external factor label to the value range of the external factor data divided into the constant band, indicating that the relationship between the external factor data in that range and the corresponding feature quantities is constant, and assigns an external factor label to the value range of the external factor data divided into the transition band, indicating that the relationship between the external factor data in that range and the corresponding feature quantities changes frequently, thereby creating external factor label data in which the value range of the external factor data corresponds to the external factor labels.

[0007] According to the present disclosure, as configured above, it is possible to provide data that aids in the construction of a diagnostic model that prevents a decrease in the accuracy of equipment degradation diagnosis due to the operating environment. As a result, it is possible to construct a diagnostic model that prevents a decrease in the accuracy of the diagnosis results of equipment degradation diagnosis performed using a diagnostic model based on feature quantities, and to realize accurate degradation diagnosis using the diagnostic model.

[0008] 8 is a diagram showing an example of the configuration of an information processing system including a label creation device according to a first embodiment. FIG. 9 is a diagram showing an example of the configuration of a diagnostic model creation device according to a first embodiment, including the label creation device according to the first embodiment. FIG. 10 is a diagram for explaining a method in which an external factor region division unit divides a range of values ​​of external factor data in an axial direction of the external factor data into a constant band and a transition band. FIG. 11 is a diagram for explaining an example of an external factor label assigned to an axis of the external factor data by an external factor label data creation unit in the first embodiment. FIG. 12 is a diagram for explaining an example of the content of external factor label data created by the external factor label data creation unit in the first embodiment. FIG. 13 is a diagram showing an example of the configuration of a degradation diagnosis device according to the first embodiment. FIG. 14 is a diagram for explaining an example of a method in which a diagnosis unit determines, from among diagnostic models output from a model acquisition unit, a diagnostic model corresponding to a feature extracted by a feature extraction unit. FIG. 15 is a flowchart for explaining the operation of the diagnostic model creation device according to the first embodiment, including the label creation device according to the first embodiment. FIG. 16 is a flowchart for explaining details of the region division process by the external factor region division unit in step ST3 of FIG. 8. FIG. 17 is a diagram for explaining problems in the prior art. 1 is a flowchart for explaining the operation of the degradation diagnosis device according to embodiment 1. FIG. 2 is a diagram for explaining an image of an example of an external factor label that an external factor label data creation unit in the label creation device assigns to an axis of external factor data when the label creation device creates external factor data for a pair of "temperature" and "amplitude value" and a pair of "temperature" and "variance" in embodiment 1. FIG. 3 is a flowchart for explaining the operation of the abnormality factor estimating device according to embodiment 1.

[0009] In order to explain the present disclosure in more detail, embodiments of the present disclosure will be described below with reference to the accompanying drawings.

[0010] First Embodiment. A label creation device according to the first embodiment provides data (hereinafter referred to as "external factor label data") that aids in the construction of a trained model (hereinafter referred to as a "diagnostic model") used in a deterioration diagnosis of a target device (hereinafter referred to as a "target device") for which a deterioration diagnosis is to be performed. The target device may be, for example, a railway platform door. Note that this is merely an example, and various devices whose deterioration is related to the surrounding environment when the target device is operating, in other words, the operating environment, may be the target device. Deterioration diagnosis of the device is performed using a diagnostic model based on feature quantities extracted from data collected from the device. These feature quantities are affected by the operating environment. Therefore, if the diagnostic model is not constructed taking into account the influence of the feature quantities on the operating environment, changes in the feature quantities caused by changes in the operating environment may result in a false diagnosis of deterioration even when there is no deterioration. The label creation device according to the first embodiment prevents such false diagnosis. More specifically, the label creation device according to the first embodiment provides external factor label data that aids in the construction of a diagnostic model that can prevent the above-described false diagnosis. In the first embodiment, the operating environment is assumed to be natural factors that cannot be controlled by humans, such as temperature, precipitation, etc. In the first embodiment below, the operating environment that can affect the deterioration of the target device as described above is also referred to as an "external factor."

[0011] The external factor label data provided by the label creation device indicates the relationship between the feature quantities used in the deterioration diagnosis of the target equipment and the external factors. When the relationship between the feature quantities used in the deterioration diagnosis of the target equipment and the external factors changes, it is estimated that the diagnostic model for accurately diagnosing the deterioration of the target equipment also changes. As described above, the driving environment indicator is likely to be a continuous value. At the boundary where the diagnostic model switches depending on the driving environment, it is possible that accurate diagnostic results cannot be obtained regardless of whether a diagnostic model is used before or after the switch. The external factor label data indicates that the relationship between the feature quantities and the external factors is such that accurate diagnostic results may not be obtained regardless of which diagnostic model is used. The external factor label data can contribute to the construction of accurate diagnostic models. In other words, it is expected that accurate diagnostic results can be obtained when diagnosing the deterioration of the target equipment using a diagnostic model created based on the external factor label data.

[0012] FIG. 1 is a diagram showing an example of the configuration of an information processing system 6 including a label creation apparatus 1 according to a first embodiment. The information processing system 6 includes the label creation apparatus 1, a diagnostic model creation apparatus 2, a degradation diagnosis apparatus 3, and a storage device 4. In the first embodiment, the storage device 4 is provided in the information processing system 6, but this is merely an example. It is not essential that the information processing system 6 includes the storage device 4; for example, the storage device 4 may be provided in a system external to the information processing system 6 and connected to the information processing system 6. The label creation apparatus 1, the diagnostic model creation apparatus 2, and the degradation diagnosis apparatus 3 are provided in, for example, target equipment. In the first embodiment, the label creation apparatus 1 is provided in the diagnostic model creation apparatus 2.

[0013] The label creating device 1 creates external factor label data based on the learning data stored in the learning data storage unit 41 of the storage device 4. The learning data is data in which feature amounts are associated with external factor data related to external factors. Details of the learning data will be described later. The label creating device 1 stores the created external factor label data in the external factor label data storage unit 42 of the storage device 4. The diagnostic model creating device 2 acquires learning data to be used for creating a diagnostic model based on the external factor label data stored in the external factor label data storage unit 42 by the label creating device 1 from the learning data storage unit 41 of the storage device 4, and creates a diagnostic model using the acquired learning data. The diagnostic model creating device 2 stores the created diagnostic model in the diagnostic model storage unit 43 of the storage device 4. Details of the label creating device 1 and the diagnostic model creating device 2 will be described later.

[0014] The degradation diagnosis device 3 is connected to a sensor 5. The sensor 5 is provided in, for example, a target device, collects data on the operating state of the target device (hereinafter referred to as "sensor data") and external factor data on external factors, and outputs the sensor data and external factor data to the degradation diagnosis device 3. The sensor 5 may be a vibration sensor, a temperature sensor, a rainfall sensor, or the like. The sensor data and external factor data are, for example, time-series data of sensor measurement values ​​for a predetermined period of time obtained by the sensor 5 at predetermined intervals. The sensor data and external factor data indicate sensor measurement values ​​such as vibration, temperature, or precipitation. Note that this is merely an example, and the sensor data may include, for example, control values ​​such as command values ​​or reference values ​​for a predetermined period of time obtained by multiple sensors 5 at predetermined intervals. Note that, for simplicity of explanation, only one sensor 5 is illustrated in FIG. 1 , but multiple sensors 5 may be present. The sensor data and external factor data may be collected from different sensors 5. The degradation diagnosis device 3 may be connected to multiple sensors 5.

[0015] The degradation diagnosis device 3 acquires diagnostic data including sensor data and external factor data from the sensor 5. The degradation diagnosis device 3 determines a diagnostic model to be used for degradation diagnosis of the target equipment based on the external factor data, and performs degradation diagnosis of the target equipment using the determined diagnostic model based on the sensor data. In the following first embodiment, as an example, the degradation diagnosis device 3 acquires vibration data related to the vibration of the target equipment from the sensor 5 as sensor data and temperature data as external factor data. The degradation diagnosis device 3 outputs the results of the degradation diagnosis to an output device (not shown). The output device is, for example, a display device or audio output device included in a personal computer (PC) installed in a management room where an operator or the like performs management work on the target equipment. The output device may also be, for example, a display device or audio output device included in a tablet terminal carried by the operator. Details of the degradation diagnosis device 3 will be described later.

[0016] The storage device 4 is configured by a hard disk drive (HDD) or a solid state drive (SSD), and includes the above-mentioned learning data storage unit 41, external factor label data storage unit 42, and diagnostic model storage unit 43. The learning data storage unit 41 stores learning data. The external factor label data storage unit 42 stores external factor label data created by the label creation device 1. The diagnostic model storage unit 43 stores the diagnostic model created by the diagnostic model creation device 2. The storage device 4 may be provided in the target device, or may be provided outside the target device in a location accessible by the label creation device 1, the diagnostic model creation device 2, and the degradation diagnosis device 3.

[0017] The label creation device 1, diagnostic model creation device 2, and degradation diagnosis device 3 according to the first embodiment will be described in detail. First, the diagnostic model creation device 2 according to the first embodiment, which is equipped with the label creation device 1 according to the first embodiment, will be described. FIG. 2 is a diagram showing an example configuration of the diagnostic model creation device 2 according to the first embodiment, which is equipped with the label creation device 1 according to the first embodiment. The diagnostic model creation device 2 includes the label creation device 1 and a diagnostic model creation unit 21. The label creation device 1 includes a learning data acquisition unit 11, an extrinsic factor region division unit 12, and an extrinsic factor label data creation unit 13. The extrinsic factor region division unit 12 includes an inflection point extraction unit 121, a provisional region setting unit 122, a relational model creation unit 123, a relational model evaluation unit 124, and a region determination unit 125.

[0018] The learning data acquisition unit 11 references the learning data storage unit 41 to acquire learning data. The learning data is created in advance and stored in the learning data storage unit 41. In the first embodiment, the learning data is data in which feature quantities extracted from sensor data acquired from the sensor 5 when the degradation diagnosis device 3 performs a degradation diagnosis on the target device are associated with external factor data related to the operating environment of the target device, i.e., external factors. For example, an administrator or the like may acquire feature quantities and external factor data extracted from sensor data collected by the sensor 5 over a predetermined past period from the target device in advance, create learning data based on the feature quantities and external factor data, and store the learning data in the learning data storage unit 41. For example, the administrator or the like may acquire feature quantities and external factor data extracted by the degradation diagnosis device 3 based on the sensor data collected from the sensor 5, and create learning data based on the feature quantities and external factor data. The extraction of feature quantities in the degradation diagnosis device 3 will be described later. Furthermore, for example, the degradation diagnosis device 3 may have a function of creating learning data based on the extracted feature quantities and store the learning data in the learning data storage unit 41. In the first embodiment, the sensor data itself may be used as the feature, i.e., the training data. The training data may include feature values ​​based on sensor data when the target device is in a normal state, or feature values ​​based on sensor data created assuming that the target device is in a normal state. It is not essential that the training data include feature values ​​based on sensor data when the target device is in an abnormal state, or feature values ​​based on sensor data created assuming that the target device is in an abnormal state. The feature values ​​included in the training data are determined in advance by an administrator or the like. The administrator or the like sets feature values ​​related to the deterioration state of the device, in other words, feature values ​​that can be used as input to a diagnostic model, as feature values ​​included in the training data. The training data acquisition unit 11 outputs the acquired training data to the external factor region division unit 12.

[0019] Based on the training data output from the training data acquisition unit 11, the external factor region division unit 12 divides the range of values ​​of the external factor data in the axial direction of the external factor data into a range of values ​​of the external factor data in which the relationship with the corresponding feature is estimated to be constant (hereinafter referred to as a "constant band") and a range of values ​​of the external factor data in which the relationship with the corresponding feature is estimated to change frequently (hereinafter referred to as a "transition band"). As described above, when the relationship between the feature and the external factor changes in the training data, the diagnostic model is estimated to change as well. In other words, when the relationship between the feature and the external factor data changes frequently in the training data, the diagnostic model created based on the training data may also change frequently. In this case, even if a frequently changing diagnostic model is used, it may not be possible to obtain accurate diagnostic results. On the other hand, when the relationship between the feature and the external factor is constant in the training data, the diagnostic model does not change frequently. In other words, it is estimated that the same diagnostic model can be used for deterioration diagnosis. This is based on the idea that, for external factors acting on feature quantities, if the relationship between the feature quantities and external factors does not change, the influence of the external factors on the feature quantities can be corrected using the same method or parameters, and therefore it is estimated that the same diagnostic model can be applied. In the first embodiment, the process performed by the external factor region dividing unit 12 to divide the range of values ​​of the external factor data in the axial direction of the external factor data into constant bands and transition bands is referred to as "region dividing process."

[0020] Here, the division into constant bands and transition bands by the extrinsic factor region dividing unit 12 will be described in detail with reference to the drawings. Fig. 3 is a diagram for explaining a method in which the extrinsic factor region dividing unit 12 divides the range of external factor data values ​​in the axial direction of the external factor data into constant bands and transition bands in the first embodiment. Fig. 3 shows the relationship between feature amounts (here, amplitude values) based on training data and external factors (here, temperature), with the vertical axis representing the feature amount values ​​and the horizontal axis representing the external factor values.

[0021] As described above, the external factor region segmentation unit 12 includes an inflection point extraction unit 121, a provisional region setting unit 122, a relational model creation unit 123, a relational model evaluation unit 124, and a region determination unit 125. First, based on the learning data output from the learning data acquisition unit 11, the inflection point extraction unit 121 extracts multiple inflection points in the axial direction of the external factor data associated with feature quantities in the learning data. In the first embodiment, an inflection point refers to a point in the axial direction of the external factor data where the relationship between the external factor data and the feature quantities changes. In FIG. 3 , the inflection points extracted by the inflection point extraction unit 121 are indicated by “F1,” “F2,” and “F3.” As shown in FIG. 3 , the inflection point extraction unit 121 extracts, for example, points (“F1,” “F3”) where the value of the feature quantity changes from increasing to decreasing and a point (“F2”) where the value changes from decreasing to increasing in the axial direction of the external factor data. The inflection point extraction unit 121 adds data indicating an inflection point to the learning data output from the learning data acquisition unit 11, and outputs the learning data after adding the data indicating the inflection point (hereinafter referred to as “learning data after adding inflection point”) to the provisional region setting unit 122.

[0022] Next, the provisional region setting unit 122 selects adjacent inflection points (referred to as the first inflection point and the second inflection point) from among the inflection points extracted by the inflection point extraction unit 121 based on the inflection point-added learning data output from the inflection point extraction unit 121. For example, the provisional region setting unit 122 sets "F2" as the first inflection point and "F3" as the second inflection point. The provisional region setting unit 122 sets the range of external factor data values ​​between the selected first and second inflection points as an inflection point range, and sets a provisional region within the inflection point range as a candidate for the constant band. The provisional region setting unit 122 sets the range from the value of the external factor data at the first inflection point to the value of the external factor data at the second inflection point (i.e., the inflection point range) as the initial region of the provisional region. In FIG. 3 , the initial region of the provisional region is indicated by "P." The provisional area setting unit 122 outputs the values ​​of the external factor data in the set provisional area and the feature values ​​corresponding to the values ​​of the external factor data to the relational model creation unit 123. If the initial area "P" is the provisional area, the provisional area setting unit 122 outputs the values ​​of the external factor data and the feature values ​​from the first inflection point "F2" to the second inflection point "F3" to the relational model creation unit 123.

[0023] Next, the relational model creation unit 123 creates a model (hereinafter referred to as a "relational model") indicating the relationship between the feature quantities and the external factor data based on the values ​​of the external factor data in the provisional region and the values ​​of the feature quantities corresponding to the values ​​of the external factor data. The relational model is expressed, for example, by a linear expression. For example, if the provisional region is the initial region "P," the relational model creation unit 123 creates a relational model based on the values ​​of the external factor data and the feature quantities from the first inflection point "F2" to the second inflection point "F3." The relational model creation unit 123 outputs the created relational model to the relational model evaluation unit 124, together with the values ​​of the external factor data in the provisional region and the values ​​of the feature quantities corresponding to the values ​​of the external factor data, which are output from the provisional region setting unit 122, i.e., together with the learning data in the provisional region.

[0024] Next, the relational model evaluation unit 124 evaluates the accuracy of the relational model using the values ​​of the external factor data included in the provisional region and the feature values ​​corresponding to the external factor data values. For example, assume that the relational model is expressed by the following linear equation (1). In the following linear equation (1), y represents the feature value and x represents the external factor data: y = x + 3 ... (1) The relational model evaluation unit 124 assigns a value of a certain external factor data within the provisional region (e.g., the range between the first inflection point "F2" and the second inflection point "F3") and the feature value corresponding to the value of the external factor data into the relational model created by the relational model creation unit 123. The relational model evaluation unit 124 then determines the error between the feature value obtained by the assignment into the relational model and the assigned feature value. For example, if external factor data "21°C" and a feature value "25 mm" are assigned to the relational model, the resulting feature value is "24 mm" with an error of "1 mm." The relational model evaluation unit 124 evaluates the relational model, for example, by comparing the calculated error with a preset threshold (hereinafter referred to as the "relational model evaluation threshold"). For example, if the error is equal to or less than the relational model evaluation threshold, the relational model evaluation unit 124 determines that the relational model has sufficiently high accuracy, and if the error is greater than the relational model evaluation threshold, the relational model evaluation unit 124 determines that the relational model has low accuracy. The relational model evaluation unit 124 outputs the evaluation result of the relational model to the region determination unit 125, together with the values ​​of the external factor data in the provisional region and the values ​​of the feature quantities corresponding to the values ​​of the external factor data.

[0025] The region determination unit 125 then determines whether the relationship model has been evaluated to have sufficiently high accuracy based on the evaluation result of the relationship model output from the relationship model evaluation unit 124. If it is determined that the accuracy of the relationship model has been evaluated to be sufficiently high, the region determination unit 125 determines that the provisional region in which the accuracy of the relationship model has been evaluated to be high is to be divided into a constant band. Then, the region determination unit 125 determines that the range of values ​​of the external factor data other than the determined constant band in the inflection point range is to be divided into a transition band. On the other hand, if it is determined that the accuracy of the relationship model has not been evaluated to be sufficiently high, i.e., that the accuracy of the relationship model has been evaluated to be low, the region determination unit 125 outputs an instruction to the provisional region setting unit 122 to narrow the range of the provisional region and reset the provisional region. The provisional region setting unit 122 resets the provisional region based on the instruction from the region determination unit 125. Specifically, the provisional region setting unit 122 narrows the range of values ​​of the external factor data at both ends of the provisional region and resets the provisional region.

[0026] If it is now assumed that the accuracy of the relational model is sufficiently high, the region determination unit 125 will determine that the provisional region between the first inflection point "F2" and the second inflection point "F3" is divided into a constant band. In this case, since there is no region other than the constant band in the provisional region, i.e., no range of values ​​of the external factor data, the region determination unit 125 does not set a transition band in the provisional region between the first inflection point "F2" and the second inflection point "F3."

[0027] If the accuracy of the current relationship model is evaluated as low, the region determination unit 125 outputs an instruction to the provisional region setting unit 122 to narrow the range of the provisional region and reset the provisional region. In this case, the provisional region setting unit 122 resets the range of external factor data indicated by "D" in FIG. 3 (the range of external factor data values ​​from point "E1" to point "E2" in FIG. 3) as the provisional region. Note that when an instruction to reset the provisional region is output, the extent to which the range of external factor data to be reset as the provisional region is narrowed from both ends of the provisional region before resetting is determined in advance. The provisional region setting unit 122 outputs the values ​​of the external factor data in the reset provisional region and the values ​​of the feature quantities corresponding to the values ​​of the external factor data to the relational model creation unit 123. The relational model creation unit 123 recreates the relational model and outputs the recreated relational model to the relational model evaluation unit 124. The relational model evaluation unit 124 reevaluates the recreated relational model, and outputs the evaluation result of the reevaluated relational model to the area determination unit 125 .

[0028] The provisional region setting unit 122, the relational model creation unit 123, and the relational model evaluation unit 124 each reset the provisional region, recreate the relational model, and reevaluate the accuracy of the relational model until the region determination unit 125 determines that the provisional region should be divided into a constant band, or until the region determination unit 125 determines that the provisional region cannot be a constant band. If the provisional region set by the provisional region setting unit 122 is smaller than a predetermined range (hereinafter referred to as the "minimum provisional region range"), the region determination unit 125 determines that the provisional region cannot be a constant band, even if the relationship model is evaluated to have sufficiently high accuracy, and sets the provisional region as a transition band. Then, the region determination unit 125 determines that the range of values ​​of external factor data other than the provisional region should also be divided into transition bands within the inflection point range.

[0029] When the accuracy of a relational model is sufficiently high, it can be said that a certain relationship between the feature and the external factor is established within the range of values ​​of the external factor data for which the relational model was created. In other words, it is possible to easily estimate the impact of external factors on the feature. A diagnostic model created from feature values ​​corresponding to external factors within this range is expected to be a diagnostic model in which the accuracy of degradation diagnosis is unlikely to decrease due to external factors when the diagnostic model is used to perform a degradation diagnosis of a target device. In other words, it is expected that the same diagnostic model can be used for feature values ​​corresponding to external factors within this range in degradation diagnosis. On the other hand, when the accuracy of a relational model is low, it can be said that a certain relationship between the feature and the external factor is not established within the range of values ​​of the external factor data for which the relational model was created. In other words, it is expected that the relationship between the feature and the external factor changes frequently. In other words, it is expected that it is difficult to estimate the impact of external factors on the feature. A diagnostic model created from training data including feature values ​​corresponding to external factors within this range is expected to be a diagnostic model in which the accuracy of degradation diagnosis is likely to decrease due to external factors when the diagnostic model is used to perform a degradation diagnosis of a target device. This is because if the relationship between a feature and an external factor acting on the feature changes frequently, it is impossible to correct the effect of the external factor on the feature using the same method or parameters, and it is estimated that the diagnostic model for performing accurate deterioration diagnosis will also change frequently. Even if the accuracy of the relationship model is sufficiently high, if the range of external factor data from which the relationship model was created is smaller than the minimum provisional domain range, it can be said that there is not enough data to evaluate the relationship between the feature and the external factor. It is doubtful whether it can be said that a diagnostic model created from training data containing feature values ​​corresponding to external factors in this range is unlikely to result in a decrease in the accuracy of deterioration diagnosis due to external factors when the diagnostic model is used to perform a deterioration diagnosis of a target device.Therefore, as described above, in this case, the area determination unit 125 determines that there is no range within the inflection point range that can be classified as a constant band, that is, that a diagnostic model created from learning data including features corresponding to external factors within this inflection point range is expected to be a diagnostic model in which, when a deterioration diagnosis of the target equipment is performed using the diagnostic model, the accuracy of the deterioration diagnosis may be reduced due to external factors.

[0030] By performing the above-described extraction of inflection points, setting of provisional regions, creation of relationship models, evaluation of the relationship models, and division into constant bands and transition bands between all inflection points, the extrinsic factor region division unit 12 divides all ranges of external factor data values ​​in the axial direction of the external factor data into constant bands and transition bands. The extrinsic factor region division unit 12 outputs data (hereinafter referred to as "region setting data") relating to the constant bands or transition bands divided within the range of external factor data values ​​in the axial direction of the external factor data to the extrinsic factor label data creation unit 13. The region setting data is data in which data indicating the range of external factor data values ​​is associated with data indicating whether the range is a constant band or a transition band.

[0031] Based on the region setting data output from the external factor region dividing unit 12, the external factor label data creating unit 13 assigns an external factor label to the range of values ​​of the external factor data divided into constant bands on the axis of the external factor data, indicating that the range is a constant range in which the relationship between the external factor data in that range and the corresponding feature is constant, i.e., a constant band, and assigns an external factor label to the range of values ​​of the external factor data divided into transition bands, indicating that the relationship with the corresponding feature is estimated to change frequently, i.e., a transition band.

[0032] FIG. 4 is a diagram illustrating an example of external factor labels assigned to the axes of external factor data by the external factor label data creation unit 13 in the first embodiment. In FIG. 4 , the external factor region division unit 12 divides the external factor (here, temperature) into a constant band (a range below 10°C), a transition band (a range from 10°C to 17°C), and a constant band (a range from 17°C to 28°C). In this case, the external factor label data creation unit 13 assigns an external factor label to the axis of the external factor data indicating that the range below 10°C is a constant band (i.e., the relationship between the feature and the external factor data is constant) and that this range is a range in which the same diagnostic model (hereinafter referred to as the "first diagnostic model") can be used. In FIG. 4 , this external factor label is indicated as the "first model application region." The external factor label data creation unit 13 also assigns an external factor label to the range of 10°C or higher and lower than 17°C along the axis of the external factor data, indicating that the range is a transition zone, i.e., a range in which the relationship between the corresponding feature and the external factor data is estimated to change frequently. In FIG. 4 , this external factor label is indicated as "No Applicable Model." The external factor label data creation unit 13 also assigns an external factor label to the range of 17°C or higher and lower than 28°C along the axis of the external factor data, indicating that the range is a constant zone (i.e., the relationship between the feature and the external factor data is constant) and that the same diagnostic model (hereinafter referred to as the "second diagnostic model"), different from the first diagnostic model, can be used. In FIG. 4 , this external factor label is indicated as the "Second Model Applicable Region." As shown in Figure 4, when there are multiple constant bands, the external factor label data creation unit 13 assigns external factor labels to each constant band so that it can be seen that different diagnostic models can be used for the learning data that is assumed to have been acquired in the driving environment indicated by the external factor data in that range.

[0033] After assigning an external factor label, the external factor label data creation unit 13 creates data related to the external factor label (hereinafter referred to as "external factor label data"). The external factor label data is data in which a value range of the external factor data is associated with an external factor label. In addition to the value range of the external factor data and the external factor label, the external factor label data may also be data in which data indicating the type of external factor (here, temperature) and data indicating the type of feature (here, amplitude value) are associated.

[0034] FIG. 5 is a diagram illustrating an example of the contents of external factor label data created by the external factor label data creation unit 13 in the first embodiment. The external factor label data shown in FIG. 5 is external factor label data created after the external factor label data creation unit 13 assigns external factor labels such as those shown in FIG. 4. In the external factor label data shown in FIG. 5, the external factor label "11" is an external factor label indicating a constant band (i.e., a constant relationship between the feature amount and the external factor data) and a range in which it is estimated that the same first diagnostic model can be used. The external factor label "12" is an external factor label indicating a constant band (i.e., a constant relationship between the feature amount and the external factor data) and a range in which it is estimated that the same second diagnostic model can be used. The external factor label "00" is an external factor label indicating a transition band (i.e., a range in which it is estimated that the relationship between the feature amount and the external factor data changes frequently).

[0035] The external factor label data creating unit 13 stores the created external factor label data in the external factor label data storage unit 42 of the storage device 4 .

[0036] Based on the external factor label data created by the external factor label data creation unit 13, the diagnostic model creation unit 21 acquires, from the learning data storage unit 41 of the storage device 4, learning data corresponding to a value range of external factor data associated with an external factor label indicating a range in which the relationship between the external factor data and the feature is estimated to be constant. If the external factor label data has the content shown in FIG. 5 , the diagnostic model creation unit 21 acquires, as learning data for creating a first diagnostic model, learning data to which external factor data indicating a temperature below 10°C is assigned from the learning data storage unit 41. Furthermore, the diagnostic model creation unit 21 acquires, as learning data for creating a second diagnostic model, learning data to which external factor data indicating a temperature between 17°C and 28°C is assigned from the learning data storage unit 41. In this way, when there are value ranges of external factor data to which different external factor labels are assigned, the diagnostic model creation unit 21 acquires learning data for each value range of the external factor data.

[0037] The diagnostic model creation unit 21 creates a diagnostic model using the acquired training data. In the above example, the diagnostic model creation unit 21 creates a first diagnostic model using training data to which external factor data indicating a temperature below 10°C is assigned, and creates a second diagnostic model using training data to which external factor data indicating a temperature between 17°C and 28°C is assigned. In the first embodiment, the diagnostic model is assumed to be a model that receives a feature as input and outputs a value indicating the state of the target device. The value indicating the state of the target device is, for example, a value indicating the degree of deterioration of the target device. The diagnostic model creation unit 21 may create the diagnostic model using a known method such as supervised learning or unsupervised learning. Note that the diagnostic model creation unit 21 does not create a diagnostic model from training data corresponding to a range of values ​​of external factor data associated with an external factor label indicating that the relationship between the external factor data and the feature within that range is estimated to change frequently.

[0038] The reliability determination unit 211 included in the diagnostic model creation unit 21 determines the reliability of the diagnostic model created by the diagnostic model creation unit 21. For example, the reliability determination unit 211 determines the reliability of the diagnostic model based on the number of training data used to create the diagnostic model by the diagnostic model creation unit 21. For example, an administrator or the like creates conditions (hereinafter referred to as "reliability determination conditions") that determine the reliability based on the number of training data, and stores the conditions in a location that can be referenced by the diagnostic model creation device 2. The administrator or the like creates the reliability determination conditions so that the reliability of the diagnostic model increases as the number of training data used increases. Note that, here, the higher the reliability, the more reliable the diagnostic model whose reliability has been determined is, in other words, the higher the accuracy. Note that the reliability determination unit 211 determines the reliability for each diagnostic model.

[0039] The diagnostic model creation unit 21 stores the created diagnostic model in the diagnostic model storage unit 43. At this time, the diagnostic model creation unit 21 associates, with the diagnostic model, data indicating the range of values ​​of the external factor data corresponding to the diagnostic model and the reliability determined by the reliability determination unit 211. The diagnostic model creation unit 21 may further associate, with the diagnostic model, data indicating the feature amount to be input.

[0040] Next, a description will be given of the degradation diagnosis device 3 according to embodiment 1. Fig. 6 is a diagram showing an example of the configuration of the degradation diagnosis device 3 according to embodiment 1. The degradation diagnosis device 3 includes a diagnostic data acquisition unit 31, a model determination unit 32, a model acquisition unit 33, a feature extraction unit 34, a diagnosis unit 35, and a diagnosis result output unit 36.

[0041] The diagnostic data acquisition unit 31 acquires diagnostic data including sensor data (here, vibration data) relating to the operating state of the target equipment and external factor data (here, temperature data) relating to the operating environment of the target equipment from the sensor 5. Of the acquired diagnostic data, the diagnostic data acquisition unit 31 outputs the external factor data to the model determination unit 32 and outputs the sensor data to the feature extraction unit 34.

[0042] The model determination unit 32 determines a diagnostic model to be used for degradation diagnosis based on the external factor data acquired by the diagnostic data acquisition unit 31 and data indicating the value range of the external factor data associated with the diagnostic model. The model determination unit 32 determines the diagnostic model by referring to the diagnostic model storage unit 43 of the storage device 4. For example, assume that the external factor data indicates an external factor (here, temperature) of 18°C. Furthermore, assume that the second diagnostic model stored in the diagnostic model storage unit 43 is associated with data indicating a value range of the external factor data from 17°C to 28°C. In this case, the model determination unit 32 determines the second diagnostic model to be used for degradation diagnosis. As described above, the diagnostic model creation unit 21 of the diagnostic model creation device 2 does not create a diagnostic model based on training data corresponding to a transition band, i.e., a range of values ​​of external factor data to which an external factor label indicating that the relationship between the external factor data and the feature values ​​in that range is estimated to change frequently, is assigned. In other words, depending on the value of the external factor data, a corresponding diagnostic model may not be created. In this case, the model determination unit 32 determines that a diagnostic model to be used for degradation diagnosis has not been created. The model determination unit 32 outputs data capable of identifying the diagnostic model determined to be used for degradation diagnosis (hereinafter referred to as "model identification data") to the model acquisition unit 33. Note that data capable of identifying the diagnostic model (e.g., an ID) was assigned to the diagnostic model by the diagnostic model creation unit 21 when the diagnostic model was created. When the model determination unit 32 determines that a diagnostic model to be used for degradation diagnosis has not been created, it outputs data notifying that a diagnostic model has not been created (hereinafter referred to as "model not created notification") to the model acquisition unit 33.

[0043] When the model determination unit 32 outputs model identification data, the model acquisition unit 33 acquires the diagnostic model that the model determination unit 32 has determined to be used for degradation diagnosis from the diagnostic model storage unit 43 of the storage device 4. When the model determination unit 32 outputs a model non-creation notification, the model acquisition unit 33 creates a diagnostic model to be used for degradation diagnosis.

[0044] Creation of a diagnostic model by the model acquisition unit 33 will be described in detail. Based on the values ​​of external factor data acquired by the diagnostic data acquisition unit 31, the model acquisition unit 33 refers to the learning data storage unit 41 of the storage device 4 to acquire learning data in which external factor data surrounding the external factor data is associated with feature quantities. The model acquisition unit 33 may acquire external factor data from the diagnostic data acquisition unit 31 via the model determination unit 32. Conditions for determining which external factor data should be used as data surrounding the external factor data acquired by the diagnostic data acquisition unit 31 (hereinafter referred to as "external factor data acquisition conditions") are created in advance by an administrator or the like and stored in a location that can be referenced by the degradation diagnosis device 3.

[0045] The external factor data acquisition condition may, for example, state that "external factor data within a range of ±0.1°C of the value of the external factor data acquired by the diagnostic data acquisition unit 31 is used as data related to external factors." In this case, for example, if the value of the external factor data acquired by the diagnostic data acquisition unit 31 is 16°C, the model acquisition unit 33 acquires, from the training data storage unit 41, training data in which external factor data indicating values ​​between 15.9°C and 16.1°C are associated with feature quantities. Furthermore, the external factor data acquisition condition may, for example, state that "10 pieces of external factor data closest to the value of the external factor data acquired by the diagnostic data acquisition unit 31 are used as data related to external factors." In this case, for example, if the value of the external factor data acquired by the diagnostic data acquisition unit 31 is 16°C, the model acquisition unit 33 identifies 10 pieces of external factor data with values ​​close to 16°C, centered around 16°C, and acquires learning data in which the identified external factor data is associated with feature values ​​from the learning data storage unit 41.

[0046] When the model acquisition unit 33 acquires the learning data, it creates a diagnostic model based on the acquired learning data. The model acquisition unit 33 may create the diagnostic model in a manner similar to the manner in which the diagnostic model creation unit 21 of the diagnostic model creation device 2 creates a diagnostic model. Furthermore, when the model acquisition unit 33 creates a diagnostic model, it may determine the reliability of the created diagnostic model. The model acquisition unit 33 may determine the reliability of the diagnostic model in a manner similar to the manner in which the reliability determination unit 211 of the diagnostic model creation device 2 determines the reliability of the diagnostic model. In this way, when a diagnostic model to be used for degradation diagnosis has not been created yet, the model acquisition unit 33 creates a diagnostic model each time based on local learning data in accordance with the external factor data acquisition conditions.

[0047] As described above, the external factor data is time-series data. The model acquisition unit 33 performs a process of acquiring a diagnostic model corresponding to the external factor data in time series. Furthermore, when the model acquisition unit 33 completes acquisition of a diagnostic model corresponding to the external factor data, it creates data (hereinafter referred to as "applicable model range data") in which the acquisition date and time of the external factor data are associated with data that can identify the corresponding diagnostic model in time series. The external factor data is assigned the acquisition date and time of the external factor data.

[0048] The model acquisition unit 33 outputs the diagnostic model acquired by referring to the diagnostic model storage unit 43 or the diagnostic model acquired by creating it based on the learning data acquired from the learning data storage unit 41, together with the created applicable model range data, to the diagnosis unit 35. When the model acquisition unit 33 creates a diagnostic model, it associates data indicating the value of the external factor data corresponding to the diagnostic model with the determined reliability. The model acquisition unit 33 may further associate data indicating the input feature amount with the diagnostic model.

[0049] The feature extraction unit 34 extracts feature quantities from the sensor data acquired by the diagnostic data acquisition unit 31. More specifically, the feature extraction unit 34 extracts feature quantities to be input to a diagnostic model from the sensor data acquired by the diagnostic data acquisition unit 31. Note that an administrator or the like may create data (hereinafter referred to as "feature quantity definition data") in advance that can identify feature quantities to be input to a diagnostic model that can be used in the degradation diagnosis device 3 to diagnose degradation of a target device, and store the data in a buffer or the like of the feature extraction unit 34. The feature extraction unit 34 identifies what feature quantities to extract in accordance with the feature quantity definition data. In this example, the feature quantity to be input to the diagnostic model is an amplitude value. The feature extraction unit 34 extracts the amplitude value of vibration as a feature quantity. The feature extraction unit 34 may extract feature quantities from the sensor data using a known feature extraction method. The feature extraction unit 34 outputs the extracted feature quantities to the diagnosis unit 35.

[0050] The diagnosing unit 35 performs a deterioration diagnosis of the target device based on the feature extracted by the feature extracting unit 34 and using the diagnostic model acquired by the model acquiring unit 33 .

[0051] First, the diagnostic unit 35 determines which of the diagnostic models output from the model acquisition unit 33 corresponds to the feature extracted by the feature extraction unit 34. Specifically, based on the feature extracted by the feature extraction unit 34, the diagnostic unit 35 determines the diagnostic model acquired by the model acquisition unit 33 for the external factor data corresponding to the feature extracted by the feature extraction unit 34 as the diagnostic model corresponding to the feature extracted by the feature extraction unit 34, i.e., the diagnostic model to be used for deterioration diagnosis of the target device based on the feature extracted by the feature extraction unit 34.

[0052] FIG. 7 is a diagram illustrating an example of a method in which the diagnosing unit 35 determines, from among the diagnostic models output from the model acquiring unit 33, a diagnostic model corresponding to the feature extracted by the feature extracting unit 34 in the first embodiment. As described above, the sensor data and the external factor data are time-series data. That is, the sensor data and the external factor data are linked by the acquisition date and time of the data. The sensor data and the external factor data are each assigned a data acquisition date and time. Based on the applicable model range data output from the model acquiring unit 33, the diagnosing unit 35 determines which diagnostic model to use for which feature for which time period by matching the acquisition date and time of the data included in the applicable model range data with the acquisition date and time of the data assigned to the feature.

[0053] 7, for example, assume that diagnostic model "B" is associated with external factor data for the time period from time t1 to t2 in the applicable model range data. In this case, the diagnostic unit 35 determines that diagnostic model "B" is the diagnostic model to be used to input the feature amounts for the time period from time t1 to t2 and perform a degradation diagnosis of the target device. In this way, for the feature amounts extracted by the feature amount extraction unit 34, the diagnostic unit 35 determines that the diagnostic model acquired based on the external factor data at the corresponding data acquisition date and time is the diagnostic model to be used for degradation diagnosis of the target device based on the feature amounts.

[0054] The diagnostic unit 35 determines a diagnostic model to be used for diagnosing deterioration of the target device based on the feature extracted by the feature extraction unit 34, and then performs deterioration diagnosis of the target device using the diagnostic model. The diagnostic model is output from the model acquisition unit 33. The diagnostic unit 35 can identify the diagnostic model based on data that can identify the diagnostic model included in the applicable model range data.

[0055] For example, the diagnostic unit 35 compares a value indicating the state of the target device obtained by inputting the feature quantities into the diagnostic model with a preset threshold (hereinafter referred to as the "deterioration diagnosis threshold"). If the value indicating the state of the target device is equal to or greater than the degradation diagnosis threshold, the diagnostic unit 35 diagnoses that the target device is degraded. If the value indicating the state of the target device is less than the degradation diagnosis threshold, the diagnostic unit 35 diagnoses that the target device is not degraded. The diagnostic unit 35 may diagnose the degradation of the target device by determining the degree of degradation of the target device, rather than by determining whether the target device is degraded. For example, the diagnostic unit 35 determines the device degradation level, which indicates the degree of degradation of the target device, based on the value indicating the state of the target device obtained by inputting the feature quantities into the diagnostic model. Note that conditions for determining the degree of degradation of the target device when a certain value indicating the state of the target device is present are created in advance and stored in a buffer or the like of the diagnostic unit 35.

[0056] The diagnosis unit 35 outputs the result of the degradation diagnosis of the target device (hereinafter referred to as the "degradation diagnosis result") to the diagnosis result output unit 36. At this time, the diagnosis unit 35 outputs the diagnostic model used in the degradation diagnosis of the target device to the diagnosis result output unit 36 ​​in association with the degradation diagnosis result.

[0057] The diagnostic result output unit 36 ​​outputs the degradation diagnosis result to the output device. At this time, the diagnostic result output unit 36 ​​may output the reliability of the diagnostic model used in the degradation diagnosis of the target device together with the degradation diagnosis result to the output device. The diagnostic result output unit 36 ​​may output data that can identify the diagnostic model used in the degradation diagnosis of the target device together with the degradation diagnosis result to the output device. The reliability of the diagnostic model and the data that can identify the diagnostic model are assigned to the diagnostic model. The diagnostic result output unit 36 ​​can determine the reliability of the diagnostic model and the data that can identify the diagnostic model based on the diagnostic model output from the diagnostic unit 35.

[0058] When the diagnostic result output unit 36 ​​outputs the degradation diagnosis result to the output device, the output device displays or outputs the degradation diagnosis result as audio. By checking the displayed or audio output result of the degradation diagnosis of the target equipment, an operator or the like can, for example, determine whether the target equipment is degraded or the degree of degradation of the target equipment. Furthermore, when the diagnostic result output unit 36 ​​outputs a reliability, the output device may display or output the reliability along with the degradation diagnosis result of the target equipment. This allows an operator or the like to understand the reliability of the degradation diagnosis result and take appropriate measures, such as inspecting the target equipment early even if the target equipment is diagnosed as not degraded if the reliability is low. Furthermore, when the diagnostic result output unit 36 ​​outputs the diagnostic model used in the degradation diagnosis, the output device may display or output audio data that can identify the diagnostic model along with the degradation diagnosis result of the target equipment. This allows an operator or the like to analyze the diagnostic model, for example.

[0059] The operations of the diagnostic model creation device 2 equipped with the label creation device 1 and the degradation diagnosis device 3 will be described. First, the operations of the label creation device 1 and the diagnostic model creation device 2 will be described. FIG. 8 is a flowchart for explaining the operation of the diagnostic model creation device 2 according to embodiment 1 equipped with the label creation device 1 according to embodiment 1. Upon receiving an operation start instruction, the diagnostic model creation device 2 starts the operation shown in the flowchart of FIG. 8. For example, an operator or the like inputs the operation start instruction from a PC in a management room. Upon receiving the operation start instruction, the control unit (not shown) of the diagnostic model creation device 2 causes the label creation device 1 and the diagnostic model creation unit 21 to start operation. With regard to the operation shown in the flowchart of FIG. 8, the processing operations of steps ST1 to ST3 are performed by the label creation device 1, and the processing operation of step ST4 is performed by the diagnostic model creation device 2.

[0060] The learning data acquiring unit 11 acquires learning data by referring to the learning data storage unit 41 (step ST1). The learning data acquiring unit 11 outputs the acquired learning data to the external factor region dividing unit 12.

[0061] The extrinsic factor region dividing unit 12 performs a region dividing process to divide the range of values ​​of the extrinsic factor data in the axial direction of the extrinsic factor data into constant bands and transition bands (step ST2) based on the learning data output from the learning data acquiring unit 11 in step ST1. The extrinsic factor region dividing unit 12 outputs the region setting data to the extrinsic factor label data creating unit 13.

[0062] Based on the region setting data output from the external factor region dividing unit 12 in step ST2, the external factor label data creating unit 13 assigns external factor labels indicating that the range of values ​​of the external factor data divided into constant bands is a constant band on the axis of the external factor data, and assigns external factor labels indicating that the range of values ​​of the external factor data divided into transition bands is a transition band. After assigning the external factor labels, the external factor label data creating unit 13 creates external factor label data (step ST3). The external factor label data creating unit 13 stores the created external factor label data in the external factor label data storage unit 42 of the storage device 4.

[0063] Based on the external factor label data created by the external factor label data creation unit 13 in step ST3, the diagnostic model creation unit 21 acquires, from the learning data storage unit 41 of the storage device 4, learning data corresponding to a range of values ​​of external factor data associated with an external factor label indicating that the range of external factor data has a constant relationship with the feature. The diagnostic model creation unit 21 then creates a diagnostic model using the acquired learning data (step ST4). In step ST4, the reliability determination unit 211 included in the diagnostic model creation unit 21 determines the reliability of the diagnostic model. The diagnostic model creation unit 21 stores the created diagnostic model in the diagnostic model storage unit 43. At this time, the diagnostic model creation unit 21 associates, with the diagnostic model, data indicating the range of values ​​of the external factor data corresponding to the diagnostic model and the reliability determined by the reliability determination unit 211. The diagnostic model creation unit 21 may further associate data indicating input feature values ​​with the diagnostic model.

[0064] 9 is a flowchart for explaining details of the region segmentation process by the extrinsic factor region segmentation unit 12 in step ST2 of Fig. 8. First, the inflection point extraction unit 121 extracts multiple inflection points in the axial directions of the extrinsic factor data that are associated with feature quantities in the training data, based on the training data output from the training data acquisition unit 11 (step ST101). The inflection point extraction unit 121 outputs the training data with the inflection points added to it to the extrinsic factor region segmentation unit 12.

[0065] The provisional region setting unit 122 selects a first inflection point and a second inflection point based on the learning data after inflection points have been added and output from the inflection point extraction unit 121. The provisional region setting unit 122 defines the range of values ​​of the external factor data between the selected first inflection point and the selected second inflection point as an inflection point range, and sets a provisional region within the inflection point range as a candidate for the constant band (step ST101). The provisional region setting unit 122 outputs the values ​​of the external factor data in the set provisional region and the values ​​of the feature quantities corresponding to the values ​​of the external factor data to the relational model creation unit 123.

[0066] The relational model creation unit 123 creates a relational model based on the values ​​of the external factor data in the provisional region and the values ​​of the feature quantities corresponding to the values ​​of the external factor data (step ST103). The relational model creation unit 123 outputs the created relational model to the relational model evaluation unit 124, together with the values ​​of the external factor data in the provisional region and the values ​​of the feature quantities corresponding to the values ​​of the external factor data, which are output from the provisional region setting unit 122.

[0067] The relational model evaluation unit 124 evaluates the accuracy of the relational model using the values ​​of the external factor data included in the provisional region and the values ​​of the feature quantities corresponding to the values ​​of the external factor data, and outputs the evaluation result of the relational model together with the values ​​of the external factor data in the provisional region and the values ​​of the feature quantities corresponding to the values ​​of the external factor data, i.e., together with the learning data in the provisional region, to the region determination unit 125. Based on the evaluation result of the relational model output from the relational model evaluation unit 124, the region determination unit 125 determines whether the relational model has been evaluated to have sufficiently high accuracy (step ST104).

[0068] If it is determined that the relationship model has been evaluated to have sufficiently high accuracy ("YES" in step ST104), the region determination unit 125 determines that the provisional region in which the relationship model was evaluated to have high accuracy is to be divided into a constant band. Then, the region determination unit 125 determines that the range of extrinsic factor data values ​​other than the determined constant band in the inflection point range is to be divided into a transition band (step ST105). The extrinsic factor region division unit 12 outputs the region setting data to the extrinsic factor label data creation unit 13.

[0069] On the other hand, if it is determined that the relationship model is not evaluated to have sufficiently high accuracy, i.e., that the accuracy of the relationship model is evaluated to be low ("NO" in step ST104), the area determination unit 125 outputs an instruction to the provisional area setting unit 122 to narrow the range of the provisional area and reset the provisional area. The provisional area setting unit 122 resets the provisional area based on the instruction from the area determination unit 125. Specifically, the provisional area setting unit 122 narrows the range of values ​​of the external factor data at both ends of the provisional area and resets the provisional area. The operation of the external factor area division unit 12 returns to the processing of step ST102.

[0070] In step ST104, the area determination unit 125 determines whether the provisional area set by the provisional area setting unit 122 is smaller than the minimum provisional area range. If the provisional area set by the provisional area setting unit 122 is smaller than the minimum provisional area range, even if the relationship model is evaluated to have sufficiently high accuracy, the area determination unit 125 determines that the provisional area cannot be a constant zone and classifies the provisional area as a transition zone. The area determination unit 125 then determines that the range of extrinsic factor data values ​​other than the provisional area within the inflection point range should also be classified as a transition zone. The extrinsic factor area division unit 12 outputs the area setting data to the extrinsic factor label data creation unit 13. The operation of the extrinsic factor area division unit 12 then ends as shown in the flowchart of FIG. 9.

[0071] When creating a model for the operating conditions of an equipment, as in the technology disclosed in the aforementioned Patent Document 1, the indicators of the operating conditions are unlikely to be continuous values, so the accuracy of the diagnostic results is unlikely to decrease at the boundary between models. In other words, multiple models can be clearly switched based on the sensor data acquired from the equipment (see the upper diagram in Figure 10). However, the indicators of the driving environment are likely to be continuous values. In this case, at the boundary between models, accurate diagnostic results may not be obtained regardless of whether a model is used before or after the switch. (See the upper diagram in Figure 10. Note that the time point at which a model switch, where accurate diagnostic results may not be obtained regardless of whether a model is used, is shown as a transition zone.) In conventional technology, it has been difficult to construct a model that takes into account the possibility that accurate diagnostic results may not be obtained regardless of whether a model is used before or after the switch due to the operating environment.

[0072] In response to this, the label creation device 1 acquires learning data in which feature quantities are associated with external factor data, and, based on the acquired learning data, divides the range of values ​​of the external factor data along the axis of the external factor data into a constant band, which is a range of values ​​of the external factor data in which the relationship with the corresponding feature quantity is estimated to be constant, and a transition band, which is a range of values ​​of the external factor data in which the relationship with the corresponding feature quantity is estimated to change frequently.The label creation device 1 then assigns external factor labels to the range of values ​​of the external factor data divided into the constant band on the axis of the external factor data, indicating that the relationship between the external factor data in that range and the corresponding feature quantity is constant, and assigns external factor labels to the range of values ​​of the external factor data divided into the transition band, indicating that the relationship between the external factor data in that range and the corresponding feature quantity changes frequently, thereby creating external factor label data in which the range of values ​​of the external factor data is associated with the external factor labels.

[0073] This allows the label creation apparatus 1 to provide data that aids in the construction of a diagnostic model that prevents a decline in the accuracy of diagnosing deterioration of the target device due to the operating environment. As a result, the label creation apparatus 1 can contribute to the construction of a diagnostic model that prevents a decline in the deterioration diagnosis of the target device due to the operating environment, and can contribute to the realization of accurate deterioration diagnosis using the diagnostic model. Furthermore, the label creation apparatus 1 can contribute to shortening the time required to design a diagnostic model. Furthermore, when creating external factor label data, the label creation apparatus 1 focuses on the relationship between external factors and feature quantities based on sensor data in the target device's normal state. In other words, the learning data used by the label creation apparatus 1 to create the external factor label data only needs to be data that corresponds to the relationship between external factors and feature quantities based on sensor data in the target device's normal state. This reduces the effort required to create the learning data. In general, preparing learning data that assumes sensor data in an abnormal state requires more effort than preparing learning data that assumes sensor data in a normal state. Furthermore, by providing external factor data, the label creating apparatus 1 can allow, for example, an operator to clearly understand which external factor caused the diagnostic model to switch and at what timing.

[0074] The diagnostic model creation device 2 acquires learning data corresponding to a range of values ​​of external factor data associated with external factor labels indicating that the relationship between the range of external factor data and the corresponding feature is constant, based on the external factor label data created by the label creation device 1, and creates a diagnostic model using the acquired learning data. By creating a diagnostic model based on the external factor label data created by the label creation device 1, the diagnostic model creation device 2 can build a diagnostic model that prevents a decline in the diagnosis of equipment deterioration caused by the operating environment.

[0075] Next, the operation of the degradation diagnosis device 3 will be described. Fig. 11 is a flowchart for explaining the operation of the degradation diagnosis device 3 according to the first embodiment. When the degradation diagnosis device 3 receives an instruction to start a degradation diagnosis of the target device, it starts the operation shown in the flowchart of Fig. 11. For example, an operator or the like inputs an instruction to start a degradation diagnosis of the target device from a PC in a management room. When the control unit (not shown) of the degradation diagnosis device 3 receives the instruction to start a degradation diagnosis of the target device, it causes the diagnostic data acquisition unit 31, the model determination unit 32, the model acquisition unit 33, the feature extraction unit 34, the diagnosis unit 35, and the diagnosis result output unit 36 ​​to start operating.

[0076] The diagnostic data acquisition unit 31 acquires diagnostic data from the sensor 5 (step ST21). Of the acquired diagnostic data, the diagnostic data acquisition unit 31 outputs the external factor data to the model determination unit 32 and outputs the sensor data to the feature extraction unit 34.

[0077] The model determination unit 32 determines a diagnostic model to be used for degradation diagnosis based on the external factor data acquired by the diagnostic data acquisition unit 31 in step ST21 and data indicating the range of values ​​of the external factor data associated with the diagnostic model (step ST22). The model determination unit 32 outputs the model identification data to the model acquisition unit 33. If the model determination unit 32 determines that a diagnostic model to be used for degradation diagnosis has not been created, it outputs a model non-creation notification to the model acquisition unit 33.

[0078] When the model determination unit 32 outputs model identification data in step ST22, the model acquisition unit 33 acquires the diagnostic model determined by the model determination unit 32 to be used for degradation diagnosis from the diagnostic model storage unit 43 of the storage device 4. When the model determination unit 32 outputs a model non-creation notification in step ST22 or data indicating that a diagnostic model has not been created, the model acquisition unit 33 acquires the diagnostic model by creating a diagnostic model to be used for degradation diagnosis (step ST23). Furthermore, when the model acquisition unit 33 has completed acquisition of the diagnostic model corresponding to the external factor data, it creates applicable model range data. The model acquisition unit 33 outputs the diagnostic model acquired by referring to the diagnostic model storage unit 43 or the diagnostic model acquired by creating it based on the learning data acquired from the learning data storage unit 41 to the diagnosis unit 35, together with the created applicable model range data.

[0079] The feature extraction unit 34 extracts feature amounts from the sensor data acquired by the diagnostic data acquisition unit 31 in step ST21 (step ST24). The feature extraction unit 34 outputs the extracted feature amounts to the diagnosis unit 35.

[0080] The diagnosing unit 35 performs a degradation diagnosis of the target device using the diagnostic model acquired by the model acquiring unit 33 in step ST23 based on the feature extracted by the feature extracting unit 34 in step ST24 (step ST25). The diagnosing unit 35 outputs a degradation diagnosis result to the diagnostic result output unit 36. At this time, the diagnosing unit 35 outputs the diagnostic model used for the degradation diagnosis of the target device to the diagnostic result output unit 36 ​​in association with the degradation diagnosis result.

[0081] The diagnostic result output unit 36 ​​outputs the degradation diagnosis result to the output device (step ST26). At this time, the diagnostic result output unit 36 ​​may output the reliability of the diagnostic model used in the degradation diagnosis of the target device together with the degradation diagnosis result to the output device. The diagnostic result output unit 36 ​​may also output data that can identify the diagnostic model used in the degradation diagnosis of the target device together with the degradation diagnosis result to the output device.

[0082] 11, the processing of steps ST22 to ST23 and the processing of step ST24 are performed in parallel, but this is merely an example. For example, the processing of step ST24 may be performed after the processing of steps ST22 to ST23, or vice versa. It is sufficient that the processing of steps ST22 to ST23 and the processing of step ST24 are completed before the processing of step ST25 is performed.

[0083] In this way, the degradation diagnosis device 3 performs degradation diagnosis of the target device using the diagnostic model created by the diagnostic model creation device 2 based on the external factor label data created by the label creation device 1. The degradation diagnosis device 3 acquires diagnostic data including sensor data and external factor data, and determines a diagnostic model to be used for degradation diagnosis based on the acquired external factor data and data indicating the value range of the external factor data associated with the diagnostic model. The degradation diagnosis device 3 acquires the diagnostic model determined to be used for degradation diagnosis. The degradation diagnosis device 3 also extracts feature quantities from the acquired sensor data. Then, based on the extracted feature quantities, the degradation diagnosis device 3 performs degradation diagnosis of the target device using the diagnostic model acquired for the external factor data corresponding to the extracted feature quantities. The degradation diagnosis device 3 outputs the result of the degradation diagnosis of the target device. The degradation diagnosis device 3 performs degradation diagnosis of the target device using a diagnostic model created in consideration of the possibility that accurate diagnostic results may not be obtained before and after switching due to the operating environment, and therefore can perform accurate degradation diagnosis.

[0084] Furthermore, when a diagnostic model to be used for degradation diagnosis has not been created, the degradation diagnosis device 3 creates a diagnostic model each time based on local learning data in accordance with the conditions for acquiring external factor data. By creating a diagnostic model using only local learning data, the degradation diagnosis device 3 can create a diagnostic model that corresponds to the characteristics of a transition zone where the optimal diagnostic model frequently changes.

[0085] In the above-described first embodiment, for the sake of simplicity, the label creation device 1 has been described as creating external factor label data for one set of data consisting of one type of feature (amplitude value) and one type of external factor (temperature). However, this is merely an example. For example, the training data may be data in which external factor data relating to multiple different types of external factors, such as temperature and precipitation, are associated with the feature. The feature may also be multiple different types of feature, such as amplitude value, variance, and standard deviation. In the training data acquired by the training data acquisition unit 11 in the label creation device 1, external factor data relating to multiple different types of external factors may be associated with the feature for each external factor. Furthermore, in the training data acquired by the training data acquisition unit 11, multiple feature values ​​may be associated with one external factor. The label creation device 1 creates external factor label data for each pair of external factor data and feature. In detail, in the label creation device 1, the external factor region division unit 12 divides the range of values ​​of the external factor data into a constant band and a transition band for each pair of external factor data and feature amount, and the external factor label data creation unit 13 assigns an external factor label indicating whether the external factor data axis is a constant band or a transition band for each pair of external factor data and feature amount, thereby creating external factor label data.

[0086] For example, suppose that the feature quantities input to the diagnostic model for the external factor "temperature" are "amplitude" and "variance." In this case, the label creation device 1 creates external factor data for the pair of "temperature" and "amplitude," and for the pair of "temperature" and "variance."

[0087] FIG. 12 is a diagram illustrating an example of an image of an external factor label that the external factor label data creation unit 13 in the label creation device 1 assigns to the axis of the external factor data when the label creation device 1 creates external factor data for a pair of “temperature” and “amplitude value” and a pair of “temperature” and “variance” in embodiment 1.

[0088] In FIG. 12 , for a pair of “air temperature” and “amplitude value,” the external factor region division unit 12 divides the external factor (air temperature) into a constant zone (a range below 10°C), a transition zone (a range from 10°C to 17°C), and a constant zone (a range from 17°C to 28°C). In this case, the external factor label data creation unit 13 assigns an external factor label to the axis of the external factor data indicating that the range below 10°C is a constant zone (i.e., the relationship between the feature and the external factor data is constant) and that it is a range in which the same diagnostic model (hereinafter referred to as the “first diagnostic model”) can be used. In FIG. 4 , this external factor label is indicated as the “first model applicable region.” In FIG. 12 , this external factor label is indicated as the “1a model applicable region.” Furthermore, the external factor label data creation unit 13 assigns an external factor label to the axis of the external factor data indicating that the range from 10°C to 17°C is a transition zone. In Fig. 12, the external factor label is shown as "No applicable model." The external factor label data creation unit 13 also assigns an external factor label indicating that the range of 17°C or higher and lower than 28°C in the axial direction of the external factor data is a constant zone (i.e., the relationship between the feature amount and the external factor data is constant) and that this range is presumed to be a range in which the same diagnostic model (second diagnostic model) as the first diagnostic model can be used. In Fig. 12, the external factor label is shown as "2b model applicable area."

[0089] Furthermore, for the pair of "air temperature" and "variance," the external factor region division unit 12 divides the external factor (air temperature) into a constant zone (a range below 8°C), a transition zone (a range between 8°C and 17°C), and a constant zone (a range between 17°C and 25°C). In this case, the external factor label data creation unit 13 assigns an external factor label to the axis of the external factor data indicating that the range below 8°C is a constant zone (i.e., the relationship between the feature and the external factor data is constant) and that it is a range in which the same diagnostic model (first diagnostic model) can be used. In FIG. 12 , this external factor label is indicated as "Model 1b application area." Furthermore, the external factor label data creation unit 13 assigns an external factor label to the axis of the external factor data indicating that the transition zone (a range between 8°C and 17°C) is a range in which the same diagnostic model (first diagnostic model) can be used. In FIG. 12 , this external factor label is indicated as "Model 1b application area." The external factor label data creation unit 13 also assigns an external factor label to the axis of the external factor data, indicating that the range of 17°C or higher and lower than 25°C is a constant zone (i.e., the relationship between the feature amount and the external factor data is constant) and that this range is estimated to be a range in which the same diagnostic model (second diagnostic model) as the first diagnostic model can be used. In Figure 12, this external factor label is shown as the "2b model application range."

[0090] In the above example, the value range of external factor data labeled as a constant band and the value range of external factor data labeled as a transition band differ depending on the type of feature. Specifically, for the temperature range of 8°C or higher and lower than 10°C, the pair of "temperature" and "amplitude" is labeled as "Model 1a applicable region," while the pair of "temperature" and "variance" is labeled as "no applicable model." In this way, when there is external factor label data corresponding to multiple feature values, in which different external factor labels are assigned to different external factor data within the same external factor data value range on the same external factor data axis, the label creation device 1 uses the external factor label data created by the label creation device 1 to reset the value range of external factor data that has a constant relationship with the corresponding feature value for diagnostic model creation.

[0091] For example, an administrator or the like may check the external factor label data based on the external factor labels assigned as shown in FIG. 12 and decide to create a first diagnostic model for a diagnostic model that uses "amplitude value" and "variance" as inputs when the external factor "temperature" is less than 10°C. The administrator or the like may reset the range of external factor data values ​​to less than 10°C for which the first diagnostic model is estimated to be usable. Then, the administrator or the like may create conditions for creating a diagnostic model (hereinafter referred to as "diagnostic model creation conditions") based on the reset range of external factor data values ​​for which the first diagnostic model is estimated to be usable. The diagnostic model creation conditions are, for example, table-format data that associates the following: a value range of the reset external factor data; an external factor label for the reset value range of the external factor data; data indicating the feature values ​​(in the above example, “amplitude” and “variance”) to be input to the diagnostic model whose applicable range is determined by the external factor label; and data indicating whether the value range of the reset external factor data in the external factor label data originally created by the label creation device 1 for the feature values ​​is a constant band or a transition band (hereinafter referred to as “feature-specific region data”). In the above example, the administrator or the like creates diagnostic model creation conditions that associate data indicating that the range of the external factor “temperature” is less than 10°C, the external factor label for the “first diagnostic model applicable region,” and the feature-specific region data. The feature-specific region data associates the feature “amplitude” with data indicating that values ​​less than 10°C are a constant band, and the feature “variance” with data indicating that values ​​less than 8°C are a constant band and values ​​between 8°C and 10°C are a transition band (see FIG. 12 ). The administrator or the like stores the diagnostic model creation conditions that have been created in the external factor label data storage unit 42 .

[0092] In the diagnostic model creation device 2, if diagnostic model creation conditions are stored in the external factor label data storage unit 42, the diagnostic model creation unit 21 creates a diagnostic model in accordance with the diagnostic model creation conditions. In the example described above, the diagnostic model creation unit 21 determines to create a first diagnostic model for external factor data below 10°C in accordance with the diagnostic model creation conditions, and creates the first diagnostic model using training data to which the external factor data below 10°C is assigned. In this case, the reliability determination unit 211 may determine the reliability of the diagnostic model based on the feature-specific region data defined in the diagnostic model creation conditions. In the example described above, the first diagnostic model is created from training data to which external factor data below 10°C is assigned. However, for the feature "variance" extracted from the training data, temperatures between 8°C and 10°C are originally within the range of values ​​of external factor data to which the label creation device 1 assigns an external factor label as a transition band. The reliability determination unit 211 determines the reliability of a diagnostic model created using learning data to which external factor data has been assigned an external factor label with a transition band by the label creation device 1, as low, for example. The reliability determination unit 211 may determine how low to assign the reliability based on, for example, the external factor data to which the label creation device 1 has assigned an external factor label with a transition band, the proportion of the assigned learning data, etc. Note that the above example is merely an example, and the reliability determination unit 211 may create another diagnostic model, for example, using "amplitude" and "variance" for a region below 8°C and "amplitude" for a region above 8°C and below 10°C.

[0093] Even when there is only one pair of feature quantity and external factor, the administrator or the like may create diagnostic model creation conditions after checking the external factor label data. The administrator or the like can create appropriate diagnostic model creation conditions based on the external factor label data.

[0094] In the above-described first embodiment, the diagnostic model creation unit 21 in the diagnostic model creation device 2 is provided with the reliability determination unit 211, but this is merely an example. The diagnostic model creation device 2 does not necessarily have the function of determining the reliability of the diagnostic model. If the diagnostic model creation device 2 does not have the function of determining the reliability of the diagnostic model, the diagnostic model creation device 2 can be configured without the reliability determination unit 211. Furthermore, in the degradation diagnosis device 3, the model acquisition unit 33 does not necessarily have the function of determining the reliability when creating a diagnostic model.

[0095] In the first embodiment described above, the label creating device 1 is provided in the diagnostic model creating device 2, but this is merely an example. The label creating device 1 and the diagnostic model creating device 2 may each be independent devices. In this case, a control unit (not shown) of the label creating device 1 receives an operation start instruction from an operator or the like, and upon receiving the operation start instruction, causes the learning data acquiring unit 11, the extrinsic factor region dividing unit 12, and the extrinsic factor label data creating unit 13 to start operating.

[0096] In the first embodiment, the label creation device 1, the diagnostic model creation device 2, and the degradation diagnosis device 3 are provided in the target device. However, this is merely an example. For example, in the label creation device 1, some or all of the learning data acquisition unit 11, the external factor region division unit 12, and the external factor label data creation unit 13 may be provided in a server or the like external to the target device. For example, the diagnostic model creation unit 21 of the diagnostic model creation device 2 may be provided in a server or the like external to the target device. For example, in the degradation diagnosis device 3, some or all of the diagnostic data acquisition unit 31, the model determination unit 32, the model acquisition unit 33, the feature extraction unit 34, the diagnosis unit 35, and the diagnostic result output unit 36 ​​may be provided in a server or the like external to the target device.

[0097] 13A and 13B are diagrams showing an example of the hardware configuration of the label creating apparatus 1 according to the first embodiment. In the first embodiment, the functions of the learning data acquiring unit 11, the external factor region dividing unit 12, the external factor label data creating unit 13, and a control unit (not shown) are realized by a processing circuit 101. That is, the label creating apparatus 1 includes the processing circuit 101 for performing control to provide data that helps build a diagnostic model that prevents a decrease in the accuracy of diagnosing equipment deterioration caused by the operating environment. The processing circuit 101 may be dedicated hardware as shown in FIG. 13A, or a processor 104 that executes a program stored in memory as shown in FIG. 13B.

[0098] When processing circuitry 101 is dedicated hardware, processing circuitry 101 may be, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an ASIC (Application Specific Integrated Circuit), an FPGA (Field-Programmable Gate Array), or a combination thereof.

[0099] When the processing circuit is the processor 104, the functions of the learning data acquisition unit 11, the extrinsic factor region division unit 12, the extrinsic factor label data creation unit 13, and the control unit (not shown) are realized by software, firmware, or a combination of software and firmware. The software or firmware is written as a program and stored in the memory 105. The processor 104 reads and executes the program stored in the memory 105 to perform the functions of the learning data acquisition unit 11, the extrinsic factor region division unit 12, the extrinsic factor label data creation unit 13, and the control unit (not shown). In other words, the label creation device 1 includes the memory 105 for storing a program that, when executed by the processor 104, results in the execution of steps ST1 to ST4 of FIG. 8 described above. The program stored in the memory 105 can also be said to cause a computer to execute the processing procedures or methods of the learning data acquisition unit 11, the extrinsic factor region division unit 12, the extrinsic factor label data creation unit 13, and the control unit (not shown). Here, the memory 105 may be, for example, a non-volatile or volatile semiconductor memory such as a RAM, a ROM (Read Only Memory), a flash memory, an EPROM (Erasable Programmable Read Only Memory), or an EEPROM (registered trademark; hereinafter omitted) (Electrically Erasable Programmable Read-Only Memory), a magnetic disk, a flexible disk, an optical disk, a compact disk, a mini disk, or a DVD (Digital Versatile Disc).

[0100] Note that the functions of the training data acquisition unit 11, the external factor region segmentation unit 12, the external factor label data creation unit 13, and the control unit (not shown) may be partially implemented by dedicated hardware and partially implemented by software or firmware. For example, the training data acquisition unit 11 may be implemented by a processing circuit 101 as dedicated hardware, while the functions of the external factor region segmentation unit 12, the external factor label data creation unit 13, and the control unit (not shown) may be implemented by a processor 104 reading and executing programs stored in a memory 105. The label creation device 1 also includes an input interface device 102 and an output interface device 103 that communicate via wired or wireless communication with devices such as the diagnostic model creation device 2 or the storage device 4.

[0101] 13A and 13B also show examples of the hardware configuration of the diagnostic model creation device 2 according to the first embodiment. In the first embodiment, the functions of the diagnostic model creation unit 21 and a control unit (not shown) are realized by a processing circuit 101. That is, the diagnostic model creation device 2 includes a processing circuit 101 for controlling the creation of a diagnostic model based on the external factor label data created by the label creation device 1. The processing circuit 101 may be dedicated hardware as shown in FIG. 13A, or a processor 104 that executes a program stored in a memory as shown in FIG. 13B.

[0102] When processing circuitry 101 is dedicated hardware, processing circuitry 101 may be, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an ASIC (Application Specific Integrated Circuit), an FPGA (Field-Programmable Gate Array), or a combination thereof.

[0103] When the processing circuit is the processor 104, the functions of the diagnostic model creation unit 21 and the control unit (not shown) are realized by software, firmware, or a combination of software and firmware. The software or firmware is written as a program and stored in the memory 105. The processor 104 executes the functions of the diagnostic model creation unit 21 and the control unit (not shown) by reading and executing the program stored in the memory 105. That is, the diagnostic model creation device 2 includes the memory 105 for storing a program that, when executed by the processor 104, results in the execution of step ST4 in FIG. 8 described above. It can also be said that the program stored in the memory 105 causes a computer to execute the processing procedures or methods of the diagnostic model creation unit 21 and the control unit (not shown).

[0104] The diagnostic model creation device 2 includes an input interface device 102 and an output interface device 103 that communicate with devices such as the label creation device 1 or the storage device 4 via wired or wireless communication.

[0105] 13A and 13B also show examples of the hardware configuration of the degradation diagnosis device 3 according to the first embodiment. In the first embodiment, the functions of the diagnostic data acquisition unit 31, the model determination unit 32, the model acquisition unit 33, the feature extraction unit 34, the diagnosis unit 35, the diagnosis result output unit 36, and a control unit (not shown) are realized by a processing circuit 101. That is, the degradation diagnosis device 3 includes a processing circuit 101 for controlling the diagnostic model creation device 2 to perform a degradation diagnosis of the target device using a diagnostic model created based on the external factor label data created by the label creation device 1. The processing circuit 101 may be dedicated hardware as shown in FIG. 13A, or a processor 104 that executes a program stored in a memory as shown in FIG. 13B.

[0106] When processing circuitry 101 is dedicated hardware, processing circuitry 101 may be, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an ASIC (Application Specific Integrated Circuit), an FPGA (Field-Programmable Gate Array), or a combination thereof.

[0107] When the processing circuit is the processor 104, the functions of the diagnostic data acquisition unit 31, the model determination unit 32, the model acquisition unit 33, the feature extraction unit 34, the diagnosis unit 35, the diagnosis result output unit 36, and a control unit (not shown) are realized by software, firmware, or a combination of software and firmware. The software or firmware is written as a program and stored in the memory 105. The processor 104 reads and executes the program stored in the memory 105 to execute the functions of the diagnostic data acquisition unit 31, the model determination unit 32, the model acquisition unit 33, the feature extraction unit 34, the diagnosis unit 35, the diagnosis result output unit 36, and a control unit (not shown). In other words, the degradation diagnosis device 3 includes the memory 105 for storing a program that, when executed by the processor 104, results in the execution of steps ST21 to ST26 of FIG. 11 described above. In addition, the program stored in memory 105 can also be said to cause the computer to execute the processing procedures or methods of the diagnostic data acquisition unit 31, the model determination unit 32, the model acquisition unit 33, the feature extraction unit 34, the diagnosis unit 35, the diagnostic result output unit 36, and a control unit not shown.

[0108] It should be noted that the functions of the diagnostic data acquisition unit 31, the model determination unit 32, the model acquisition unit 33, the feature extraction unit 34, the diagnosis unit 35, the diagnosis result output unit 36, and a control unit (not shown) may be partially realized by dedicated hardware and partially realized by software or firmware. The degradation diagnosis device 3 also includes an input interface device 102 and an output interface device 103 that perform wired or wireless communication with devices such as the sensor 5 or the storage device 4.

[0109] As described above, the label creation device 1 according to the first embodiment includes a learning data acquisition unit 11 that acquires learning data in which feature amounts that can be used to create a diagnostic model used in the deterioration diagnosis of a target device that is a deterioration diagnosis target are associated with external factor data related to the driving environment, and a constant band, which is a range of values ​​of external factor data in the axial direction of the external factor data, in which the relationship with the corresponding feature amount is estimated to be constant, and a transition band, which is a range of values ​​of external factor data in which the relationship with the corresponding feature amount is estimated to change frequently, based on the learning data acquired by the learning data acquisition unit 11. The label creation device 1 is configured to include an external factor region segmentation unit 12 that segments the external factor data, and an external factor label data creation unit 13 that creates external factor label data in which the value ranges of the external factor data correspond to the external factor labels by assigning external factor labels to the constant bands of the external factor data, indicating that the relationship between the external factor data and the corresponding feature is constant, and assigning external factor labels to the transition bands of the external factor data, indicating that the relationship between the external factor data and the corresponding feature changes frequently. Therefore, the label creation device 1 can provide data that helps build a diagnostic model that prevents a decrease in the accuracy of diagnosing deterioration of a target device due to the operating environment. As a result, the label creation device 1 can contribute to building a diagnostic model that prevents a decrease in deterioration diagnosis of a target device due to the operating environment, and can contribute to achieving accurate deterioration diagnosis using the diagnostic model. The label creation device 1 can also contribute to reducing the time required for designing a diagnostic model. Furthermore, when creating external factor label data, the label creating device 1 focuses on the relationship between external factors and feature quantities based on sensor data when the target device is in a normal state. In other words, the learning data used by the label creating device 1 when creating external factor label data only needs to be data that corresponds to at least the relationship between external factors and feature quantities based on sensor data when the target device is in a normal state, so the effort required to create the learning data can be reduced.Furthermore, by providing external factor data, the label creating apparatus 1 can allow, for example, an operator to clearly understand which external factor caused the diagnostic model to switch and at what timing.

[0110] The diagnostic model creation device 2 according to the first embodiment includes a learning data acquisition unit 11 that acquires learning data in which feature quantities that can be used to create a diagnostic model for degradation diagnosis of a target device to be subjected to degradation diagnosis are associated with external factor data related to the operating environment; an external factor region division unit 12 that divides, based on the learning data acquired by the learning data acquisition unit 11, a range of values ​​of external factor data in an axial direction of the external factor data into a constant band, which is a range of values ​​of external factor data in which a relationship with a corresponding feature quantity is estimated to be constant, and a transition band, which is a range of values ​​of external factor data in which a relationship with a corresponding feature quantity is estimated to change frequently; and a range of values ​​of external factor data divided into a constant band with respect to the axis of the external factor data, in which a relationship between the external factor data in the range and the corresponding feature quantity is constant. The label creation device 1 includes an external factor label data creation unit 13 that creates external factor label data in which the range of values ​​of the external factor data is associated with the external factor labels, by assigning an external factor label indicating that the relationship between the external factor data in that range and the corresponding feature value changes frequently to a range of values ​​of the external factor data divided into transition bands, and assigning an external factor label indicating that the relationship between the external factor data in that range and the corresponding feature value changes frequently to a range of values ​​of the external factor data divided into transition bands, and a diagnostic model creation unit 21 that acquires learning data corresponding to the range of values ​​of the external factor data associated with the external factor labels indicating that the relationship between the external factor data in that range and the corresponding feature value is constant, based on the external factor label data created by the external factor label data creation unit 13 in the label creation device 1, and creates a diagnostic model using the acquired learning data. The diagnostic model creation device 2 creates a diagnostic model based on the external factor label data created by the label creation device 1, thereby enabling the construction of a diagnostic model that prevents a decline in the diagnosis of equipment deterioration caused by the operating environment.

[0111] The degradation diagnosis device 3 according to the first embodiment includes a learning data acquisition unit 11 that acquires learning data in which feature quantities that can be used to create a diagnostic model used in degradation diagnosis of a target device to be diagnosed are associated with external factor data related to the operating environment; an external factor region division unit 12 that divides, based on the learning data acquired by the learning data acquisition unit 11, a range of values ​​of external factor data in an axial direction of the external factor data into a constant band, which is a range of values ​​of external factor data in which the relationship with the corresponding feature quantities is estimated to be constant, and a transition band, which is a range of values ​​of external factor data in which the relationship with the corresponding feature quantities is estimated to change frequently; and an external factor label that indicates, for the axis of the external factor data, the range of values ​​of external factor data divided into the constant band is a range in which the relationship between the external factor data in that range and the corresponding feature quantities is constant. The label creating device 1 includes an external factor label data creating unit 13 that assigns external factor labels to the range of values ​​of the external factor data divided into transition bands, and assigns external factor labels to the range of values ​​of the external factor data divided into transition bands, indicating that the relationship between the external factor data in that range and the corresponding feature changes frequently, thereby creating external factor label data in which the value range of the external factor data is associated with the external factor labels; and the diagnostic model creating device 2 includes a diagnostic model creating unit 21 that acquires learning data corresponding to the range of values ​​of the external factor data associated with the external factor labels, indicating that the relationship between the external factor data in that range and the corresponding feature is constant, based on the external factor label data created by the external factor label data creating unit 13 in the label creating device 1, and creates a diagnostic model using the acquired learning data.

[0112] The degradation diagnosis device 3 according to the first embodiment is configured to include a diagnostic data acquisition unit 31 that acquires diagnostic data including sensor data relating to the operating state of the target equipment and external factor data relating to the operating environment of the target equipment; a model determination unit 32 that determines a diagnostic model to be used for degradation diagnosis based on the external factor data acquired by the diagnostic data acquisition unit 31 and data indicating the range of values ​​of the external factor data associated with the diagnostic model; a model acquisition unit 33 that acquires the diagnostic model determined by the model determination unit 32 to be used for degradation diagnosis; a feature extraction unit 34 that extracts features from the sensor data acquired by the diagnostic data acquisition unit 31; a diagnosis unit 35 that performs degradation diagnosis of the target equipment based on the features extracted by the feature extraction unit 34, using the diagnostic model acquired by the model acquisition unit 33 for the external factor data corresponding to the features; and a diagnosis result output unit 36 ​​that outputs the results of the degradation diagnosis of the target equipment by the diagnosis unit 35. The deterioration diagnosis device 3 performs deterioration diagnosis of the target equipment using a diagnostic model created taking into account the possibility that accurate diagnostic results may not be obtained before and after switching due to the operating environment, and therefore can perform accurate deterioration diagnosis.

[0113] Furthermore, the information processing system 6 according to the first embodiment is configured to include the above-described label creation device 1, diagnostic model creation device 2, and degradation diagnosis device 3. Therefore, the information processing system 6 can provide data that aids in the construction of a diagnostic model that prevents a decrease in the accuracy of diagnosing degradation of a target device due to the driving environment. Furthermore, the information processing system 6 can construct a diagnostic model that prevents a decrease in the accuracy of diagnosing degradation of a target device due to the driving environment, and can also realize accurate degradation diagnosis using the diagnostic model.

[0114] In addition, in the present disclosure, any of the components of the embodiments may be modified or omitted.

[0115] The label creation device according to the present disclosure can provide data that aids in the construction of a diagnostic model that prevents a decrease in the accuracy of diagnosing equipment deterioration caused by the operating environment.

[0116] 1 Label creation device, 11 Learning data acquisition unit, 12 External factor region division unit, 121 Inflection point extraction unit, 122 Provisional region setting unit, 123 Relational model creation unit, 124 Relational model evaluation unit, 125 Region determination unit, 13 External factor label data creation unit, 2 Diagnostic model creation device, 21 Diagnostic model creation unit, 211 Reliability determination unit, 3 Deterioration diagnosis device, 31 Diagnostic data acquisition unit, 32 Model determination unit, 33 Model acquisition unit, 34 Feature extraction unit, 35 Diagnosis unit, 36 Diagnostic result output unit, 4 Storage device, 41 Learning data storage unit, 42 External factor label data storage unit, 43 Diagnostic model storage unit, 5 Sensor, 6 Information processing system, 101 Processing circuit, 102 Input interface device, 103 Output interface device, 104 Processor, 105 Memory.

Claims

1. A learning data acquisition unit that acquires learning data in which features that can be used to create a diagnostic model used in the deterioration diagnosis of a target device that is to be diagnosed with deterioration are associated with external factor data related to the operating environment; and an external factor region division unit that divides, based on the learning data acquired by the learning data acquisition unit, a constant band, which is a range of values ​​of the external factor data in the axial direction of the external factor data, in which the relationship with the corresponding feature is estimated to be constant, and a transition band, which is a range of values ​​of the external factor data in which the relationship with the corresponding feature is estimated to change frequently; and an external factor label data creation unit that creates external factor label data in which the range of values ​​of the external factor data divided into the constant band is associated with the external factor label, the external factor label indicating that the relationship between the external factor data in that range and the corresponding feature is constant, and the external factor label indicating that the relationship between the external factor data in that range and the corresponding feature changes frequently, the external factor label data indicating that the relationship between the external factor data in that range and the corresponding feature changes frequently, the external factor label data indicating that the relationship between the external factor data in that range and the corresponding feature is frequently changed ... is associated with the external factor label, the external factor label data creating unit creating external factor label data in which the external factor data value range is associated with the external factor label, the external factor label indicating that the relationship between the external factor data in that range and the corresponding feature is constant, the external factor label indicating that the relationship between the external factor data in that range and the corresponding feature is constant, the external factor label indicating that the relationship between the external factor data in that range and the corresponding feature changes frequently, the external factor label indicating that the relationship between the external factor data in that range and the corresponding feature changes frequently, the external factor label data creating unit creating external factor label data in which the external factor data value range is associated with the external factor label.

2. The external factor region division unit comprises: an inflection point extraction unit which extracts a plurality of inflection points at which the relationship between the feature amount and the external factor data changes in the axial direction of the external factor data; a provisional region setting unit which selects a first inflection point and a second inflection point adjacent to each other from the plurality of inflection points extracted by the inflection point extraction unit, and sets a provisional region which is a candidate for the constant band in an inflection point range which is a range of values ​​of the external factor data between the selected first inflection point and the second inflection point; a relationship model creation unit which creates a relationship model which indicates the relationship between the feature amount and the external factor data based on the values ​​of the external factor data in the provisional region and the values ​​of the feature amount corresponding to the values ​​of the external factor data; and a relationship model evaluation unit which evaluates accuracy of the relationship model using the values ​​of the external factor data included in the provisional region and the values ​​of the feature amount corresponding to the values ​​of the external factor data.

2. The label creation device according to claim 1, further comprising: a region determination unit that, when the accuracy of the relational model is evaluated to be high, determines the provisional region when the accuracy of the relational model is evaluated to be high as the constant band, and determines a range of values ​​of the external factor data other than the determined constant band as the transition band, and, when the accuracy of the relational model is evaluated to be low, instructs the provisional region setting unit to re-set the provisional region by narrowing the range of the provisional region, and has the relational model creation unit create the relational model again and the relational model evaluation unit evaluate the accuracy of the relational model again.

3. The label creation device described in claim 1 or claim 2, characterized in that the external factor data may include a plurality of pieces of external factor data related to the driving environment, the feature extraction unit may extract a plurality of the features, the external factor region division unit divides the constant band and the transition band within the range of values ​​of the external factor data for each pair of the external factor data and the feature, and the external factor label data creation unit creates the external factor label data for each pair of the external factor data and the feature.

4. A diagnostic model creation device comprising: a label creation device according to any one of claims 1 to 3; and a diagnostic model creation unit which acquires, based on the external factor label data created by the external factor label data creation unit, the learning data corresponding to a range of values ​​of the external factor data to which the external factor labels indicating that the relationship between the range of the external factor data and the corresponding feature amount is constant, and creates the diagnostic model using the acquired learning data.

5. The diagnostic model creation device according to claim 4, characterized in that the diagnostic model creation unit associates, with the diagnostic model, data indicating a range of values ​​of the external factor data corresponding to the learning data used to create the diagnostic model.

6. A diagnostic model creation device as described in claim 4 or claim 5, characterized in that the diagnostic model creation unit is provided with a reliability determination unit that determines the reliability of the diagnostic model based on the number of learning data used to create the diagnostic model.

7. A degradation diagnosis device for performing the degradation diagnosis of the target equipment using the diagnostic model created by the diagnostic model creation device according to any one of claims 4 to 6, wherein the diagnostic model is associated with data indicating a value range of the external factor data corresponding to the learning data used in creating the diagnostic model, a diagnostic data acquisition unit for acquiring diagnostic data including sensor data on the operating state of the target equipment and the external factor data on the operating environment of the target equipment, a model determination unit for determining the diagnostic model to be used for the degradation diagnosis based on the external factor data acquired by the diagnostic data acquisition unit and data indicating a value range of the external factor data associated with the diagnostic model, a model acquisition unit for acquiring the diagnostic model determined by the model determination unit to be used for the degradation diagnosis, a feature extraction unit for extracting feature amounts from the sensor data acquired by the diagnostic data acquisition unit, and a diagnosis unit for performing the degradation diagnosis of the target equipment using the diagnostic model acquired by the model acquisition unit for the external factor data corresponding to the feature amount based on the feature amount extracted by the feature extraction unit. a diagnosis result output unit that outputs a result of the deterioration diagnosis of the target device by the diagnosis unit.

8. The degradation diagnosis device described in claim 7, characterized in that the diagnostic model is provided with data indicating its reliability determined based on the number of learning data used to create the diagnostic model, and the diagnostic result output unit outputs data indicating the reliability of the diagnostic model used in the degradation diagnosis together with the result of the degradation diagnosis.

9. An information processing system comprising: a label creation device according to any one of claims 1 to 3; a diagnostic model creation device according to any one of claims 4 to 6; and a degradation diagnosis device according to claim 7 or claim 8.

10. A step in which a learning data acquisition unit acquires learning data in which features that can be used to create a diagnostic model to be used in the deterioration diagnosis of a target device to be diagnosed are associated with external factor data related to the operating environment; and a step in which an external factor region division unit divides a range of values ​​of the external factor data in the axial direction of the external factor data into a constant band, which is a range of values ​​of the external factor data in which the relationship with the corresponding feature is estimated to be constant, and a transition band, which is a range of values ​​of the external factor data in which the relationship with the corresponding feature is estimated to change frequently, based on the learning data acquired by the learning data acquisition unit. an external factor label data creation unit assigns, on an axis of the external factor data, an external factor label indicating that a relationship between the external factor data in the constant band and the corresponding feature is constant, and assigns, on a value range of the external factor data divided into the transition band, an external factor label indicating that a relationship between the external factor data in the constant band and the corresponding feature changes frequently, thereby creating external factor label data in which the value range of the external factor data and the external factor label correspond to each other.

Images