Information processing device, information processing method, and program

Abstract

To provide technology that enables highly accurate anomaly diagnosis. [Solution] An information processing device according to one aspect of the present disclosure includes a preprocessing unit that performs a predetermined preprocessing according to the type of measurement value on first measurement data, which are among measurement data that measure the normal state of an abnormality target, and whose measurement values ​​constituting the measurement data are not constant or do not follow a predetermined distribution; and a model creation unit that creates an abnormality diagnosis model that models the normal state based on the first measurement data after the preprocessing and second measurement data other than the first measurement data.

Description

[Technical Field]

[0001] This disclosure relates to an information processing device, an information processing method, and a program. [Background technology]

[0002] A technology is known that performs anomaly diagnosis on equipment such as factories and plants, using measurement data from sensors installed on these facilities. In such anomaly diagnosis, a method called multivariate statistical process control (Non-Patent Literature 1) is often used. [Prior art documents] [Non-patent literature]

[0003] [Non-Patent Document 1] Manabu Kano, "Multivariate Statistical Process Management," Internet<URL:http: / / manabukano.brilliant-future.net / research / report / Report2005_MSPC.pdf> [Overview of the project] [Problems that the invention aims to solve]

[0004] However, multivariate statistical process control is based on continuous measurement data whose values ​​are stationary or follow a normal distribution under normal conditions. For example, it could not utilize measurement data where the values ​​were not stationary under normal conditions or did not follow a normal distribution. As a result, it was sometimes impossible to achieve sufficient accuracy in diagnosing anomalies.

[0005] This disclosure is made in view of the above points and aims to provide a technology that enables accurate anomaly diagnosis. [Means for solving the problem]

[0006] An information processing device according to one aspect of the present disclosure includes a preprocessing unit that performs a predetermined preprocessing according to the type of measurement value on first measurement data, which are measurement data that measure the normal state of an abnormality target, and whose measurement values ​​constituting the measurement data are not constant or do not follow a predetermined distribution; and a model creation unit that creates an abnormality diagnosis model that models the normal state based on the first measurement data after the preprocessing and second measurement data other than the first measurement data. [Effects of the Invention]

[0007] Technology is provided that enables highly accurate anomaly diagnosis. [Brief explanation of the drawing]

[0008] [Figure 1] This figure shows an example of the hardware configuration of the anomaly diagnosis device according to the first embodiment. [Figure 2] This figure shows an example of the functional configuration of an abnormality diagnosis device according to the first embodiment. [Figure 3] This is a flowchart showing an example of the model creation process according to the first embodiment. [Figure 4] This figure shows an example of measurement data where the measured values ​​are expressed as cumulative values. [Figure 5] This figure shows an example of measurement data where the measured value is represented as a digital pulse value. [Figure 6] This is a flowchart showing an example of an anomaly diagnosis process according to the first embodiment. [Figure 7] This figure shows an example of the functional configuration of an abnormality diagnosis device according to the second embodiment. [Figure 8] A flowchart showing an example (part 1) of the model creation process according to the second embodiment. [Figure 9] This is a flowchart showing an example (part 2) of the model creation process according to the second embodiment. [Figure 10] A flowchart showing an example (part 3) of the model creation process according to the second embodiment. [Modes for carrying out the invention]

[0009] The first and second embodiments of the present invention will be described in detail below with reference to the drawings. In each of the following embodiments, an anomaly diagnosis device 10 that achieves highly accurate anomaly diagnosis from measurement data of sensors installed in facilities such as factories and plants will be described. Hereinafter, facilities such as factories and plants that are the target of anomaly diagnosis will be referred to as "the object of diagnosis." However, the object of diagnosis is not limited to facilities such as factories and plants, but may also be, for example, equipment or devices. Specific examples of plants include petrochemical plants, power plants, steel plants, and food processing plants.

[0010] Here, the anomaly diagnosis device 10 according to each of the following embodiments uses a method called Multivariate Statistical Process Control (MSPC) to perform anomaly diagnosis on the target of diagnosis. In this case, the anomaly diagnosis device 10 according to each of the following embodiments performs a predetermined preprocessing on the measurement values ​​contained in the measurement data, which are not stationary or do not follow a normal distribution, to convert them into continuous and stationary or normally distributed measurement data. As a result, measurement data that could not be used for anomaly diagnosis by MSPC in the past can be used, and as a result, an improvement in the accuracy of anomaly diagnosis can be expected. Stationary measurement data means measurement data whose value is constant, but it does not necessarily have to be strictly constant. For example, if the value contained in the measurement data (i.e., the measurement value) is within a certain error, it may be considered "stationary measurement data". Similarly, the measurement values ​​contained in the measurement data do not need to strictly follow a normal distribution, and if they follow a normal distribution within a certain error, they may be considered "measurement data that follows a normal distribution". Furthermore, continuous means that it takes continuous values ​​for a certain unit quantity within a certain range of real numbers or integers.

[0011] However, multivariate statistical process control is just one example of a method for realizing anomaly diagnosis of the target of diagnosis, and the anomaly diagnosis device 10 in each of the following embodiments may realize anomaly diagnosis using methods other than multivariate statistical process control. Specific examples of methods for realizing anomaly diagnosis include, for example, unsupervised methods such as Isolation Forest and One Class SVM, and supervised methods such as neural networks.

[0012] Here, various sensors are installed on the object to be diagnosed, and these sensors measure physical quantities that represent the state of the object. The time-series data composed of these measured physical quantities is the measurement data. Hereafter, with N being the number of types of measured values, the measurement data composed of the nth (1≦n≦N)th measured value will also be called the "nth measurement data," and {x n We will represent this as (t)}. Also, the data consisting of the first to Nth measured values ​​will be called "operation data", and x(t) = (x1(t),···,x N We will represent this as (t), where t is the index representing the time. Note that x is the variable that takes the nth (1≦n≦N)th measurement. n These are also called, for example, "process variables" or "state variables."

[0013] For simplicity, in the following, under normal conditions for the subject of diagnosis, the first to M measurement data points will be considered stationary or follow a normal distribution, while the (M+1) to N measurement data points will be considered non-stationary or not follow a normal distribution.

[0014] In the following discussion, MSPC will be used as the method for diagnosing anomalies, but it is also possible to use, for example, Univariate Statistical Process Control (USPC) instead of MSPC.

[0015] [First Embodiment] The first embodiment will be described below.

[0016] <Example of hardware configuration of the abnormality diagnosis device 10 according to the first embodiment> An example of the hardware configuration of the anomaly diagnosis device 10 according to the first embodiment will be described with reference to Figure 1. Figure 1 is a diagram showing an example of the hardware configuration of the anomaly diagnosis device 10 according to the first embodiment.

[0017] As shown in Figure 1, the abnormality diagnosis device 10 according to the first embodiment includes an input device 101, a display device 102, an external I / F 103, a communication I / F 104, a RAM (Random Access Memory) 105, a ROM (Read Only Memory) 106, an auxiliary storage device 107, and a processor 108. Each of these hardware components is connected to each other via a bus 109 for communication.

[0018] The input device 101 is, for example, a keyboard, mouse, touch panel, or physical button. The display device 102 is, for example, a display or display panel. The abnormality diagnosis device 10 does not necessarily have to have at least one of the input device 101 and the display device 102.

[0019] External I / F 103 is an interface with external devices such as recording media 103a. Examples of recording media 103a include CD (Compact Disc), DVD (Digital Versatile Disk), SD memory card (Secure Digital memory card), and USB (Universal Serial Bus) memory card.

[0020] The communication interface 104 is an interface for connecting to a communication network. The RAM 105 is a volatile semiconductor memory (storage device) that temporarily holds programs and data. The ROM 106 is a non-volatile semiconductor memory (storage device) that can retain programs and data even when the power is turned off. The auxiliary storage device 107 is a non-volatile storage device such as an HDD (Hard Disk Drive), SSD (Solid State Drive), or flash memory. The processor 108 is a processing unit such as a CPU (Central Processing Unit) or GPU (Graphics Processing Unit).

[0021] Note that the hardware configuration shown in Figure 1 is just one example, and the hardware configuration of the anomaly diagnosis device 10 is not limited to this. For example, the anomaly diagnosis device 10 may have multiple auxiliary storage devices 107 or multiple processors 108, it may not have some of the hardware shown, or it may have various other hardware components besides the hardware shown.

[0022] <Example of the functional configuration of the abnormality diagnosis device 10 according to the first embodiment> An example of the functional configuration of the abnormality diagnosis device 10 according to the first embodiment will be described with reference to Figure 2. Figure 2 is a diagram showing an example of the functional configuration of the abnormality diagnosis device 10 according to the first embodiment.

[0023] As shown in Figure 2, the anomaly diagnosis device 10 according to the first embodiment includes an input unit 201, a pre-processing unit 202, a model creation unit 203, an anomaly diagnosis unit 204, and an output unit 205. Each of these units is realized, for example, by a process in which one or more programs installed in the anomaly diagnosis device 10 are executed by a processor 108 or the like. The anomaly diagnosis device 10 according to the first embodiment also includes a data storage unit 301 and a model storage unit 302. Each of these storage units is realized, for example, by a storage area such as an auxiliary storage device 107. However, for example, at least one of the storage units, the data storage unit 301 and the model storage unit 302, may be realized by a storage area such as a storage device (e.g., a storage device of a database server) that is connected to the anomaly diagnosis device 10 in a manner that allows communication.

[0024] The input unit 201 receives a model creation dataset, consisting of normal operating data of the vehicle being diagnosed, from the data storage unit 301 when creating an anomaly diagnosis model. The input unit 201 also receives the diagnostic target data, which is the operating data of the vehicle being diagnosed, when performing an anomaly diagnosis on the vehicle being diagnosed.

[0025] When creating an anomaly diagnosis model, the preprocessing unit 202 performs predetermined preprocessing on the measured values ​​of the M+1th to Nth measurement data included in the model creation dataset, according to the type of measured value. Similarly, when performing an anomaly diagnosis on the target of diagnosis, the preprocessing unit 202 performs predetermined preprocessing on the M+1th to Nth measurement values ​​included in the data to be diagnosed, according to the type of measured value. Hereinafter, the data on which preprocessing has been performed on the measured value of the nth (M+1≦n≦N)th measurement data will also be referred to as the "nth preprocessed measurement data". Similarly, the data on which preprocessing has been performed on the M+1th to Nth measurement values ​​of the data to be diagnosed will also be referred to as the "preprocessed data to be diagnosed".

[0026] When creating an anomaly diagnosis model, the model creation unit 203 creates an anomaly diagnosis model using MSPC based on the nth (1≦n≦M)th measurement data and the nth (M+1≦n≦N)th pre-processed measurement data included in the model creation dataset. The model creation unit 203 also saves the anomaly diagnosis model to the model storage unit 302.

[0027] The anomaly diagnosis unit 204 performs an anomaly diagnosis on the target of diagnosis based on the pre-processed data to be diagnosed and the anomaly diagnosis model stored in the model storage unit 302.

[0028] The output unit 205 outputs the results of the abnormality diagnosis performed by the abnormality diagnosis unit 204 to a predetermined output destination. The predetermined output destination is not limited to a specific destination, but examples include a display device 102 such as a display, the storage area of ​​the auxiliary storage device 107, other programs or devices, etc.

[0029] The data storage unit 301 stores the model creation dataset. Hereinafter, the model creation dataset will be defined as D = {x(t) | t = 1, ..., T}. Here, T is the index of the time that the driving data x(t) included in the model creation dataset D can take.

[0030] The model storage unit 302 stores the anomaly diagnosis model created by the model creation unit 203.

[0031] <Example of model creation process according to the first embodiment> The following describes an example of the model creation process for creating an anomaly diagnosis model, with reference to Figure 3. Figure 3 is a flowchart of an example of the model creation process according to the first embodiment. Generally, the model creation process is executed when the target for diagnosis is offline, but it is not limited to this; for example, it may be executed in the background when the target for diagnosis is online.

[0032] The input unit 201 inputs the model creation dataset D stored in the data storage unit 301 (step S101).

[0033] The preprocessing unit 202 performs predetermined preprocessing on the measured values of the measurement data from the (M + 1)-th to the N-th included in the model creation dataset D input in step S101 above, according to the type of the measured value (step S102). Specifically, the preprocessing unit 202 converts each measured value x n (t) included in the measurement data {x n (t)} by a filter realized by a transfer function corresponding to the type. Hereinafter, assuming that the transfer function is represented as F(z), the preprocessing in the case where the type of the measured value included in a certain measurement data {x n (t)} is an "integrated value" (or may be called an "accumulated value") and the case where it is a "digital pulse value" will be described.

[0034] (1) Case of integrated value The integrated value means the case where the measured value x n (t) at time t is represented as x n (t)=u n (t - 1)+···+u n (t) using certain predetermined measured values u n (t - 1),···,u n (1). The integrated value appears, for example, when the measured value x n (t) is some counter value (e.g., a counter value representing the number of occurrences of a warning, etc.). When the measured value x n (t) is a counter value representing the number of occurrences of a warning, examples of u(s) (1≦s≦t - 1) include measured values that take 1 when a warning occurs at time s and 0 when no warning occurs (that is, measured values representing the occurrence or non-occurrence of a warning). An example of measurement data in which the measured value x n (t) is represented by an integrated value is shown in FIG. 4. As shown in FIG. 4, when the measured value x n (t) at time t is represented by an integrated value, the measurement data {x n (t)} composed of those measured values x n(t) becomes measurement data that is non-steady and does not follow a normal distribution.

[0035] In this case, the preprocessing unit 202 performs preprocessing on each measurement value x n (t) using a transfer function representing differentiation and a transfer function representing smoothing (e.g., first-order lag, etc.). Specifically, with the transfer function representing differentiation as F1(z) = 1 - z -1 and the transfer function representing first-order lag as F2(z) = (1 - a) / (1 - az -1 )(where 0 < a < 1 is the weight), the preprocessing unit 202 uses the transfer function defined as F(z) = F1(z)F2(z) to obtain the preprocessed measurement value as x n '(t) = F(z)x n (t). Here, z represents the operation of advancing one time step, and z -1 represents the operation of delaying one time step. This means that the preprocessed measurement value x n (t) is the smoothed difference between the measurement value x n (t) and its previous value x n (t - 1).

[0036] As a result, it becomes possible to convert the measurement data {x n (t)} in which the measurement value x n (t) at time t is represented by an integrated value into measurement data {x n '(t)} that is generally steady and generally follows a normal distribution, and it can be used for creating an anomaly diagnosis model by MSPC.

[0037] (2) Digital pulse value The digital pulse value is, for example, the case where the measurement value x n (t) at time t takes 0 or 1. The digital pulse value appears, for example, as the ON / OFF of a signal such as a stop signal or a start signal, or a signal representing the occurrence of some event. An example of measurement data in which the measurement value x n (t) is represented by a digital pulse value is shown in FIG. 5. As shown in FIG. 5, when the measurement value x n (t) at time t is represented by a digital pulse value, the measurement data {x n (t)} composed of those measurement values xn (t) is stationary, but the measurement data does not follow a normal distribution.

[0038] In this case, the preprocessing unit 202 uses a transfer function representing the integral to determine each measured value x n Perform preprocessing on (t). Specifically, the transfer function representing the integral is F3(z)=1 / (1-z -1 The preprocessing unit 202 uses a transfer function defined as F(z)=F3(z) to determine the measured value after preprocessing, x n '(t)=F(z)x n Let (t) be this. n The previous value of (t) x n (t-1) and measured value x n (t) is the sum of the pre-processed measured value x n This means to set it to (t).

[0039] Furthermore, the preprocessing unit 202 uses a transfer function defined as F(z)=F2(z) to determine the measured value after preprocessing, x n '(t)=F(z)x n (t) may also be used. This is the measured value x n (t) is smoothed (specifically, with a first-order lag) to obtain the pre-processed measurement value x n This means to set it to (t).

[0040] This results in the measured value x at time t. n (t) is the measurement data represented by the cumulative value {x n (t)} is measured data {x} which is roughly stationary and follows a roughly normal distribution. n It becomes possible to convert it to '(t)', which can then be used to create anomaly diagnosis models using MSPC.

[0041] Note that the transfer functions described in (1) and (2) above are just examples. Depending on the type of measured value, it is possible to transform the measured value using filters implemented with various transfer functions. Generally, a transfer function is defined as F(z)=(az+b) / (cz+d) or F(z)=(bz) for predetermined a, b, c, and d. -1 +a) / (dz -1This is a function that can be expressed as a rational expression with respect to +c).

[0042] The model creation unit 203 creates an anomaly diagnosis model using MSPC based on the nth (1≦n≦M)th measurement data and the nth (M+1≦n≦N)th pre-processed measurement data (step S103). That is, the model creation unit 203 creates {(x1(t),···,x M (t),x M+1 '(t),···,x N Using '(t))|t=1,···,T}, an anomaly diagnosis model is created using MSPC. Here, for n=M+1,···,N, x n (t) is the measured value after preprocessing. Note that the method for creating an anomaly diagnosis model using MSPC is an existing method, so please refer to, for example, Non-Patent Document 1.

[0043] The model creation unit 203 saves the anomaly diagnosis model created in step S103 above to the model storage unit 302 (step S104).

[0044] <Example of abnormality diagnosis processing according to the first embodiment> The following describes an example of an anomaly diagnosis process for diagnosing an anomaly in the target of diagnosis, with reference to Figure 6. Figure 6 is a flowchart of an example of an anomaly diagnosis process according to the first embodiment. Generally, the anomaly diagnosis process is executed when the target of diagnosis is online. The data to be diagnosed is x(t)=(x1(t),···,x N (t)) is given.

[0045] The input unit 201 receives the given diagnostic data x(t) (step S201).

[0046] The preprocessing unit 202 performs predetermined preprocessing on the M+1th to Nth measurement values ​​included in the diagnostic target data x(t) input in step S201 above, according to the type of measurement value (step S202). At this time, the preprocessing unit 202 may, if necessary, use some of the data from the previous diagnostic target data x(t-1),...,x(1) and the previous preprocessed diagnostic target data x'(t-1),...,x'(1) to perform predetermined preprocessing on the M+1th to Nth measurement values ​​included in the diagnostic target data x(t), similar to step S102 in Figure 3. Hereinafter, the measurement values ​​x included in the diagnostic target data x(t) n (t)(M+1≦n≦N) The pre-processed measurement value x n Let '(t)' be (M+1≦n≦N). Also, the data to be diagnosed after preprocessing is x'(t)=(x1'(t),···,x N Let's assume (t).

[0047] The anomaly diagnosis unit 204 performs an anomaly diagnosis of the target based on the preprocessed data to be diagnosed x'(t) and the anomaly diagnosis model stored in the model storage unit 302 (step S203). For example, the anomaly diagnosis unit 204 uses the preprocessed data to be diagnosed x'(t) and the anomaly diagnosis model to determine predetermined anomaly index values ​​(e.g., Q statistic, T 2 After calculating the statistical value, if the abnormality index value exceeds a predetermined threshold, the abnormality diagnosis result is considered abnormal; otherwise, the abnormality diagnosis result is considered normal.

[0048] The output unit 205 outputs the abnormality diagnosis result obtained in step S103 to a predetermined output destination (step S204).

[0049] [Second Embodiment] The second embodiment will now be described. In the second embodiment, each transfer function that implements the filter used in preprocessing is expressed using parameters, and the case in which the values ​​of these parameters are adjusted using predetermined index values ​​(e.g., index values ​​for evaluating the normal distribution of the measured values ​​after preprocessing, index values ​​for evaluating the anomaly diagnosis performance of the anomaly diagnosis model) will be described. As a result, in the second embodiment, it is expected that an anomaly diagnosis model with high anomaly diagnosis performance can be created.

[0050] Hereafter, we will refer to the index value used to evaluate the normality of measured values ​​as the "normality index value," and the index value used to evaluate the anomaly diagnosis performance of an anomaly diagnosis model as the "anomaly diagnosis performance index value."

[0051] In the second embodiment, the differences from the first embodiment will be described primarily, and the explanation of components that may be the same as in the first embodiment will be omitted.

[0052] <Example of hardware configuration of the abnormality diagnosis device 10 according to the first embodiment> The hardware configuration of the anomaly diagnosis device 10 according to the second embodiment may be the same as that of the first embodiment, so its description will be omitted.

[0053] <Example of the functional configuration of the abnormality diagnosis device 10 according to the second embodiment> An example of the functional configuration of the anomaly diagnosis device 10 according to the second embodiment will be described with reference to Figure 7. Figure 7 is a diagram showing an example of the functional configuration of the anomaly diagnosis device 10 according to the second embodiment.

[0054] As shown in Figure 7, the anomaly diagnosis device 10 according to the second embodiment includes, in addition to the parts described in the first embodiment, an evaluation unit 206 and a parameter adjustment unit 207. The evaluation unit 206 and the parameter adjustment unit 207 are realized, for example, by a process in which one or more programs installed in the anomaly diagnosis device 10 are executed by a processor 108 or the like.

[0055] The input unit 201 obtains a model evaluation dataset from the data storage unit 301 for the evaluation unit 206 to calculate an anomaly diagnosis performance index value. The model evaluation dataset is a dataset used to evaluate the anomaly diagnosis performance of the anomaly diagnosis model. The model evaluation dataset consists of pairs (x(t), y(t)) of driving data x(t) and ground truth data y(t) that indicates whether the target was normal or abnormal when the driving data x(t) was obtained from the target. Hereinafter, the model evaluation dataset will be referred to as D'.

[0056] The evaluation unit 206 calculates at least one of the normal distribution index value and the abnormality diagnosis performance index value.

[0057] The parameter adjustment unit 207 adjusts the parameters of each transfer function that realizes the filter used for preprocessing by the preprocessing unit 202, based on at least one of the normal distribution index value and the anomaly diagnosis performance index value calculated by the evaluation unit 206.

[0058] The data storage unit 301 stores a model evaluation dataset D' in addition to the model creation dataset D.

[0059] <Example of model creation process according to the second embodiment> ≪Example of model creation process according to the second embodiment (Part 1)≫ Below, we will explain an example of a model creation process for creating an anomaly diagnosis model after adjusting the parameters of the transfer function using the normal distribution evaluation index value, with reference to Figure 8. Figure 8 is a flowchart of an example (part 1) of the model creation process according to the second embodiment.

[0060] The input unit 201 receives the model creation dataset D stored in the data storage unit 301 (step S301).

[0061] The preprocessing unit 202 performs predetermined preprocessing on the measured values ​​of the M+1th to Nth measurement data included in the model creation dataset D input in step S301, according to the type of the measured value (step S302). Specifically, the preprocessing unit 202 processes the measured data {x n Depending on the type of measurement value included in (t), a filter is implemented using the transfer function F(z;a) corresponding to that type, which filters each measurement value x n Transform (t), where a is the set of parameters included in the transfer function F(z;a). Hereafter, the set of parameters a will also simply be called the parameters.

[0062] Here, the transfer function F(z;a) can be expressed, for example, as follows:

[0063] • When the transfer function F(z;a) is a transfer function F1(z;a1) that represents the derivative. F1(z;a1)=1-a 11 z -1 Here, a1 = {a 11} is a set of parameters.

[0064] Furthermore, the transfer function F1(z;a1) representing the derivative can be expressed more generally as follows.

[0065] F1(z;a1)=a 10 +a 11 z -1 +a 12 z -2 +···+a 1m z -m Here, a1 = {a 10 ,a 11 ,a 12 ,···,a 1m} is a set of parameters.

[0066] • When the transfer function F(z;a) is a transfer function F2(z;a2) representing a first-order lag. F2(z;a2)=(1-a 20 ) / (1-a 21 z-1 ) Here, a2 = {a 20 ,a 21} is a set of parameters.

[0067] Furthermore, the transfer function F2(z;a2) can be generalized to represent the transfer function of the m-th order lag.

[0068] • When the transfer function F(z;a) is an integral transfer function F3(z;a3) F3(z;a3)=1 / (1-a 31 z -1 ) Here, a3 = {a 31} is a set of parameters.

[0069] Furthermore, since the transfer function F3(z;a3) is the inverse operation of the transfer function F1(z;a1) representing differentiation, it can be generalized in the same way as the transfer function F1(z;a1) representing differentiation.

[0070] The evaluation unit 206 calculates a normal distribution index value based on the pre-processed measurement data from the M+1th to the Nth (step S303). The normal distribution index value calculated in this step is then used to determine the normal distribution index value. (1) Let's assume that.

[0071] Note that the normality index value s (1) For example, values ​​obtained by evaluating the normality of the distribution of the M+1th to Nth preprocessed measurement data using the Kolmogorov-Smirnov method, QQ plot, or skewness / kurtosis can be used. Below are the normality index values ​​s. (1) The larger the value of this parameter, the closer the distribution of the pre-processed measurement data from the M+1th to the Nth position is to a normal distribution (i.e., the higher the normality).

[0072] The parameter adjustment unit 207 adjusts the normality index value s calculated in step S303 above. (1) Based on this, it is determined whether or not to adjust the parameter a of the transfer function F(z;a) (step S305). For example, the parameter adjustment unit 207 adjusts the normality index value s(1) is less than a predetermined threshold value th (1) it is determined that the parameter a is adjusted, and if not, it may be determined that the parameter a is not adjusted.

[0073] When it is determined in step S305 above that the parameter a is adjusted, the parameter adjustment unit 207 uses the normal distribution index value s calculated in step S303 above (1) to adjust the parameter a of the transfer function F(z; a) (step S305). That is, the parameter adjustment unit 207 uses the normal distribution index value s calculated in step S303 above (1) to update the value of the parameter a of the transfer function F(z; a) so that it becomes higher. The parameter adjustment unit 207, for example, uses the normal distribution index value s calculated in step S303 above (1) as an objective function, and may update the value of the parameter a of the transfer function F(z; a) so as to maximize the objective function. For solving the parameter a that maximizes the objective function, for example, a known optimization method such as the steepest descent method or the simplex method may be used.

[0074] On the other hand, when it is determined in step S305 above that the parameter a is not adjusted, the model creation unit 203 creates an abnormality diagnosis model (step S306) in the same manner as step S103 in FIG. 3.

[0075] The model creation unit 203 stores the abnormality diagnosis model created in step S306 above in the model storage unit 302 (step S307).

[0076] ≪Example of model creation process according to the second embodiment (part 2)≫ Hereinafter, an example of a model creation process for creating an abnormality diagnosis model after adjusting the parameters of a transfer function using an abnormality diagnosis performance index value will be described with reference to FIG. 9. FIG. 9 is a flowchart showing an example of a model creation process (part 2) according to the second embodiment.

[0077] Steps S401 to S402 in FIG. 9 may be the same as steps S301 to S302 in FIG. 8 respectively, so the description thereof is omitted.

[0078] Subsequent to step S402, the model creation unit 203 creates an abnormality diagnosis model (step S403) in the same manner as step S103 in FIG. 3.

[0079] The input unit 201 inputs the model evaluation dataset D' stored in the data storage unit 301 (step S404).

[0080] The evaluation unit 206 calculates an abnormality diagnosis performance index value for the abnormality diagnosis model created in step S403 using the model evaluation dataset D' input in step S404 above (step S405). That is, the evaluation unit 206 uses each (x(t), y(t)) ∈ D' to calculate an index value representing the prediction accuracy of the abnormality diagnosis of the abnormality diagnosis model as the abnormality diagnosis performance index value s (2) and calculates it. Note that as the abnormality diagnosis performance index value s (2) it is possible to use, for example, the F1 score or the like. Hereinafter, it is assumed that the higher the value of the abnormality diagnosis performance index value s (2) the higher the abnormality diagnosis performance of the abnormality diagnosis model.

[0081] The parameter adjustment unit 207 determines whether to adjust the parameter a of the transfer function F(z; a) based on the abnormality diagnosis performance index value s (2) calculated in step S405 above (step S406). For example, the parameter adjustment unit 207 determines to adjust the parameter a when the abnormality diagnosis performance index value s (2) is less than a predetermined threshold th (2) and determines not to adjust the parameter a otherwise.

[0082] If it is determined in step S406 above to adjust the parameter a, the parameter adjustment unit 207 uses the abnormality diagnosis performance index value s (2)Based on this, the parameter a of the transfer function F(z;a) is adjusted (step S407). That is, the parameter adjustment unit 207 adjusts the abnormality diagnosis performance index value s calculated in step S405 above. (2) The parameter a of the transfer function F(z;a) is updated to increase the value of the parameter a. The parameter adjustment unit 207 adjusts the abnormality diagnosis performance index value s calculated in step S405 above, for example. (2) With the objective function as F(z;a), the value of parameter a in the transfer function F(z;a) should be updated to maximize this objective function. Known optimization methods such as the steepest gradient method or the simplex method can be used to find the parameter a that maximizes this objective function.

[0083] On the other hand, if it is determined in step S406 that parameter a should not be adjusted, the model creation unit 203 saves the anomaly diagnosis model created in step S403 to the model storage unit 302 (step S408).

[0084] ≪Example of model creation process according to the second embodiment (part 3)≫ Below, we will explain an example of a model creation process for creating an anomaly diagnosis model by adjusting the parameters of the transfer function using both the normal distribution evaluation index value and the anomaly diagnosis performance index value, with reference to Figure 10. Figure 10 is a flowchart of an example (part 3) of the model creation process according to the second embodiment.

[0085] Steps S501 to S505 in Figure 10 can be treated the same as steps S301 to S305 in Figure 8, respectively, so their explanation is omitted.

[0086] If it is determined in step S504 that parameter a does not need to be adjusted, the model creation unit 203 creates an anomaly diagnosis model (step S506) in the same manner as in step S103 in Figure 3.

[0087] Steps S507 to S510 in Figure 10 can be treated the same as steps S404 to S407 in Figure 9, respectively, so their explanation is omitted.

[0088] If it is determined in step S509 that parameter a does not need to be adjusted, the model creation unit 203 saves the anomaly diagnosis model created in step S506 to the model storage unit 302 (step S511).

[0089] <Modified example of the second embodiment> In the model creation process shown in Figure 10, the normality index value s (1) After adjusting parameter a using the method, the abnormality diagnosis performance index value s (2) Parameter a was adjusted using this method, but this can be done simultaneously. That is, the evaluation unit 206 calculates the normality index value s (1) and abnormality diagnosis performance index value s (2) After calculating the normal distribution index value s, the parameter adjustment unit 207 adjusts the normal distribution index value s (1) and abnormality diagnosis performance index value s (2) Parameter a may be adjusted using the following: In this case, the parameter adjustment unit 207 may, for example, use αs (1) +(1-α)s (2) The objective function is defined as a weight (where α satisfies 0 ≤ α ≤ 1). The value of parameter a in the transfer function F(z;a) should be updated to maximize this objective function. To determine whether or not to adjust parameter a, one can use, for example, whether or not the value of the objective function has converged.

[0090] <Summary> As described above, the anomaly diagnosis device 10 according to the first and second embodiments converts measurement data that could not be used in conventional MSPC (i.e., measurement data where the measured values ​​are not steady under normal conditions or do not follow a normal distribution) so that it can be used in MSPC. Then, using this converted measurement data, an anomaly diagnosis model is created that models the normal state of the target to be diagnosed using the MSPC method. This is expected to enable the creation of an anomaly diagnosis model that can diagnose anomalies with higher accuracy than conventional methods.

[0091] The present invention is not limited to the embodiments specifically disclosed above, and various modifications, changes, and combinations with known technologies are possible as long as they do not deviate from the spirit described in the claims. [Explanation of Symbols]

[0092] 10. Anomaly Diagnosis Device 101 Input Device 102 Display device 103 External I / F 103a Recording medium 104 Communication I / F 105 RAM 106 ROM 107 Auxiliary storage 108 processors 109 Bus 201 Input section 202 Pre-processing section 203 Model Creation Department 204 Department of Abnormal Diagnosis 205 Output section 301 Data Storage Unit 302 Model Memory Unit

Claims

1. A preprocessing unit that performs a predetermined preprocessing according to the type of measurement value on first measurement data from which measurement values ​​constituting the measurement data are not constant or do not follow a predetermined distribution, among the measurement data that measure the normal state of the target of abnormal diagnosis, A model creation unit creates an abnormality diagnosis model that models the normal state based on the first measurement data after the preprocessing and the second measurement data other than the first measurement data. An information processing device having

2. The aforementioned preprocessing is a filtering process implemented with a predetermined transfer function, The aforementioned pre-processing unit, If the type of the measured value is an integrated value, a filtering process is performed using a transfer function representing differentiation and a transfer function representing smoothing. The information processing apparatus according to claim 1, wherein, when the measured value is a digital pulse value, it performs filtering using a transfer function that represents integration or a transfer function that represents smoothing.

3. The aforementioned pre-processing unit, Of the third measurement data that measures the state of the object to be diagnosed as abnormal, a fourth measurement data in which the measurement values ​​constituting the third measurement data are not steady or do not follow a predetermined distribution is subjected to a predetermined preprocessing according to the type of measurement value. The aforementioned information processing device is The information processing apparatus according to claim 2, further comprising: an abnormality diagnosis unit that performs an abnormality diagnosis on the target of abnormality diagnosis based on the fourth measurement data after the preprocessing and the abnormality diagnosis model.

4. The information processing apparatus according to any one of claims 1 to 3, wherein the distribution is a normal distribution.

5. The aforementioned preprocessing is a filtering process implemented by a predetermined transfer function that includes parameters, An evaluation unit calculates a predetermined index value that evaluates at least one of the distribution of the first measurement data after the preprocessing and the performance of the anomaly diagnosis model, A parameter adjustment unit updates the value of the parameter based on the predetermined index value, The information processing apparatus according to claim 1, having the following features.

6. The evaluation unit, The information processing apparatus according to claim 5, wherein at least one of a first index value that evaluates the normality of the distribution of the first measurement data after the preprocessing and a second index value that evaluates the anomaly diagnosis accuracy of the anomaly diagnosis model is calculated as the predetermined index value.

7. The parameter adjustment unit is The information processing apparatus according to claim 5 or 6, which updates the value of the parameter so that the value of the predetermined index value increases.

8. A preprocessing procedure is performed on first measurement data, which are measurement data that measure the normal state of the target of abnormal diagnosis, in which the measurement values ​​constituting the measurement data are not constant or do not follow a predetermined distribution, according to the type of measurement value. A model creation procedure for creating an abnormality diagnosis model that models the normal state based on the first measurement data after the preprocessing and the second measurement data other than the first measurement data, A method of information processing performed by a computer.

9. A preprocessing procedure is performed on first measurement data, which are measurement data that measure the normal state of the target of abnormal diagnosis, in which the measurement values ​​constituting the measurement data are not constant or do not follow a predetermined distribution, according to the type of measurement value. A model creation procedure for creating an abnormality diagnosis model that models the normal state based on the first measurement data after the preprocessing and the second measurement data other than the first measurement data, A program that causes a computer to execute something.

Images