DECLINE SYSTEM

Abstract

System for deterioration monitoring, comprehensive: a normal model creation unit (102) that creates a normal model for another device, which is a model for the other device that differs from a target device in a normal state, based on normal data of the other device, which are operating data of the other device in a normal state; a unit (104) for creating a deterioration detection model, which creates a deterioration detection model for the other device based on the normal data of the other device and deterioration data of the other device, wherein the deterioration detection model is a model for determining the deterioration of the other device from operating data of the other device and the deterioration data are operating data of the other device in a deterioration state; a unit (106) for restructuring a normal model, which creates a normal model for the target device, which is a model of the target device in a normal state, based on the normal model for the other device created by the normal model creation unit and normal target device data, which are operating data of the target device in a normal state; a unit (107) for restructuring a deterioration detection model, which creates a deterioration detection model for the target device based on the deterioration detection model for the other device created by the unit for creating a deterioration detection model and the normal model for the target device created by the unit (106) for restructuring a normal model, wherein the deterioration detection model for the target device is a model for determining deterioration of the target device from operational data of the target device; and a deterioration detection unit (109) that determines the deterioration of the target device based on operational data of the target device and the deterioration detection model for the target device created by the unit (107) for restructuring a deterioration detection model.

Description

TECHNICAL FIELD

[0001] The invention relates to a deterioration detection system that detects deterioration of a device due to aging. GENERAL STATE OF THE ART

[0002] Various devices such as devices in railway transport, devices in power generation plants or devices in production lines require preventive maintenance and methods for detecting the deterioration of the devices (deterioration detection methods) have been investigated.

[0003] Generally, to perform high-precision deterioration detection on a device, operating data must be collected in both a normal state and a deterioration state for each target device (hereinafter referred to as a target device). However, in real devices, deterioration occurs slowly for many devices, and in many cases, the timing of an abnormality cannot be accurately determined, making it difficult to collect sufficient operating data in a deterioration state.

[0004] Therefore, there is a method in which devices are grouped by model, etc., based on device specifications, and a diagnosis model is created for each group using operating data of all devices in the same group (see, for example, Patent Literature 1). When detecting deterioration of a target device, a group to which the target device belongs is first determined based on the specifications, and the deterioration is detected using a diagnosis model for the group. Therefore, deterioration detection can be performed even when the amount of operating data for individual devices is small. Patent Literature 2 describes an abnormality prediction system based on a deviation of measured data from normal data of a normal model.Patent literature 3 deals with the detection and identification of a disturbance in a process. LIST OF CITIONSPATENT LITERATURE Patent literature 1: JP 2016 - 091 378 A Patent literature 2: JP 2006 - 343 063 A Patent literature 3: US 2014 / 0365179 A1 SUMMARY OF THE INVENTION TECHNICAL TASK

[0005] As described above, in a conventional method, a diagnostic model is created using data of all devices belonging to the same group as a target device. Therefore, in the conventional method, to obtain a diagnostic model that accurately represents the target device, the operating data of each device in the group and the operating data of the target device must be similar operating data. If the operating data of each device in the same group as the target device includes data exhibiting different behaviors than the operating data of the target device, a created diagnostic model will not accurately represent the behavior of the target device, and accurate deterioration detection cannot be performed.

[0006] In addition, in general, even for identical models, for devices with large individual differences, large differences in assembly conditions, large differences in environmental conditions, or the like, the deterioration cannot be accurately detected even if a diagnostic model constructed using operating data of the identical models is applied to each device.

[0007] Furthermore, if the conditions are subdivided into small pieces to account for individual differences, assembly conditions, environmental conditions, or the like, and a diagnostic model is created for each condition, an immense amount and type of operating data is required to create each diagnostic model.

[0008] The invention is intended to achieve the above-described object, and an object of the invention is to provide a deterioration detection system that can detect deterioration of a target device without using operational data in a deterioration state of the target device. SOLUTION TO THE TASK

[0009] A deterioration detection system according to the present invention comprises: a normal model creation unit that creates a normal model for another device, which is a model for the other device different from a target device in a normal state, based on normal data of the other device, which is operating data of the other device in a normal state; a deterioration determination model creation unit that creates a deterioration determination model for the other device based on the normal data of the other device and deterioration data of the other device, wherein the deterioration determination model is a model for determining the deterioration of the other device from operating data of the other device, and the deterioration data is operating data of the other device in a deterioration state;a normal model restructuring unit that creates a normal model for the target device, which is a model of the target device in a normal state, based on the normal model for the other device created by the normal model creating unit and normal data of the target device, which is operation data of the target device in a normal state; a deterioration determination model restructuring unit that creates a deterioration determination model for the target device based on the deterioration determination model for the other device created by the deterioration determination model creating unit and the normal model for the target device created by the normal model restructuring unit;and a deterioration determination unit that determines deterioration of the target device based on operation data of the target device and the deterioration determination model for the target device created by the deterioration determination model restructuring unit; ADVANTAGEOUS EFFECTS OF THE INVENTION

[0010] According to the invention, with the configuration described above, deterioration of a target device can be detected without using operating data of the target device in a deterioration state. BRIEF DESCRIPTION OF THE DRAWINGS Fig. 1 is a diagram showing an exemplary configuration of a deterioration detection system according to a first embodiment of the invention. Fig. 2 is a diagram showing an exemplary configuration of a deterioration determination model creation unit in the first embodiment of the invention. Fig. 3 is a diagram showing an exemplary configuration of a normal model restructuring unit in the first embodiment of the invention. Fig. 4 is a diagram showing an exemplary configuration of a deterioration determination model restructuring unit in the first embodiment of the invention. Fig. 5 is a flowchart illustrating the exemplary operation of the deterioration detection system according to the first embodiment of the invention. Fig. 6 is a flowchart showing an exemplary operation of a normal model creating unit in the first embodiment of the invention. Fig. 7 is a flowchart illustrating the exemplary operation of the deterioration determination model creation unit in the first embodiment of the invention. Fig. 8 is a flowchart showing the exemplary operation of the normal model restructuring unit in the first embodiment of the invention. Fig. 9 is a flowchart illustrating the exemplary operation of the deterioration determination model restructuring unit in the first embodiment of the invention. Fig. 10 is a flowchart illustrating the exemplary operation of a deterioration determination unit in the first embodiment of the invention. Fig. 11 is a diagram showing an exemplary configuration of a deterioration detection system according to a second embodiment of the invention. Fig. 12 is a diagram showing an exemplary configuration of a model evaluation unit in the second embodiment of the invention. Fig. 13 is a flowchart showing an exemplary process of evaluating and modifying a deterioration determination model by the deterioration detection system according to the second embodiment of the invention. Fig. 14 is a flowchart illustrating the exemplary operation of the model evaluation unit in the second embodiment of the invention. Fig. 15 is a flowchart illustrating the exemplary operation of a model modification unit in the second embodiment of the invention. Fig. 16 is a diagram showing an exemplary hardware configuration of the deterioration detection system according to the first and second embodiments of the invention. DESCRIPTION OF EMBODIMENTS

[0011] Embodiments of the invention will be described in detail below with reference to the accompanying drawings. First embodiment.

[0012] Fig. 1 is a diagram showing an exemplary configuration of a deterioration detection system 1 according to a first embodiment of the invention.

[0013] The deterioration detection system 1 detects deterioration of a device serving as a target (hereinafter referred to as a target device). Note that the target device is, for example, a device in railway transportation such as a platform screen door, a device mounted on a train such as a brake, a device in power generation plants, or a device in production lines. As shown in Fig. 1, the deterioration detection system 1 includes a deterioration feature derivation unit 101, a normal model creation unit 102, a normal model storage unit 103, a deterioration determination model creation unit 104, a deterioration determination model storage unit 105, a normal model restructuring unit 106, a deterioration determination model restructuring unit 107, a deterioration determination model storage unit 108, a deterioration determination unit 109, and a detection result output unit 110.

[0014] The deterioration feature derivation unit 101 derives feature values ​​from input operation data. Operation data to be input to the deterioration feature derivation unit 101 may be any of operation data in a normal state (normal data) and operation data in a deterioration state (deterioration data) collected from another device (hereinafter referred to as the other device) different from a target device, operation data in a normal state (normal data) collected from the target device, and operation data (acquisition target data) collected from the target device. Note that the normal data and the acquisition target data of the target device are collected sequentially.The characteristic value to be derived by the unit for deriving deterioration characteristics 101 is also a characteristic value with which a deterioration can be made detectable.

[0015] The deterioration feature derivation unit 101 first derives feature values ​​from normal data and deterioration data collected from the other device. Next, the deterioration feature derivation unit 101 derives feature values ​​from pieces of normal data sequentially collected from the target device immediately after the target device starts operating. Next, the deterioration feature derivation unit 101 derives feature values ​​from pieces of acquisition target data sequentially collected from the target device.

[0016] Note that a method for deriving a feature value that can detect deterioration varies depending on the configuration, physical knowledge, preliminary analysis, or the like. A feature value derived from normal data by the deterioration feature derivation unit 101 is hereinafter referred to as normal feature value data. A feature value derived from deterioration data by the deterioration feature derivation unit 101 is referred to as deterioration feature value data. A feature value derived from detection target data by the deterioration feature derivation unit 101 is referred to as detection target feature value data. Note that the other device may be, for example, a test device or a device for which data is collected in advance in another environment.

[0017] It should be noted that Fig. 1 shows a case where the deterioration detection system 1 includes the deterioration characteristic derivation unit 101. However, this is not intended to be limiting, and the deterioration detection system 1 may not include the deterioration characteristic derivation unit 101.

[0018] The normal model creation unit 102 creates a model for the other device in a normal state (normal model) based on normal feature value data of the other device derived by the deterioration feature derivation unit 101. For example, a behavior that can be expressed in normal feature value data, a pattern that can be expressed in normal feature value data, a range of behavior or a range of pattern, and a correlation between feature values ​​derived from one or more pieces of normal feature value data represented by equations, graphs, or networks can be used as the normal model.If a normal model is created using a graph structure that represents a correlation between feature values, the graph structure can be represented, for example, with a covariance matrix of feature vectors or a precision matrix, which is the inverse of the covariance matrix.

[0019] The normal model storage unit 103 stores data representing the normal model for the other device created by the normal model creating unit 102.

[0020] The deterioration determination model creation unit 104 creates a deterioration determination model (deterioration determination model) for the other device from operation data of the other device based on the normal feature value data and deterioration feature value data of the other device derived by the deterioration feature deriving unit 101. In model creation, the deterioration determination model creation unit 104 may create a deterioration determination model for the other device based on the normal model for the other device read from the normal model storage unit 103, in addition to the above-described normal feature value data and deterioration feature value data of the other device. Details of the deterioration determination model creation unit 104 will be described later.

[0021] The deterioration determination model storing unit 105 stores data representing the deterioration determination model for the other device created by the deterioration determination model creating unit 104.

[0022] The deterioration detection system 1 creates a normal model for the other device and a deterioration detection model for the other device before creating a model for the target device and detecting deterioration of the target device.

[0023] The normal model restructuring unit 106 creates a model for the target device in a normal state (normal model) based on the normal model for the other device read from the normal model storage unit 103 and normal feature value data of the target device derived by the deterioration feature deriving unit 101. Details of the normal model restructuring unit 106 will be described later.

[0024] The deterioration determination model restructuring unit 107 creates a deterioration determination model (deterioration determination model) from operation data of the target device based on the deterioration determination model for the other device read from the deterioration determination model storage unit 105 and the normal model for the target device created by the normal model restructuring unit 106. In model creation, the deterioration determination model restructuring unit 107 may create a deterioration determination model for the target device based on normal feature value data of the target device derived by the deterioration feature deriving unit 101, in addition to the above-described deterioration determination model for the other device and the normal model for the target device.Details of the deterioration assessment model restructuring unit 107 will be described later.

[0025] The deterioration determination model storing unit 108 stores data representing the deterioration determination model for the target device created by the deterioration determination model restructuring unit 107.

[0026] The deterioration determination unit 109 determines the deterioration of the target device based on the detection target feature value data derived by the deterioration feature deriving unit 101 and the deterioration determination model for the target device read from the deterioration determination model storage unit 108. In the determination, the deterioration determination unit 109 determines whether the target device is normal or deteriorated by applying the deterioration determination model for the target device to the detection target feature value data.

[0027] The detection result output unit 110 outputs data representing a result of the determination made by the deterioration determination unit 109 to the outside.

[0028] Next, an exemplary design of the deterioration determination model creation unit 104 will be described with reference to Fig. 2. As described in Fig. 2, the deterioration determination model creation unit 104 includes a deterioration degree calculation unit 1041 and a discriminator creation unit 1042.

[0029] The deterioration degree calculation unit 1041 calculates a deterioration degree of normal feature value data and a deterioration degree of deterioration feature value data of the other device by applying a normal model for the other device to both the normal feature value data and the deterioration feature value data.

[0030] The discriminator creation unit 1042 creates a discriminator that can distinguish between the degree of deterioration of the normal feature value data and the degree of deterioration of the deterioration feature value data calculated by the deterioration degree calculation unit 1041, and uses the discriminator as a deterioration determination model.

[0031] Next, an exemplary design of the normal model restructuring unit 106 will be described with reference to Fig. 3 described.

[0032] As in Fig. 3, the normal model restructuring unit 106 includes a model unit 1061, a parameter updating unit 1062, and a convergence determining unit 1063.

[0033] In the model unit 1061, a normal model for the device other than a normal model for a target device is included.

[0034] The parameter updating unit 1062 updates a parameter of the normal model for the target device included in the modeling unit 1061 using normal feature value data of the target device. The parameter updating unit 1062 updates the parameter of the normal model for the target device each time normal data of the target device is collected, and the data is collected sequentially until the convergence determination unit 1063 determines that the parameter has converged.

[0035] Because there are individual differences between the other device and the target device, or differences in assembly conditions or environmental conditions, the data from the other device and the target device generally show different trends. Therefore, if a model created using data from the other device is applied to the target device in its original state, it is difficult to accurately capture the deterioration occurring in the target device.

[0036] The deterioration detection system 1 thus updates a parameter of a model created for the other device using normal data of the target device. The deterioration detection system 1 can therefore restructure a model to make it suitable for the target device.

[0037] Additionally, in many cases, it is difficult to collect a sufficient amount of data for training for all target devices in advance. The deterioration detection system 1 thus updates a parameter of a model previously created for the other device using data sequentially collected from the target device, thereby obtaining a model for the target device in-process. Consequently, even if sufficient data from the target device is not available in advance, the deterioration detection system 1 can restructure a model with high accuracy.

[0038] The convergence determination unit 1063 determines whether the parameter updated by the parameter update unit 1062 has converged. Each time normal data of the target device is collected, with the data being collected one by one, the convergence determination unit 1063 determines whether the parameter of the normal model updated by the parameter update unit 1062 has converged. The convergence determination unit 1063 repeats this process until it determines that the parameter has converged.

[0039] Note that, for example, there is a method in which the update is completed when a certain number of parameter updates have been performed. However, with this method, there is a possibility that an update is completed before the parameter converges, that is, before a model is restructured. Thus, the convergence determination unit 1063 performs a parameter convergence determination, which enables the deterioration detection system 1 to create an accurate model.

[0040] The parameter updating unit 1062 and the convergence determination unit 1063 start parameter updating and parameter convergence determination when collecting data from the target device in normal operation.

[0041] The parameter update unit 1062 terminates the parameter update when it determines that the parameter has converged, or when the target device has been running for a certain period of time and the collected data is highly likely to be non-normal data. Whether or not the collected data is highly likely to be non-normal data is determined based on knowledge of the device design or physical knowledge, etc.

[0042] Next, an exemplary design of the unit for restructuring a deterioration determination model 107 is described with reference to Fig. 4 described.

[0043] As in Fig. 4, the deterioration determination model restructuring unit 107 includes a model unit 1071, a deterioration degree calculation unit 1072, a deterioration determination unit (second deterioration determination unit) 1073, and a model updating unit 1074.

[0044] In the model unit 1071, a deterioration determination model for the other device is included as a deterioration determination model for the target device.

[0045] The deterioration degree calculation unit 1072 calculates a deterioration degree of normal feature value data of the target device by applying a normal model for the target device to the normal feature value data. The deterioration degree calculation unit 1072 calculates the deterioration degree until the normal model restructuring unit 106 completes model creation. Each time normal data of the target device is collected, with the data being collected one by one, the deterioration degree calculation unit 1072 calculates a deterioration degree of normal feature value data using a parameter of a normal model updated by the parameter updating unit 1062.

[0046] The deterioration determination unit 1073 performs a deterioration determination by applying the deterioration determination model for the target device included in the model unit 1071 to the degree of deterioration calculated by the deterioration degree calculation unit 1072.

[0047] The model updating unit 1074 updates the deterioration determination model for the target device included in the model unit 1071 based on a result of the determination made by the deterioration determination unit 1073.

[0048] Next, the exemplary operation of the deterioration detection system 1 according to the first embodiment will be described with reference to Fig. 5 described.

[0049] In the conventional method, since only model information of a target device is used, accurate deterioration detection cannot be performed due to factors such as device-specific individual differences. While it is difficult to collect deterioration data of a target device using an actual device, it may be possible to sequentially collect normal data of the target device. Therefore, the deterioration detection system 1 according to the first embodiment modifies the normal model and the deterioration detection model for the other device using normal data of a target device, thereby creating a normal model and a deterioration detection model that can more accurately represent the target device. Therefore, the deterioration detection system 1 according to the first embodiment improves the accuracy of deterioration detection.

[0050] In the exemplary operation of the deterioration detection system 1 according to the first embodiment, as shown in Fig. 5, first, the deterioration feature deriving unit 101 derives feature values ​​from inputted operation data (step ST501).

[0051] The normal model creation unit 102 then creates a normal model for the other device based on normal feature value data of the other device derived by the deterioration feature derivation unit 101 (step ST502). Data representing the normal model for the other device created by the normal model creation unit 102 is stored in the normal model storage unit 103. The detailed exemplary operation of the normal model creation unit 102 will be described later.

[0052] The deterioration determination model creation unit 104 then creates a deterioration detection model for the other device based on the normal feature value data and deterioration feature value data of the other device derived by the deterioration feature deriving unit 101 and the normal model for the other device read from the normal model storage unit 103 (step ST503). Data representing the deterioration determination model for the other device created by the deterioration determination model creation unit 104 is stored in the deterioration determination model storage unit 105. The detailed exemplary operation of the deterioration determination model creation unit 104 will be described later.

[0053] The normal model restructuring unit 106 then creates a normal model for a target device based on the normal model for the other device read from the normal model storage unit 103 and normal feature value data of the target device derived by the deterioration feature deriving unit 101 (step ST504). The detailed exemplary operation of the normal model restructuring unit 106 will be described later.

[0054] The deterioration determination model restructuring unit 107 then creates a deterioration determination model for the target device based on the deterioration determination model for the other device read from the deterioration determination model storage unit 105, the normal model for the target device created by the normal model restructuring unit 106, and normal feature value data of the target device derived by the deterioration feature deriving unit 101 (step ST505). Data representing the deterioration determination model for the target device created by the deterioration determination model restructuring unit 107 is stored in the deterioration determination model storage unit 108.The detailed exemplary operation of the deterioration determination model restructuring unit 107 will be described later.

[0055] The deterioration determination unit 109 then determines deterioration of the target device based on detection target feature value data derived by the deterioration feature deriving unit 101 and the deterioration determination model for the target device read from the deterioration determination model storage unit 108 (step ST506).

[0056] The detection result output unit 110 then outputs data representing a result of the determination made by the deterioration determination unit 109 to the outside (step ST507).

[0057] Next, the exemplary operation of the normal model creating unit 102 will be described with reference to Fig. 6 described.

[0058] In the exemplary operation of the normal model creation unit 102 as shown in Fig. 6, the normal model creating unit 102 first acquires normal feature value data of the other device derived by the deterioration feature deriving unit 101 (step ST601).

[0059] The normal model creating unit 102 then creates a normal model for the other device based on the normal feature value data of the other device (step ST602).

[0060] The normal model creating unit 102 then outputs data representing the normal model for the other device (step ST603).

[0061] Next, the exemplary operation of the deterioration determination model creation unit 104 will be described with reference to Fig. 7 described.

[0062] In the exemplary operation of the unit for creating a deterioration determination model 104, as described in Fig. 7, the deterioration determination model creation unit 104 first acquires normal feature value data and deterioration feature value data of the other device derived by the deterioration feature deriving unit 101 (step ST701).

[0063] The deterioration determination model creation unit 104 reads out a normal model for the other device stored in the normal model storage unit 103 (step ST702).

[0064] The deterioration degree calculation unit 1041 then calculates a deterioration degree of the normal feature value data and a deterioration degree of the deterioration feature value data of the other device by applying the normal model for the other device to both the normal feature value data and the deterioration feature value data (step ST703).

[0065] As a method for calculating a degree of deterioration, for example, when the normal model is a graph structure representing a correlation between feature values, the calculation method using the following equation (1) can be considered. In equation (1), x(k) ∈ R rrepresents a k-th r-dimensional feature vector in all feature value data, a(k) represents the degree of anomaly (degree of deterioration) of x(k), and P(x) represents a probability distribution of x. k represents a time point at which data is acquired, and the value of k is sequentially updated each time data is acquired. a(k)=−log(P(x(k)))

[0066] Assuming a normal distribution P(x(k)) = N(x(k) |µ, Σ), the degree of deterioration is represented by the Mahalanobis distance of the following equation (2). Where µ represents the mean of x and Σ represents the covariance matrix of x. a(k)=(x(k)−μ)T∑−1(x(k)−μ)

[0067] The method for calculating a degree of deterioration described above is only an example and other methods may be used and the degree of deterioration may be a multidimensional vector.

[0068] The discriminator creation unit 1042 then creates a discriminator that can distinguish between the degree of deterioration of the normal feature value data and the degree of deterioration of the deterioration feature value data calculated by the deterioration degree calculation unit 1041, and uses the discriminator as a deterioration determination model (step ST704).

[0069] In a method of constructing a discriminator, for example, when the degree of deterioration is calculated using equation (2), a discriminator can be obtained by calculating a threshold that most effectively distinguishes between the degree of deterioration of normal feature value data and the degree of deterioration of deterioration feature value data by applying a threshold optimization curve to the degree of deterioration.

[0070] The deterioration determination model creation unit 104 then outputs data representing the deterioration determination model for the other device (step ST705).

[0071] Next, the exemplary operation of the normal model restructuring unit 106 will be described with reference to Fig. 8. In the exemplary operation of the unit for restructuring a normal model 106, as shown in Fig. As shown in Figure 8, the normal model restructuring unit 106 first reads a normal model for the other device stored in the normal model storage unit 103 (step ST801). The normal model for the other device is then included in the model unit 1061 as a normal model for a target device.

[0072] The normal model restructuring unit 106 acquires normal feature value data of the target device as training data (step ST802).

[0073] The parameter updating unit 1062 then updates a parameter of the normal model for the target device included in the model unit 1061 using the training data (step ST803).

[0074] For example, assume that the normal model included in the model unit 1061 is a graph structure representing a correlation between feature values, and a sample mean, a sample covariance matrix, and a graph structure obtained with parameter updating are represented by µ (dach), Σ (dach), and A (dach), respectively.

[0075] µ (dach) (k), which is a sample mean at the current time, is calculated here as shown in the following equation (3). In equation (3), µ represents an i-element of µ. μ^(k)=1k[kμi(k−1)+x(k)]

[0076] Σ (dach) (k), which is a sample covariance matrix at the current time, is calculated as shown in the following equation (4). In equation (4), σ ij 2 represents an (i, j)-element of Σ (dach). σij2(k)=1k[kσij2(k−1)+(xi(k)−μi(k−1))(xj(k)−μj(k))] 1≤i,j≤r

[0077] A(dach)(k), which is a graph structure at the current time, is calculated from µ(dach)(k), which is a sample mean at the current time, Σ(dach)(k), which is a sample covariance matrix at the current time, and A(dach)(k-1), which is a graph structure obtained last time. For the initial values ​​of µ(dach), Σ(dach), and A(dach), the value of P, which is a normal model for the other device, is used.

[0078] For A (dach) (k), when a covariance matrix is ​​used as the graph structure, A (dach) (k) = Σ (dach) (k), and when a precision matrix is ​​used as the graph structure, A (dach) (k) = Σ (dach) -1 (k).

[0079] When a sparse graph is used as the graph structure, L1 optimization (the following expression (5)) solves an optimization problem with Σ (dach) = Σ (dach) (k), where AA (dach) (k-1) is the initial value of which A (dach) (k) can be obtained. In expression (5), ρ represents a regularization parameter. maxΛ[In(detΛ)−tr(∑^Λ)−ρ‖Λ‖1]

[0080] The normal model restructuring unit 106 then outputs data representing the normal model for the target device (step ST804).

[0081] The convergence determination unit 1063 then determines whether the parameter updated by the parameter updating unit 1062 has converged (step ST805).

[0082] If the convergence determination unit 1063 determines that the parameter has not converged at this step ST805, the flow returns to step ST802 and repeats the above-described process. On the other hand, if the convergence determination unit 1063 determines that the parameter has converged at step ST805, the flow ends.

[0083] An update of the normal model by the normal model restructuring unit 106 ends when the parameter of the normal model has converged and the value is almost unchanged. As a method for determining whether the parameter of the normal model has converged, when the normal model is represented by, for example, A (dach) (k) which is a square matrix r × r, it can be determined that the parameter has converged if A (dach) (k) satisfies the following expression (6). In expression (6), ε is a sufficiently small constant value, and tr(·) represents the sum (trace) of diagonal elements of the matrix. |tr(Λ^(k))−tr(Λ^(k−1))|≤ε

[0084] In the normal model restructuring unit 106, there may be a situation where normal data of the target device can no longer be acquired before the convergence determination unit 1063 determines that the parameter has converged. In this case, even if the parameter has not converged, the normal model restructuring unit 106 outputs data representing a normal model for the target device calculated immediately before normal data can no longer be acquired.

[0085] Next, the exemplary operation of the deterioration determination model restructuring unit 107 will be described with reference to Fig. 9 described.

[0086] In the exemplary operation of the unit for restructuring a deterioration assessment model 107, as described in Fig. As shown in Figure 9, the deterioration determination model restructuring unit 107 first reads a deterioration determination model for the other device stored in the deterioration determination model storage unit 105 (step ST901). Then, the model unit 1071 contains the deterioration determination model for the other device as the deterioration determination model for the target device.

[0087] The deterioration determination model restructuring unit 107 acquires normal feature value data of the target device as training data (step ST902).

[0088] The deterioration determination model restructuring unit 107 reads out a normal model for the target device created by the normal model restructuring unit 106 (step ST903).

[0089] The deterioration degree calculation unit 1072 then calculates a deterioration degree of the training data by applying the normal model for the target device to the training data (step ST904).

[0090] In this calculation, the deterioration determination model restructuring unit 107 calculates a(k), which is the degree of deterioration of x(k), which is normal feature value data of the target device input at the current time, by applying a normal model for the target device updated using x(k) to x(k).

[0091] The deterioration determination unit 1073 then performs a deterioration determination by applying the deterioration determination model included in the model unit 1071 to the degree of deterioration calculated by the deterioration degree calculation unit 1072 (step ST905).

[0092] The model updating unit 1074 then updates the deterioration determination model included in the model unit 1071 based on a result of the determination made by the deterioration determination unit 1073 (step ST906).

[0093] In the update, the deterioration determination model restructuring unit 107 updates the deterioration determination model by comparing P'(a), which is a probability distribution of the degree of deterioration of normal feature value data of the other device, with P(a), which is a probability distribution of the degree of deterioration of normal feature value data of the target device, and updates a separation plane of a discriminator based on a difference between the probability distributions of the other device and the target device.

[0094] As a method for updating a dividing plane, for example, a method can be considered in which, between P'(a) and P(a), which are probability distributions, a difference between the probability distributions is quantified using Kullback-Leibler divergence, etc., and the dividing plane transition occurs to a greater extent as the difference becomes larger.

[0095] As for the direction along which the parting plane transition occurs, for example, a method is used in which X, which is an error-prone detection rate of deterioration detection, is calculated in advance for the other device, and Y, which is an error-prone detection rate of deterioration detection, is calculated by performing deterioration detection on sequentially collected training data using a current deterioration detection model, and then transitioning in a direction in which Y becomes smaller when X < Y and in a direction in which Y becomes larger when X > Y.

[0096] The deterioration determination model restructuring unit 107 then determines whether a parameter of the normal model for the target device has converged (step ST907).

[0097] If the deterioration determination model restructuring unit 107 determines that the parameter of the normal model for the target device has not converged at this step ST907, the flow returns to step ST902 and repeats the above-described operation.

[0098] On the other hand, when the deterioration determination model restructuring unit 107 determines at this step ST907 that the parameter of the normal model for the target device has converged, the deterioration determination model restructuring unit 107 outputs data representing the deterioration determination model for the target device (step ST908).

[0099] Next, the exemplary operation of the deterioration determination unit 109 will be described with reference to Fig. 10 described.

[0100] In the exemplary operation of the deterioration determination unit 109 as shown in Fig. 10, the deterioration determination unit 109 first acquires detection target feature value data of the target device derived by the deterioration feature deriving unit 101 (step ST1001).

[0101] Subsequently, the deterioration determination unit 109 reads out a deterioration determination model for the target device stored in the deterioration determination model storage unit 108 (step ST1002).

[0102] Subsequently, the deterioration determination unit 109 determines whether or not there is deterioration of the target device by applying the deterioration determination model for the target device to the detection target feature value data (step ST1003).

[0103] Subsequently, the deterioration determination unit 109 outputs data representing a result of the determination (step ST1004).

[0104] As described above, the deterioration detection system 1 according to the first embodiment includes the normal model creation unit 102 that creates a normal model for the other device based on normal data of the other device; the deterioration determination model creation unit 104 that creates a deterioration determination model for the other device based on the normal data and deterioration data of the other device; the normal model restructuring unit 106 that creates a normal model for a target device based on the normal model created by the normal model creation unit 102 and normal data of the target device;the deterioration determination model restructuring unit 107, which creates a deterioration determination model for the target device based on the deterioration determination model created by the deterioration determination model creation unit 104 and the normal model created by the normal model restructuring unit 106; and the deterioration determination unit 109, which determines deterioration of the target device based on operating data of the target device and the deterioration determination model created by the deterioration determination model restructuring unit 107. With this configuration, the deterioration detection system 1 according to the first embodiment can detect deterioration of the target device without using deterioration data of the target device. Second embodiment.

[0105] In the deterioration detection system 1 according to the first embodiment, it is assumed that immediately after the target device is mounted and started to operate, there is no deterioration of the target device, and the operating data collected from the target device is normal. However, even immediately after the target device is mounted and started to operate, the operating data collected from the device may not be normal due to an initial failure, etc.

[0106] The deterioration detection system 1 can thus be supplemented with the following function: the accuracy of a deterioration detection is calculated based on a deterioration detection model for a target device, which is sequentially updated; an alarm is issued when the accuracy is outside an acceptable range; and the deterioration detection model is modified based on the result of a review by an inspector or quality inspector. The deterioration detection system 1 with the above-described function is described below.

[0107] Fig. Fig. 11 is a diagram showing an exemplary configuration of a deterioration detection system 1 according to a second embodiment of the invention. The deterioration detection system 1 according to the second embodiment shown in Fig. 11 is obtained by supplementing the deterioration detection system 1 according to the first embodiment shown in Fig. 1, to obtain a model evaluation unit 111, an input acceptance unit 112, a model modification unit 113, and a normal model storage unit 114. Other components are identical to those in the first embodiment, and the same reference numerals are used, and only different components will be described.

[0108] The model evaluation unit 111 issues an alarm when the accuracy of a deterioration determination is outside an allowable range based on normal feature value data of a target device derived by the deterioration feature deriving unit 101 and a deterioration determination model created by the deterioration determination model restructuring unit 107. An exemplary configuration of the model evaluation unit 111 will be described later.

[0109] The input accepting unit 112 accepts, after an alarm is issued by the model evaluation unit 111, inputs indicating a judgment as to whether the alarm is true or false.

[0110] The model modification unit 113 modifies the deterioration determination model created by the deterioration determination model restructuring unit 107 based on the inputs accepted by the input acceptance unit 112. In the modification, the model modification unit 113 may, for example, modify the deterioration determination model created by the deterioration determination model restructuring unit 107 based on a normal model read from the normal model storage unit 114.

[0111] The deterioration determination model storage unit 108 stores data representing the deterioration determination model modified by the model modification unit 113.

[0112] The normal model storage unit 114 stores data representing a normal model for the target device created by the normal model restructuring unit 106. The normal model storage unit 114 is not an essential part of the deterioration detection system 1 and need not be used.

[0113] Next, an exemplary configuration of the model modification unit 111 will be described with reference to Fig. 12 described.

[0114] As in Fig. 12, the model evaluation unit 111 includes a deterioration determination unit 1111, an accuracy calculation unit 1112, a comparison unit 1113, and an alarm output unit 1114.

[0115] The deterioration determination unit 1111 performs deterioration determination by applying a deterioration determination model for a target device to normal feature value data of the target device.

[0116] The accuracy calculation unit 1112 calculates the accuracy of the deterioration determination based on a result of the deterioration determination made by the deterioration determination unit 1111.

[0117] The comparison unit 1113 determines whether the deterioration determination accuracy calculated by the accuracy calculation unit 1112 is within an allowable range. The allowable range is determined based on, for example, the deterioration determination accuracy value obtained when a deterioration determination model for the other device is applied in advance to normal feature value data of the other device.

[0118] The alarm output unit 1114 outputs an alarm externally when the comparison unit 1113 determines that the accuracy of the deterioration determination is outside the allowable range.

[0119] Next, an exemplary process for evaluating and modifying a deterioration determination model by the deterioration detection system 1 according to the second embodiment will be described with reference to FIG. Fig. 13 described.

[0120] In the exemplary process for evaluating and modifying a deterioration determination model by the deterioration detection system 1 according to the second embodiment, as shown in Fig. As shown in FIG. 13, first, the model evaluation unit 111 issues an alarm based on the normal feature value data of a target device derived by the deterioration feature deriving unit 101 and a deterioration determination model created by the deterioration determination model restructuring unit 107 when the accuracy of a deterioration determination is outside an allowable range (step ST1301). The detailed exemplary operation of the model evaluation unit 111 will be described later.

[0121] The input accepting unit 112 then accepts inputs indicating whether the alarm is true or false after the model evaluation unit 111 issues an alarm (step ST1302). Thus, after the model evaluation unit 111 issues an alarm, an inspector or quality inspector performs an inspection on the target device and makes an input indicating whether the alarm is erroneous or not.

[0122] The model modification unit 113 then modifies the deterioration determination model created by the deterioration determination model restructuring unit 107 based on the inputs accepted by the input acceptance unit 112 (step ST1303). The detailed exemplary operation of the model evaluation unit 113 will be described later.

[0123] Next, the exemplary operation of the model evaluation unit 111 will be described with reference to Fig. 14 described.

[0124] In the exemplary operation of the model evaluation unit 111, as described in Fig. 14, first, the model evaluation unit 111 acquires normal feature value data of the target device (step ST1401).

[0125] In addition, the model evaluation unit 111 reads out a deterioration determination model for the target device created by the deterioration determination model restructuring unit 107 (step ST1402).

[0126] The deterioration determination unit 1111 then performs deterioration determination by applying the deterioration determination model for the target device to the normal feature value data of the target device (step ST1403).

[0127] The accuracy calculation unit 1112 then calculates the accuracy of the deterioration determination made by the deterioration determination unit 1111 based on a result of the deterioration determination (step ST1404).

[0128] The comparison unit 1113 then determines whether the accuracy of the deterioration determination calculated by the accuracy calculation unit 1112 is within an allowable range (step ST1405).

[0129] If the comparison unit 1113 determines at this step ST1405 that the accuracy of the deterioration determination is within the allowable range, the flow returns to step ST1401 and repeats the above-described process.

[0130] For example, data from the target device that was flagged as normal may actually be abnormal due to an initial failure, etc. In this case, the deterioration detection system 1 can create an accurate model by determining whether the accuracy of a deterioration detection is within the acceptable range and modifying the flag if the accuracy is outside the acceptable range.

[0131] The comparison unit 1113 continues to sequentially determine whether the accuracy of a deterioration detection is within the allowable range while data labeled "normal" is being input from the target device. The comparison unit 1113 then terminates the process when no more data labeled "normal" is input.

[0132] On the other hand, when the comparison unit 1113 determines at step ST1405 that the accuracy of the deterioration determination is out of the allowable range, the alarm output unit 1114 outputs an alarm externally (step ST1406).

[0133] Next, the exemplary operation of the model modification unit 113 will be described with reference to Fig. 15. In the exemplary operation of the model modification unit 113 as shown in Fig. 15, the model modification unit 113 first acquires inputs accepted by the input acceptance unit 112 (step ST1501).

[0134] The model modification unit 113 then determines whether the input indicates an erroneous alarm (step ST1502).

[0135] If the model modification unit 113 determines in this step ST1502 that the input indicates an erroneous alarm, the model modification unit 113 modifies a deterioration determination model so that the erroneous alarm does not occur (step ST1503). During the modification, the model modification unit 113 may, for example, adjust a dividing plane of the deterioration determination model so that normal feature value data is correctly determined as normal. The flow then proceeds to step ST1505.

[0136] On the other hand, if the model modification unit 113 determines in step ST1502 that the input does not indicate a faulty alarm, the model modification unit 113 modifies the deterioration determination model by correcting normal feature value data (training data) of the target device acquired during an alarm generation period to be deterioration feature value data (step ST1504). With the modification, the model modification unit 113 can modify the deterioration determination model, for example, by retrieving a normal model obtained immediately before the alarm generation period from the normal model storage unit 114 and using the normal model as the current normal model.

[0137] Subsequently, the model modification unit 113 outputs data representing the modified deterioration determination model (step ST1505).

[0138] As described above, the deterioration detection system 1 according to the second embodiment includes: a model evaluation unit 111 that issues an alarm when the accuracy of a deterioration determination is outside an allowable range based on the normal data of the target device and the deterioration determination model for the target device created by the deterioration determination model restructuring unit 107; an input acceptance unit 112 that, after an alarm is issued by the model evaluation unit 111, accepts inputs indicating a judgment as to whether the alarm is true or false;and a model modification unit 113 that modifies the deterioration determination model for the target device created by the deterioration determination model restructuring unit 107 based on the inputs accepted by the input acceptance unit 112. The deterioration determination unit 109 determines the deterioration of the target device based on operating data of the target device and the deterioration determination model for the target device modified by the model modification unit 113. With this configuration, the deterioration detection system 1 according to the second embodiment can detect deterioration of the target device even when normal data of the target device is abnormal.

[0139] Finally, an exemplary hardware configuration of the deterioration detection system 1 according to the first and second embodiments will be described with reference to Fig. 16. Although an exemplary hardware configuration of the deterioration detection system 1 according to the first embodiment is shown below, an exemplary hardware configuration of the deterioration detection system 1 according to the second embodiment is also identical.

[0140] As in Fig. 16, processes performed by the deterioration feature deriving unit 101, the normal model creating unit 102, the deterioration determination model creating unit 104, the normal model restructuring unit 106, the deterioration determination model restructuring unit 107, and the deterioration determination unit 109 are executed by a CPU (central processing unit, also referred to as a main processor, processing unit, arithmetic unit, microprocessor, microcomputer, processor, or digital signal processor (DSP)) 202, which reads programs stored in a memory 201.

[0141] Functions of the deterioration feature derivation unit 101, the normal model creation unit 102, the deterioration detection model creation unit 104, the normal model restructuring unit 106, the deterioration detection model restructuring unit 107, and the deterioration detection unit 109 are implemented with software, firmware, or a combination of software and firmware. Software and firmware are described as programs and stored in the memory 201. The CPU 202 implements the function of each unit by reading and executing the programs stored in the memory 201. The deterioration detection system 1 thus includes the memory 201 for storing programs with which, for example, each Fig.5 to 10 is thus performed when the programs are executed by the CPU 202. In addition, it can also be said that the programs cause a computer to execute operations and processes of the deterioration feature deriving unit 101, the normal model creating unit 102, the deterioration determination model creating unit 104, the normal model restructuring unit 106, the deterioration determination model restructuring unit 107, and the deterioration determination unit 109.

[0142] A parameter of a normal model stored in the normal model storage unit 103, a parameter of a deterioration determination model stored in the deterioration determination model storage unit 105, and a parameter of a deterioration determination model stored in the deterioration determination model storage unit 108 are each stored in a DB 204 via a DB input and output interface 203.

[0143] Data representing a result of a detection output from the detection result output unit 110 may be stored in the DB 204 or may be displayed on a display 206 via a display interface 205.

[0144] An input device 207 is used to input normal data and deterioration data collected from the other device, and normal data and acquisition target data collected from a target device. Various operating data input to the input device 207 are stored in the DB 204 or the memory 201 via an input interface 208.

[0145] The memory 201 and the DB 204 correspond, for example, to a non-volatile or volatile semiconductor memory such as a random access memory (RAM), a read-only memory (ROM), a flash memory, an erasable programmable ROM (EPROM) or an electrically erasable programmable ROM (EEPROM), a magnetic storage disk, a floppy disk, an optical storage disk, a compact disc, a MiniDisc or a digital versatile disc (DVD).

[0146] It should be noted that in the invention of this application, a free combination of the embodiments, modifications to any component of any of the embodiments, or omission of any component in any of the embodiments are possible within the scope of the invention. INDUSTRIAL APPLICABILITY

[0147] A deterioration detection system according to the present invention includes a normal model creating unit, a deterioration determination model creating unit, a normal model restructuring unit, a deterioration determination model restructuring unit, and a deterioration determination unit, and can detect deterioration of a target device without using operation data in a deterioration state of the target device, and is thus suitable as a deterioration detection system that detects deterioration of a device due to aging. LIST OF REFERENCE SYMBOLS

[0148] 1: Deterioration detection system, 101: Deterioration feature derivation unit, 102: Normal model creation unit, 103: Normal model storage unit, 104: Deterioration detection model creation unit, 105: Deterioration detection model storage unit, 106: Normal model restructuring unit, 107: Deterioration detection model restructuring unit, 108: Deterioration detection model storage unit, 109: Deterioration detection unit, 110: Detection result output unit, 111: Model evaluation unit, 112: Input acceptance unit, 113: Model modification unit, 114: Normal model storage unit, 201: Memory, 202: CPU, 203: DB input and output interface, 204: DB, 205: Display interface, 206: Display, 207: Input device, 208: Input interface, 1041: Unit for calculating a degree of deterioration,1042: Discriminator creation unit, 1061: Model unit, 1062: Parameter update unit, 1063: Convergence determination unit, 1071: Model unit, 1072: Deterioration degree calculation unit, 1073: Deterioration determination unit, 1074: Model update unit, 1111: Deterioration determination unit (second deterioration determination unit), 1112: Accuracy calculation unit, 1113: Comparison unit, 1114: Alarm output unit,

Claims

[1] Deterioration detection system comprising: a normal model creating unit (102) that creates a normal model for another device, which is a model for the other device different from a target device in a normal state, based on normal data of the other device, which is operation data of the other device in a normal state; a deterioration determination model creation unit (104) that creates a deterioration determination model for the other device based on the normal data of the other device and deterioration data of the other device, wherein the deterioration determination model is a model for determining the deterioration of the other device from operating data of the other device, and the deterioration data is operating data of the other device in a deterioration state; a normal model restructuring unit (106) that creates a normal model for the target device, which is a model for the target device in a normal state, based on the normal model for the other device created by the normal model creating unit and normal data of the target device, which is operation data of the target device in a normal state; a deterioration determination model restructuring unit (107) that creates a deterioration determination model for the target device based on the deterioration determination model for the other device created by the deterioration determination model creating unit and the normal model for the target device created by the normal model restructuring unit (106), wherein the deterioration determination model for the target device is a model for determining deterioration of the target device from operating data of the target device; and a deterioration determination unit (109) that determines the deterioration of the target device based on operation data of the target device and the deterioration determination model for the target device created by the deterioration determination model restructuring unit (107). [2] The deterioration detection system according to claim 1, wherein the normal model for the other device created by the normal model creating unit (102) is created using a pattern that can be exhibited in the normal data of the other device or a correlation between feature values derived from one or more pieces of normal data. [3] The deterioration detection system according to claim 1, wherein the deterioration determination model creating unit (104) creates the deterioration determination model for the other device based on the normal data of the other device, the deterioration data of the other device, and the normal model for the other device created by the normal model creating unit (102). [4] The deterioration detection system according to claim 1, wherein the deterioration detection model creation unit (104) comprises: a deterioration degree calculation unit (1041) that calculates a deterioration degree of the normal data of the other device and a deterioration degree of the deterioration data of the other device by applying the normal model for the other device created by the normal model creation unit (102) to both the normal data of the other device and the deterioration data of the other device, and a discriminator creation unit (1042) that creates a discriminator that can distinguish between the degree of deterioration of the normal data and the degree of deterioration of the deterioration data, the degree of deterioration being calculated by the deterioration degree calculation unit (1041), and uses the discriminator as a deterioration determination model. [5] The deterioration detection system according to claim 1, wherein the normal model restructuring unit (106) comprises: a model unit (1061) in which the normal model for the other device created by the normal model creating unit is included as the normal model for the target device; a parameter updating unit (1062) that updates a parameter of the normal model for the target device included in the model unit (1061) using the normal data of the target device; and a convergence determination unit (1063) that determines whether the parameter updated by the parameter update unit (1062) has converged, and wherein the parameter updating unit (1062) updates the parameter until the convergence determining unit (1063) determines that the parameter has converged. [6] The deterioration detection system according to claim 1, wherein the deterioration determination model restructuring unit (104) creates the deterioration determination model for the target device based on the deterioration determination model for the other device created by the deterioration determination model creating unit, the normal model for the target device created by the normal model restructuring unit (106), and the normal data of the target device. [7] The deterioration detection system according to claim 1, wherein the deterioration detection model restructuring unit (107) comprises: a model unit (1071) in which the deterioration determination model for the other device created by the deterioration determination model creation unit is included as a deterioration determination model for the target device; a deterioration degree calculating unit (1072) that calculates a degree of deterioration of the normal data of the target device by applying the normal model for the target device created by the normal model restructuring unit (106) to the normal data of the target device; a second deterioration determination unit (1073) that performs a deterioration determination by applying the deterioration determination model included in the model unit (1071) to the degree of deterioration calculated by the deterioration degree calculation unit (1072); and a model updating unit (1074) that updates the deterioration determination model included in the model unit (1071) based on a result of the deterioration determination made by the second deterioration determination unit (1073), and wherein the deterioration degree calculating unit (1072) calculates the deterioration degree until the normal model restructuring unit (106) completes the creation of the normal model for the target device. [8] The deterioration detection system of claim 1, further comprising: a model evaluation unit (111) that issues an alarm when the accuracy of a deterioration determination is outside an allowable range based on the normal data of the target device and the deterioration determination model for the target device created by the deterioration determination model restructuring unit (107); an input accepting unit (112) which, after the alarm is issued by the model evaluation unit (111), accepts inputs indicating a judgment as to whether the alarm is true or false, and a model modification unit (113) that modifies the deterioration determination model for the target device created by the deterioration determination model restructuring unit (107) based on the inputs accepted by the input acceptance unit (112), wherein the deterioration determination unit (109) determines deterioration of the target device based on operation data of the target device and the deterioration determination model for the target device modified by the model modification unit (113). [9] The deterioration detection system according to claim 1 or 8, wherein the target device is a platform screen door.

Images