Failure Probability Evaluation System

Abstract

To select time-series operating data associated with faults from time-series operating data measured from a large number of sensors and predict the remaining life of components constituting a machine system with higher accuracy.SOLUTION: The present invention is a fault probability evaluation system for evaluating the fault probability of fault in a component constituting a machine system. The fault probability evaluation system comprises: a maintenance history database that stores the maintenance history data of the machine system; an operation database that stores a plurality of operation data that indicates the operating state of the component; a scatter degree calculation unit that calculates a scatter degree, on the basis of the operation data and the maintenance history data, which indicates the degree of dispersion of a fault probability function for calculating fault probability for each of the operation data and which corresponds to the correlation with fault in the component; and an operation data selection unit that selects operation data in accordance with the calculated scatter degree. The fault probability evaluation system uses the selected operation data preferentially and realizes evaluation of the fault probability.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a technique for evaluating the failure probability of an object, and in particular to failure diagnosis and prediction (signs) including the calculation of the failure probability. The object here includes equipment, facilities, mechanical systems, and the components that make up these. [Background technology]

[0002] In various types of industrial machinery, such as power generation equipment and transportation equipment, which are types of mechanical systems, it is desirable for the mechanical systems to perform their specified functions normally. To achieve this, it is important to properly understand and manage the failure risk of each component that makes up the industrial machinery and to carry out maintenance such as repairs and replacements of each component at the appropriate time. Furthermore, when managing and operating multiple identical models of industrial machinery, statistical analysis of past maintenance records makes it possible to predict the number of corrective and preventive maintenances that may be required in the future, as well as the remaining lifespan of the target equipment. Here, maintenance records refer to data that records the details of maintenance and the time of occurrence in pairs.

[0003] In statistical analysis using maintenance records, the failure probability density function f(t), failure probability function F(t), failure rate function λ(t), etc., which are required to predict the number of corrective maintenance and preventive maintenance operations that may occur in the future, as well as the remaining lifespan of the target equipment, are estimated. The technology for this is shown in Non-Patent Document 1, etc. Here, t is the operating time of the equipment. When the failure probability density function f, failure probability function F, and failure rate function λ are functions of the variable t, their respective relationships can be expressed by (Equation 1) and (Equation 2).

[0004]

number

[0005]

number

[0006] Non-Patent Document 1 assumes that the operating status of a mechanical system is static, i.e., does not fluctuate in time or space. However, the operating status of a mechanical system is generally not constant in time or space. For example, the operating status of a wind turbine changes from moment to moment depending on wind conditions, and the load also varies depending on its location. Furthermore, the load on construction machinery, for example, changes depending on the work environment, such as the different tasks performed daily, the operating characteristics of the operator, and soil characteristics. Therefore, evaluation of failure probability functions or failure rate functions based simply on operating time has limitations in the accuracy of estimating the number of maintenance visits and remaining lifespan.

[0007] In recent years, various sensors have been attached to mechanical systems, making it easy to access the time-series operation data measured by these sensors via a network. In order to improve the above-mentioned estimation accuracy, Patent Document 1 evaluates the operating conditions that differ for each individual machine from the time-series operation data measured by the sensors and takes this into appropriate consideration, thereby enabling more accurate estimation of the number of maintenance visits and remaining lifespan. [Prior art documents] [Non-patent literature]

[0008] [Non-Patent Document 1] Yasuyoshi Fukui, "Introduction to Reliability Engineering", Morikita Publishing Co., Ltd., 2006 [Patent documents]

[0009] [Patent Document 1] Japanese Patent Application Publication No. 2019-160128 Summary of the Invention [Problem to be solved by the invention]

[0010] In recent mechanical systems, it is not uncommon for a large number of sensors, ranging from hundreds to thousands, to be attached to equipment. However, not all time-series operation data is necessarily related to failures, and a large amount of time-series operation data unrelated to failures is also included. Therefore, in order to further improve the estimation accuracy of the number of maintenance visits and remaining life, it is desirable to pre-select only time-series operation data related to failures and perform failure probability evaluation. However, Patent Document 1 does not disclose a technology for pre-selecting only time-series operation data related to failures. Therefore, an object of the present invention is to further improve the estimation accuracy of maintenance-related information for an object, such as the number of maintenance visits and remaining life. [Means for solving the problem]

[0011] In order to solve the above problems, the present invention The more specific configuration of A failure probability evaluation system for evaluating the failure probability of a component constituting a mechanical system includes a maintenance history database for storing maintenance history data of the mechanical system, and a plurality of databases showing the operating status of the components. Time series Stores operation data Time series a working database; Time series Failure probability function for calculating the failure probability for each operational data a time-series operation data selection unit that selects time-series operation data with small variations in time series operation data; a failure probability evaluation unit that inputs the selected time-series operation data and the maintenance history data, identifies a correspondence relationship of learning data and a correspondence relationship of test data, which are correspondence relationships of failure probabilities with the input time-series operation data, and calculates a prediction performance index of a lifespan model from the correspondence relationship of the learning data and the correspondence relationship of the test data; The calculated Predictive Performance Indicators Depending on Time series Selecting Operational Data Time series An operation data selection unit is provided, Time series This is a failure probability evaluation system that prioritizes the use of operational data to evaluate the failure probability.

[0012] The present invention also includes a failure probability evaluation method using the failure probability evaluation system, a failure probability evaluation program that causes the failure probability evaluation system to function as a computer, and a storage medium that stores the program. [Effects of the Invention]

[0013] According to the present invention, it is possible to improve the accuracy of estimation regarding the maintenance of an object compared to the prior art. Problems, configurations, and effects other than those described above will become apparent from the following description of the embodiments. [Brief explanation of the drawings]

[0014] [Figure 1] System configuration diagram for Example 1 [Figure 2] FIG. 10 is a diagram showing an example of maintenance history data used in the first embodiment. [Figure 3] FIG. 10 is a diagram showing an example of time-series operation data and a division flag table stored in the time-series operation database 2 used in the first embodiment. [Figure 4] FIG. 1 shows an example of survival analysis data used in Example 1. [Figure 5] FIG. 10 is a diagram showing a failure probability function and its coefficient of variation for the cumulative value of time-series operation data in Example 1. [Figure 6] Flowchart for determining the number of time-series operational data Ns in the first embodiment [Figure 7] Graph showing the relationship between the deviation rate R and the number of selected time-series operational data Ns in Example 1 [Figure 8] Graph showing the relationship between failure probability and cumulative damage in Example 1 [Figure 9] System configuration diagram including a modification of the failure probability evaluation unit in the first embodiment [Figure 10] FIG. 1 is a diagram showing an example of a graphical user interface (GUI) on a display unit in the first embodiment. [Figure 11] System configuration diagram for Example 2 [Figure 12] System configuration diagram for Example 3 [Figure 13] Hardware configuration diagram of the failure probability evaluation system in Example 4 DETAILED DESCRIPTION OF THE INVENTION

[0015] A failure probability evaluation system 100 according to an embodiment of the present invention will be described below. The failure probability evaluation system 100 according to this embodiment is for evaluating the failure probability of failures in components constituting a mechanical system, and includes a maintenance history database that stores maintenance history data of the mechanical system, an operation database that stores a plurality of operation data indicating the operating statuses of the components, a dispersion calculation unit that indicates the degree of dispersion of a failure probability function for calculating the failure probability for each of the operation data based on the operation data and the maintenance history data and calculates a dispersion according to the correlation with failures in the components, and an operation data selection unit that selects operation data according to the calculated dispersion, and prioritizes the use of the selected operation data to evaluate the failure probability.

[0016] According to this embodiment, by selecting operation data of sensors related to failures from a large amount of operation data and using it to evaluate the failure probability, it is possible to predict the failure probability of components of a mechanical system with high accuracy.

[0017] Below, examples showing more specific examples of this embodiment will be described with reference to the drawings. The failure probability assessment system 100 is realized on a calculator (computer), and its functions are executed by a processing device in accordance with a program. However, its functions may also be realized by dedicated hardware, and the present invention is not limited to the use of a program (software).

[0018] In the following examples, a wind power generator 1 is used as the mechanical system to evaluate the failure probability of the wind power generator 1 and its components. However, the application of the present invention is not limited to the wind power generator 1.

[0019] Furthermore, in this embodiment, the above-mentioned (Equation 1) and (Equation 2) are also used, and as is clear from these, if one of the failure probability density function f, the failure probability function F, and the failure rate function λ can be identified, the other functions can be calculated. Therefore, in the following examples, when identifying the failure probability distribution, the failure probability density function f and the failure probability function F are used selectively as necessary. [Example]

[0020] 1 is a system configuration diagram according to the first embodiment. A failure probability evaluation system 100 according to the present embodiment includes a maintenance history database 3, a time-series operation database 2, a time-series operation data selection unit 13, and a failure probability evaluation unit 10. The time-series operation database 2 and the maintenance history database 3 may be provided outside the failure probability evaluation system 100, or may be configured as a single database.

[0021] First, the maintenance history database 3 and the maintenance history data 6 stored therein will be described. The maintenance history database 3 shown in FIG. 1 stores maintenance history data 6 for components of the wind power generator 1. FIG. 2 is a diagram showing an example of the maintenance history data 6 used in this embodiment. As shown in FIG. 2, the maintenance history data 6 is data in which the individual on which maintenance was performed is associated with the date and time of the maintenance. In this embodiment, the site name and unit number to which the wind power generator 1, which is the target, belongs are used as information for identifying the individual. Note that an individual number or the like may also be used as information for identifying the individual. This also applies to each data item described later. Furthermore, it is desirable that the maintenance history data 6 include a maintenance history including events (maintenance events) related to the maintenance performed on the individual. As shown in FIG. 2, in this embodiment, the items of maintenance content, part name, and content are used as the maintenance history.

[0022] Here, the maintenance content indicates the type of maintenance that has been performed, and in this embodiment, "preventive maintenance" and "post-maintenance" are used. "Preventive maintenance" refers to maintenance such as part replacement carried out before a failure occurs, such as through periodic replacement. "Post-maintenance" refers to maintenance such as part replacement carried out after an abnormality or failure has occurred. Furthermore, the part name identifies the part or location that has been maintained or where the abnormality occurred. Furthermore, the content indicates the content of the maintenance that has been performed.

[0023] The maintenance history data 6 may also include the following items. For example, in the case of preventive maintenance, the reason for the maintenance can be used. In the case of corrective maintenance, the occurrence of the failure can be used.

[0024] Furthermore, if the wind power generator 1 has a function for automatically detecting the maintenance history, the wind power generator 1 and the maintenance history database 3 may be connected via a network, and the maintenance history data 6 may be automatically stored in the maintenance history database 3. Alternatively, the worker 4 in charge of maintenance may register the maintenance history in the maintenance history database 3. With this configuration, the maintenance history including the maintenance details for multiple component parts is stored in the maintenance history database 3.

[0025] Next, we will explain the time-series operation database 2 and the time-series operation data 5 stored therein. In this embodiment, time-series operation data 5, such as operation data on the components of the wind power generator 1, is accumulated in the time-series operation database 2 via communication means such as a network. At this time, the collection interval for each data is not particularly limited, but can be set to the maximum interval within a range that does not lose information on the physical phenomenon of interest. In this case, it is possible to collect data that is meaningful for failure probability evaluation while reducing the data volume. For example, in the failure probability evaluation of the wind power generator 1 in this embodiment, a daily interval is ideal because relatively long-term predictions, such as several months or several years, are handled. Note that the time-series operation data 5 is a type of operation data.

[0026] Furthermore, the time-series operation data 5 may be measurement values ​​sampled at any interval, but it is more preferable to use statistical values ​​such as maximum, minimum, average, and standard deviation within the collection interval. This makes it possible to maximize the use of information while significantly reducing the amount of data. For example, average wind speed per day may be considered. Furthermore, the information stored in the time-series operation database 2 is not limited to information obtained from the components themselves. For example, weather data such as temperature measured by meteorological observation equipment installed near the components is also useful for evaluating the load state of the components.

[0027] FIG. 3 illustrates an example of the time-series operation data 5 and the division flag table 41 stored in the time-series operation database 2 used in this embodiment. First, FIG. 3(a) illustrates an example of the time-series operation data 5 output to the time-series operation data selector 13. As shown in FIG. 3(a), the time-series operation data 5 includes the following fields: date and time, site name, unit number, and sensors 1 to N. The date and time indicate the date and time of measurement by sensors 1 to N. The site name and unit number are the same as those in the maintenance history data 6 and indicate the measurement targets of sensors 1 to N. Sensors 1 to N each indicate the sensors that measure the time-series operation data 5. In the wind power generator 1, sensor 1 measures the average wind speed as the time-series operation data 5. In addition, other sensors 2 to N measure time-series operation data other than the average wind speed. That is, in this embodiment, a total of N sensors 1 to N measure N pieces of time-series operation data. In the example of FIG. 3(a), the data collection interval (measurement interval) is one day. For convenience, information other than that obtained directly from components, such as temperature measured by weather observation equipment, is also referred to as a sensor here.

[0028] Before describing the division flag table 41 shown in FIG. 3(b) in detail, its necessity will be explained. As with general machine learning, a failure probability evaluation device also needs to have high prediction accuracy not only for data used in learning but also for unknown data. Furthermore, in order to evaluate prediction performance for unknown data, it is necessary to divide the training data used in learning by the failure probability evaluation unit 10 (described later) and test data used as a substitute for the unknown data, and compare the failure probabilities for each. Furthermore, in the division flag table 41, a division flag is associated with each object (wind turbine generator 1). Specifically, in FIG. 3(b), the site name and unit number, which are examples of information for identifying an individual object, are used.

[0029] Therefore, in this embodiment, a division flag (learning or test) is assigned to the time-series operation data 5 according to the site name and the machine number based on the division flag table 41. As a result, the time-series operation data 5 is divided into learning data and test data. Specifically, a division flag according to the individual number in FIG. 3(b) is applied to the time-series operation data 5 in FIG. 3(a). In this way, the time-series operation data 5 is divided into learning data and test data. By similarly applying FIG. 3(b) to the maintenance history data 6, the maintenance history data 6 can also be divided into learning data and test data. The division method is not particularly limited, but it is preferable to divide the data so that there is no bias in the number of failure occurrences between learning data and test data.

[0030] Next, the time-series operation data selection unit 13 will be described. The time-series operation data selection unit 13 has a feature amount calculation unit 14, a failure correlation calculation unit 15, and a selected time-series operation data number determination unit 16. The time-series operation data selection unit 13 also selects time-series operation data of a sensor related to a failure from the time-series operation data 5. Then, the time-series operation data selection unit 13 adds a division flag to the selected time-series operation data using a division flag table 41, and outputs this and the maintenance history data 6.

[0031] For example, when there are 100 pieces of time-series operation data 5 for each sensor, the time-series operation data selection unit 13 selects 10 pieces of time-series operation data for sensors related to the failure. Then, the time-series operation data selection unit 13 outputs the maintenance history data and dataset 9 and dataset 91 of the selected time-series operation data. Here, dataset 9 includes both learning data and test data, and dataset 91 includes only learning data. The feature calculation unit 14, the failure correlation calculation unit 15, and the selected time-series operation data number determination unit 16, as well as a specific method for selecting time-series operation data for sensors related to the failure, will be described later.

[0032] Next, the failure probability evaluation unit 10 will be described. The failure probability evaluation unit 10 has a life modeling unit 17, a life model recording unit 18, and a failure probability output unit 19. The life modeling unit 17 receives a data set 91 (learning data) as input, generates a life model 20, and records it in the life model recording unit 18. The failure probability output unit 19 receives a data set 9 (learning data, test data) as input, calls a life model 21 recorded in the life model recording unit 18, and outputs a failure probability 11.

[0033] Here, the lifespan model 20 refers to a machine learning model or a statistical model that calculates a failure probability from a selected set of time-series operation data. These models take time-series operation data (operation information) as input and output a corresponding failure probability. Each model will be specifically described below.

[0034] The machine learning model is a model that directly calculates the failure probability up to the time of input of time-series operation data (operation information) using a machine learning method such as a neural network (deep learning, etc.) or random forest. Note that the machine learning method is not limited to the above method. The parameters of the trained machine learning model (for example, the weight coefficients of neurons in each layer of a neural network) are recorded in the life model recording unit 18.

[0035] A statistical model is a model that generates explanatory variables from a set of time-series operation data, learns the correspondence between failure probability and explanatory variables using probability and statistical methods based on maintenance history data, and calculates failure probability from the time-series operation data based on the correspondence. When using a statistical model to parametrically express the correspondence as a probability distribution, the parameters of the probability distribution are determined using methods such as maximum likelihood estimation or Bayesian estimation. When a probability distribution is not assumed using a statistical model, non-parametric methods such as the Kaplan-Meier method can be used. Note that the probability and statistical methods are not limited to those mentioned above.

[0036] Furthermore, when learning a statistical model in the life modeling unit 17, it is desirable to convert the time-series operation data according to the failure mechanism and use it as an explanatory variable for the statistical model. For example, when thermal load is the dominant factor in the deterioration of an object, such as material deterioration, the explanatory variable can be specified as follows. The Arrhenius equation shown in (Equation 3) converts the time-series operation data T of temperature for learning data into the amount of deterioration D occurring at time i, which is proportional to the speed of the deterioration reaction. i and the cumulative value D=ΣDi up to the occurrence of maintenance can be used as an explanatory variable.

[0037]

number

[0038] where A is a constant corresponding to the degradation reaction rate, R a is the gas constant, E a is the activation energy. When expressing the correspondence between the explanatory variable D and the failure probability as a parametric probability distribution F(D), for example, the Weibull distribution F shown in (Equation 4) can be assumed.

[0039]

number

[0040] Here, the parameters α and β are the shape parameter and scale parameter of the Weibull distribution, respectively. Here, the unknown parameters A and E of the Arrhenius equation are calculated from the cumulative value D (explanatory variable) for all individuals and the maintenance history data. a In order to identify the parameters α and β of the failure probability function F (Weibull distribution), the techniques described in Non-Patent Document 1 and Patent Document 1 can be used. In addition, the life model recording unit 18 stores the parameters A and E of the explanatory variables as a statistical model. a , and the type of probability function used for identification (Weibull distribution in the above example). Note that functions identified as failure probability functions include, but are not limited to, the Weibull distribution, gamma distribution, log-normal distribution, etc.

[0041] Furthermore, it is known that in wind turbines 1, the square of the average wind speed is proportional to the wind load acting on the blades, etc. Therefore, the cumulative value of the square of the average wind speed may be used as the explanatory variable. Furthermore, when calculating the explanatory variables, a statistical model showing the correspondence with the failure probability (objective variable) may be constructed using not only the converted values ​​of the physical quantities described above, but also cumulative values ​​calculated using a counting method such as the rainflow method as explanatory variables. The rainflow method is a method for counting the stress frequency of components subjected to repeated fluctuating loads, and is effective when the failure mechanism is fatigue fracture.

[0042] Although several simple explanatory variables have been exemplified above, it is also possible to generate explanatory variables from multiple time-series operation data. Even if the failure mechanism and physical laws for the time-series operation data 5 of each sensor are unknown, the explanatory variable D(x) can be automatically learned from a combination of time-series operation data x of the target equipment and used in a statistical model. This method will be described later in the description of the desirable failure probability evaluation unit 10 of this embodiment (FIG. 9).

[0043] Although specific examples of machine learning models and statistical models have been shown above as lifespan models, the learning method for lifespan models is not limited to these.

[0044] Next, the failure probability output unit 19 receives as input the maintenance history data including the learning data and the test data, the selected data set 9, and the lifespan model 21 stored in the lifespan model recording unit 18. Then, the failure probability output unit 19 outputs the correspondence relationships F1 and F2 between the time-series operation data (input) of the learning data and the test data and the failure probability (output).

[0045] Here, when a machine learning model is used as the lifespan model, the failure probability can be output directly for the selected data set (input). Therefore, the above correspondence is expressed as a set of pairs of input and output values ​​obtained for each individual.

[0046] Furthermore, when a statistical model is used as a lifespan model, the correspondence between the maintenance history data and the set of selected data (input) and the failure probability (output) is expressed by a failure probability function in a parametric model, or by a Kaplan-Meier curve, for example, in a non-parametric model.

[0047] Next, the feature amount calculation unit 14, the failure correlation calculation unit 15, and the selected time-series operation data number determination unit 16 will be specifically described.

[0048] First, the feature calculation unit 14 will be described. The feature calculation unit 14 receives the time-series operation data 5 and maintenance history data 6 of each sensor as input. The feature calculation unit 14 then calculates feature quantities for these data and outputs survival analysis data 7. While there are no particular limitations on the method for calculating feature quantities, it is desirable to use the cumulative value of the time-series operation data of each sensor. Damage to an object or its components can accumulate during use, resulting in a failure. In the case of such a failure, such as a wear-out failure, the cumulative value of the time-series operation data is considered to be correlated with the failure. While there are no limitations on the method for calculating the cumulative value, it is desirable for the cumulative value to be a monotonically increasing value because damage accumulation is an irreversible process. Therefore, if the time-series operation data 5 is negative, it is desirable to convert it using a function that does not take negative values, such as an absolute value function or a soft-plus function, before calculating the cumulative value. It is desirable for the feature calculation unit 14 to identify a division flag for the input data using a division flag table 41 and include this in the survival analysis data 7.

[0049] As a result of the above, the cumulative value can be made to increase monotonically. Furthermore, depending on the failure mechanism, as explained in the life modeling section, further improvement in accuracy can be achieved by applying transformations and counting methods to the time-series operation data itself and then calculating the cumulative value. For example, physical knowledge regarding the failure mechanism can be incorporated by applying the Arrhenius equation to temperature data, the squared value to average wind speed data, and the rainflow method to stress data. Note that the calculation of the cumulative value may use a combination of the above transformations and counting methods, or others.

[0050] FIG. 4 shows an example of the survival analysis data 7 used in this embodiment. As shown in FIG. 4, the survival analysis data 7 records cumulative values ​​obtained by accumulating moment-to-moment values ​​of the time-series operation data 5 in FIG. 3, that is, cumulative values ​​of the time-series operation data of each sensor at the time of measurement. Furthermore, a survival analysis flag (survival or failure) is added to the survival analysis data 7 according to the maintenance event based on the maintenance history data 6. As the survival analysis flag, a survival flag is added in the case of proactive maintenance, and a failure flag is added in the case of reactive maintenance.

[0051] Furthermore, for units that have never undergone maintenance since their initial operation and are still in operation (such as Unit 1 at Site XX in Figure 4), there is no record of them in the maintenance history database. In this case, "No maintenance" is recorded in the maintenance event column in Figure 4. In the case of "No maintenance," a survival flag is added as a survival analysis flag, just like in the case of pre-maintenance. This results in survival analysis data 7, in which the cumulative value of the time-series operation data of each sensor at the time of measurement for each unit is paired with the survival analysis flag (survival or failure). The feature calculation unit 14 applies the division flag table 41 to the survival analysis data 7, linking the units (site name and their unit number) to add a division column, and can divide the data into training data and test data.

[0052] Next, the failure correlation calculation unit 15 will be described. Note that the operation in the failure correlation calculation unit 15 is performed only on the learning data. This is because time-series operation data is selected based on the output value of the failure correlation calculation unit 15 (the correlation between the feature amount of the time-series operation data of each sensor and the failure) and a lifespan model is learned. In other words, the test data used for verifying prediction accuracy should not be used in the failure correlation calculation unit 15. Note that the failure correlation calculation unit 15 can function as a dispersion calculation unit that indicates the degree of dispersion of the failure probability function and calculates the dispersion according to the correlation with component failures.

[0053] The failure correlation calculation unit 15 receives as input the learning data in the survival analysis data 7 whose division flag is "learning," and calculates the correlation between the time-series operation data of each sensor and failures, and the correlation between operation time and failures, using the feature values ​​calculated by the feature value calculation unit 14. Specifically, the failure correlation calculation unit 15 uses the cumulative value of the time-series operation data of each sensor as the feature value. The failure correlation calculation unit 15 then outputs a list 8 that estimates the variance of the failure probability function for the cumulative value and the variance of the failure probability function based on the total operation time. As explained in the description of the lifespan modeling unit 17, in the data set of the survival analysis data 7, for example, only the cumulative value of the time-series operation data (average wind speed) of sensor 1 is focused on. Then, a known technique can be used to estimate a failure probability function that fits the focused variable as an explanatory variable. Therefore, the failure correlation calculation unit 15 can use similar processing to evaluate the failure probability function that matches the cumulative value and total operating time of the time-series operation data of each sensor in the survival analysis data 7, and obtain the number of failure probability functions (N) + 1. Once the failure probability function is determined, its variation can be easily quantified using, for example, the coefficient of variation. For example, the mean E(x) and variance V(x) of a random variable x that follows the failure probability function of the two-variable Weibull distribution shown in (Equation 4) can be calculated using the following (Equation 5) and (Equation 6).

[0054]

number

[0055]

number

[0056] From the above, the coefficient of variation C v is a function of only the shape parameter α as shown in the following (Equation 7).

[0057]

number

[0058] FIG. 5 illustrates the failure probability function and its coefficient of variation for the cumulative time-series operation data in this embodiment. Specifically, FIG. 5(a) illustrates a failure probability function 22 that fits the cumulative time-series operation data (cumulative average wind speed) of sensor 1, which is related to the failure. FIG. 5(b) illustrates a failure probability function 23 that fits the cumulative time-series operation data of sensor N, which is not related to the failure. FIG. 5(c) illustrates a failure probability function 24 based on the operating time. As shown in FIG. 5(a), in the case of wind power generator 1, the failure probability of components (e.g., gearboxes and blades) increases sharply when the cumulative value of sensor 1's time-series operation data (average wind speed) exceeds a certain value. This is because average wind speed is a physical quantity that correlates with the wind load on the wind power generator. Note that FIG. 5(a) illustrates that the variance of the failure probability function decreases when average wind speed is used as the reference.

[0059] As shown in Figure 5(b), there is no correlation (smaller) between the magnitude of the cumulative value and the failure probability. This is because the time-series operation data of sensor N is not related to failures or has only a small correlation. Therefore, even if the cumulative value of the time-series operation data of sensor N increases, the failure probability function does not necessarily increase suddenly, and the variance of the failure probability function increases.

[0060] As shown in Figure 5(c), the operating time-based failure probability function 24 is identified for the cumulative value of operating time (corresponding to the number of operating days in Figure 4) without using time-series operating data. To improve the prediction accuracy by utilizing time-series operating information compared to time-axis-based failure probability evaluation, the following condition must be satisfied. In other words, the variance of the failure probability function for the cumulative value of the time-series operating data of the sensor related to the failure must be at least smaller than the variance of the failure probability function for the total operating time. Figure 5(d) shows List 8. This quantifies the variance of the failure probability function for the cumulative value of the time-series operating data of each sensor and for the total operating time as a coefficient of variation, which is output from the failure correlation calculation unit 15 to the selected time-series operating data number determination unit 16. Figures 5(a) to 5(d) can be displayed on the display unit 12.

[0061] Next, the selected time-series operation data number determination unit 16 will be described. The selected time-series operation data number determination unit 16 receives as input the list 8, the maintenance history data 6, a correspondence relationship F1 between the time-series operation data (input) of the learning data and the failure probability (output), and a correspondence relationship F2 between the time-series operation data (input) of the test data and the failure probability (output). Then, the selected time-series operation data number determination unit 16 determines the maintenance history data 6 and N s The selected time-series operational data number determination unit 16 outputs the data set 9 of the time-series operational data. The selected time-series operational data number determination unit 16 can function as an operational data selection unit that selects operational data.

[0062] Next, the number of selected time-series operational data N using each of the above configurations s The procedure for obtaining the above will be described with reference to the flowchart shown in Fig. 6. First, in step S50, the time-series operation data selection unit 13 selects N top time-series operation data with small variations in the failure probability function. s This means that from the columns of time-series operation data of each sensor shown in Figure 3, all but the columns of selected time-series operation data of the top sensors with the smallest coefficient of variation are deleted, and only the selected time-series operation data of the top sensors with the smallest variation are selected. s The maximum value of N max An example of this is the total number N of time-series operational data.

[0063] Here, in order to reduce the calculation time, N s It is also effective to set the maximum value of N as the total number of time-series operation data corresponding to the feature quantity whose failure probability function has smaller variance than the operating time. The minimum value is 1. s The range is initially 1 to N max In this case, the range may be roughly searched for in which the predicted performance index (described later) improves by varying the value in logarithmic intervals up to 1, and then the range may be searched for in detail in increments of 1.

[0064] In step S51, the failure probability evaluation unit 10 inputs the selected time-series operation data and maintenance history data 6, and acquires a correspondence relationship between the time-series operation data (input) and the failure probability (output). As a result, the failure probability evaluation unit 10 identifies a correspondence relationship F1 between the learning data and the test data, and a correspondence relationship F2 between the test data and the training data.

[0065] In step S52, the failure probability evaluation unit 10 calculates a predictive performance index of the lifespan model from the correspondence relationship F1 of the learning data and the correspondence relationship F2 of the test data. When a machine learning model is used as the lifespan model, the above correspondence relationship is a set of pairs of time-series operation data (input) and failure probability (output). In this case, the dissimilarity between the sets is calculated and used as the evaluation index. Possible dissimilarity measures include Minkowski distance (e.g., Euclidean distance or Manhattan distance), Mahalanobis distance, and cosine similarity multiplied by a negative value.

[0066] Furthermore, when a statistical model is used as the lifespan model, the above correspondence can be obtained as a failure probability function or Kaplan-Meier curve that defines the input / output relationship between time-series operating data (input) and failure probability (output). The deviation rate R between the failure probability function and Kaplan-Meier curve can be used as an evaluation index. The deviation rate R can be calculated using equation 8. Note that this deviation rate R is an example of the degree of dispersion that indicates correlation.

[0067]

number

[0068] The deviation rate R can be calculated by calculating the difference in failure probability at each point even if the correspondence relationship is not obtained as a continuous function. In addition, when the failure probability function is obtained as a continuous probability distribution function, the Kullback-Leibler divergence D, which quantifies the difference between the two distribution functions, can be calculated. KL(F1||F2) can be used. In (Equation 9), we show an example of the formula for calculating the Kullback-Leibler divergence in the case of a continuous probability distribution.

[0069]

number

[0070] Furthermore, the performance index of the lifespan model is not limited to the above indexes. For example, the difference in the variability of the correspondence relationship between the time-series operation data (input) and the failure probability (output) (for example, the difference between the coefficient of variation of the failure probability function F1 of the training data and the coefficient of variation of the failure probability function F2 of the test data) can also be used.

[0071] In addition, in steps S50 to S52, the number of selected time-series operation data N s In step S53, the time-series operation data selection unit 13 selects the number N of selected time-series operation data that provides the best evaluation index. s Determine.

[0072] The effects of this embodiment will be explained below using virtual data as an example. A machine system that acquires time-series operation data from a large number of sensors is simulated, and the number of time-series operation data measured by the virtual sensors is set to 100. Of these, the time-series operation data from 10 sensors is from damage sensors related to failures, and the remaining 90 is from dummy sensors that are not related to failures. Failures in the virtual data are generated according to the Weibull distribution of cumulative damage, which is expressed as the linear sum of the 10 time-series operation data from the damage sensors.

[0073] The life model used is a damage model D(x) obtained by a damage model generation / update unit 303 in the failure probability evaluation unit 10, which is desirable for this embodiment and will be described later with reference to Fig. 9. In addition, the failure probability evaluation unit identifies a Weibull distribution as the correspondence relationship between the failure probability (output) and the time-series operating data (input) for learning and testing.

[0074] Here, FIG. 7 shows the relationship between the deviation rate R and the number of selected time-series operational data N in this embodiment. s In Fig. 7, the Y axis represents the deviation rate R, and the X axis represents the number of selected time-series operational data N s The dashed line in Figure 7 shows the deviation rate based on the total operating time, and the thick solid line shows the line with a deviation rate of 0. The closer the deviation rate is to 0, the better the forecasting performance. In Figure 7, the deviation rate is smallest when the number of selected time-series operating data is 10. Therefore, when the number of selected time-series operating data is N a This is the case when no time-series operation data is selected (N s =100), when time-series operation data is selected (N w = 10) shows that the deviation rate R is reduced and the prediction performance is improved.

[0075]

number

[0076] 8A and 8B are graphs showing the relationship between the failure probability and the cumulative damage in this embodiment. In Fig. 8A, the time-series operation data selection unit 13 selects the top time-series operation data N with the smallest variation in the failure probability function. s The failure probability evaluation result 27 when N = 10 is selected is shown. Also, in Fig. 8(b), the failure probability evaluation result 27 when N = 10 is selected is shown. sThe failure probability evaluation results for graphs 28 with 100 failures are shown. Figures 8(a) and 8(b) compare the failure probability function F1 (solid line A) during training and the failure probability function F2 (dashed line B) during testing. In Figure 5(a), the deviation between solid line A and dashed line B is small, indicating highly accurate predictions. In Figure 5(b), the deviation between solid line A and dashed line B is large, indicating a decline in prediction performance. This indicates that selecting time-series operational data related to failures prevents overfitting to the training data and improves prediction accuracy (generalization performance) for unknown data. Figures 8(a) and 8(b) can be displayed on the display unit 12. As described above, in this embodiment, the selected time-series operation data number determination unit 16 selects the operation history data with the smallest deviation rate (degree of deviation) between the correspondence relationship between the operation history data at the time of learning (for learning) and the correspondence relationship between the operation history data at the time of testing (for testing). As described above, this correspondence relationship is expressed by a set of pairs of a predetermined number of top time-series operation data items having a high correlation with failures among the time-series operation data items and failure probabilities, or by a parametric or non-parametric failure probability function.

[0077] Next, a modified example of the failure probability evaluation unit 10 in this embodiment, different from that shown in FIG. 1, will be described. FIG. 9 is a system configuration diagram including a modified example of the failure probability evaluation unit 10 in this embodiment. In this modified example, the time-series operation data of the sensor related to wear-out failures is selected in the time-series operation data selection unit 13. Therefore, in the failure probability evaluation unit 10, an explanatory variable D(x t ) and construct a statistical model for the explanatory variables, we can expect to achieve even greater results.

[0078] The details of the failure probability evaluation unit 10 of this modified example are shown below. The failure probability evaluation unit 10 has a life modeling unit 17, a life model recording unit 18, and a failure probability output unit 19. The life modeling unit 17 is made up of an explanatory variable generation / update unit 30 and a failure probability function identification unit 33.

[0079] First, the life modeling unit 17 in this modification will be described. Here, the life modeling unit 17 has an explanatory variable generating / updating unit 30 and a failure probability function identifying unit 33, and outputs an explanatory variable D(x). Note that this function can be realized by the technology described in Patent Document 1. In addition, the explanatory variable D(X t ) is the time series operation data X t It is shown as a function of (Equation 10).

[0080] Here, d(x) is the damage to the equipment per unit time, and x is an operation data vector that represents the time-series operation data set at a certain moment. In this modification, since wear-out failures are the target, the time integral D(x) of the damage per unit time d(x) is taken as the explanatory variable that leads to equipment failure. In this modification, the form of the formula for calculating the explanatory variable D(x) is not particularly limited. For example, as shown in (Equation 11), the explanatory variable D(x) is calculated by multiplying the selected N s The mathematical formula expressed as a linear combination of individual time-series operational data is the simplest, and the optimization calculation requires relatively low computational costs.

[0081]

number

[0082] where C=(c1,c2,...cNs) is the coefficient vector that represents the weighting of each time-series operational data, and x=(x 1, x 2, …,xNs) is the number of N s These are the instantaneous values ​​of time-series operational data.

[0083] Next, the failure probability function identification unit 33 will be described. The processing of the failure probability function identification unit 33, that is, the method of identifying the failure probability function, is publicly known, as is the failure probability output unit 19 and the failure correlation calculation unit 15. When generating the explanatory variable D(x), the explanatory variable generation / update unit 30 repeatedly calls the failure probability function identification unit 33 to minimize the variation in the failure probability function.

[0084] Next, the explanatory variable generation and update unit 30 will be described. The explanatory variable generation and update unit 30 receives the time-series operation data and maintenance history data 6 used for failure probability evaluation as input, and outputs the explanatory variable D(x) and standard survival analysis data 32. Here, the explanatory variable-standard survival analysis data 32 is the survival analysis data 7 with the column of the explanatory variable D(x) added. The survival analysis data 32 can obtain a failure probability function by identifying the failure probability function described above. Therefore, the variability of the failure probability function, such as the coefficient of variation, can be easily evaluated.

[0085] Furthermore, the explanatory variable generation and update unit 30 automatically searches for explanatory variables that take into account time-series operation data so as to minimize the variance 31 of the failure probability function. The explanatory variables obtained as a result are reflected in the survival analysis data 7, and damage model-based survival analysis data 32 is generated. The search for explanatory variables can be reduced to an optimization problem in which the variance is the objective function and the explanatory variables D(x) are the parameters.

[0086] Also, if the failure mechanism is known to some extent, a method may be adopted in which the user defines in advance only the shape of the equation in accordance with the failure mechanism and searches for the coefficients.The conversion method and counting method described in the description of the life modeling unit 17 and the feature calculation unit 14, or a combination thereof, can be used.

[0087] The optimization calculation method for obtaining the coefficient vector representing the weighting of each time-series operation data is not particularly limited. However, since the objective function may be non-convex, it is desirable to use metaheuristics such as a genetic algorithm or particle swarm optimization.

[0088] The explanatory variable D(x) learned as described above is stored in the life model recording unit 18, and then is called up as appropriate by the failure probability output unit 19 and used to calculate the failure probability.

[0089] Next, the display unit 12 will be described. The display unit 12 receives the failure probability up to the present time from the failure probability output unit 19, the variance of the failure probability function of the cumulative value of the time-series operation data of each sensor from the failure correlation calculation unit 15 in the time-series operation data selection unit 13, and a list 8 that estimates the variance of the failure probability function based on the operating time. The display unit 12 then displays these. Specifically, the display unit 12 is composed of a computer that implements a screen drawing program and a display device, but the computer used here may be different from the computer that performs the functions of the above-mentioned calculation unit.

[0090] Next, the display on the display unit 12 will be described. FIG. 10 is a diagram illustrating an example of a graphical user interface (GUI 40) on the display unit 12 in this embodiment. The illustrated GUI 40 displays the correlation with the failure, the sensor number, the sensor name, and the coefficient of variation for time-series operation data that is highly correlated with a failure. Here, the correlation with the failure indicates the degree of correlation with the equipment failure, with higher correlations indicating higher correlations. The sensor number and the sensor name indicate the sensor that measured the failure. Furthermore, the coefficient of variation indicates the variance of the failure probability function. Therefore, FIG. 10 displays the sensor numbers, sensor names, and coefficients of variation for the top (1st to 5th) sensor data highly correlated with a failure from the time-series operation data. In this display, the time-series operation data selected by the time-series operation data selection unit 13 is displayed on the screen of the display unit 12. This allows the user to quantitatively grasp information such as what factors are causing the component failure and how much the factors are correlated with the failure compared to the total operation time. Therefore, in this embodiment, it is possible to provide useful information not only for the maintenance and operation of the wind power generator 1 but also for the design and development of the wind power generator. [Example]

[0091] In the first embodiment, the failure probability for an occurred failure is evaluated, but in the second embodiment, the future failure probability is predicted and evaluated. FIG. 11 is a system configuration diagram in the second embodiment. In this embodiment, the future failure probability is predicted and evaluated in the failure probability evaluation unit 10. The difference from the failure probability evaluation unit 10 shown in FIG. 9 in the first embodiment is that an operation status prediction unit 34 is provided. The operation status prediction unit 34 receives the time-series operation data 5 and the learned lifespan model 21 as input, and outputs an explanatory variable D(x) at a future point in time.

[0092] The method for calculating future failure probability is explained below. For units that are surviving at the present time (t0), the failure probability after an arbitrary time has elapsed is calculated separately. First, the predicted value 35 of the explanatory variable D(x) after an arbitrary time (Δt) has elapsed is predicted in advance by the operation status prediction unit 34. At this time, predictions must be made individually for each unit. The simplest method is to assume that the average value of the time-series operation data 5 recorded up to the present time will continue into the future. When using operation data that is seasonally dependent, such as wind conditions or temperature, it is desirable to make estimates by referring to seasonal trends and forecasts by weather forecasting agencies.

[0093] Alternatively, several future operation scenarios may be assumed, and time-series operation data may be generated arbitrarily for each of them. In addition to the above, time-series prediction may be performed using a state space model such as a Kalman filter. Note that the specific prediction method is not limited to the above.

[0094] The estimated value of the estimated or generated time-series operation data can be applied to the lifespan model 21 recorded in the lifespan model recording unit 18 to calculate the explanatory variable value 35 D(t0+Δt) at a future time point. Based on this and the explanatory variable D(t0) at the current time, the probability P that an individual currently in operation will fail after an arbitrary time (Δt) has elapsed can be calculated as a conditional probability in accordance with (Equation 12) in the failure probability output unit 19.

[0095]

number

[0096] Here, F(D) is the failure probability function that identifies the correspondence between time-series operational data (input) and failure probability (output). Failure probability is the expected number of failure events that will occur after a given amount of time (Δt) has passed. This value increases if Δt is made longer, but it is usually desirable to set Δt based on the expected interval between regular inspections of the mechanical system, and within such a set range of Δt, it is unlikely that the failure probability of each individual will be close to 1.0. However, it is possible for the total failure probability across all individuals to exceed 1.0. This total value is the expected number of failure events that will occur across all individuals under consideration.

[0097] Therefore, by reflecting the total value of the failure probability in, for example, the parts inventory management system 36, it is possible to optimize the inventory status of replacement parts. Alternatively, if an operation planning system 37 is connected to the wind power generator system, a method of changing the operation plan using the failure probability may be adopted. For example, if the failure probability until a future scheduled periodic inspection is higher than expected, changing the operation plan to actively shut down the wind power generator or reduce its output can extend the life of the parts. This change in the operation plan will change the future operating status, which will inevitably change the future accumulated damage. In this case, it is desirable to reflect the operation plan in the calculation of future accumulated damage in the operation status prediction unit. This configuration allows the user to easily check the relationship between changes in the operation plan and changes in the failure probability.

[0098] The calculation method of this embodiment and the current and future failure probabilities calculated using this method can be applied to the management (inventory management and operation planning) of machinery systems and facilities such as factories and plants other than wind power generator systems. An example of this application will be described in Example 3. [Example]

[0099] In Example 3, the future failure probability calculated in Example 2 is applied to machinery insurance. As an example, a case will be described in which machinery insurance rates are determined using the failure probability. Specifically, before the start of insurance operations, this example is applied to insured equipment to construct a lifespan prediction model. Machinery insurance rates are designed based on this lifespan prediction model. To reduce calculation costs, model generation and updating are not necessarily performed during the insurance period (usually one year). At the time of insurance renewal (rate calculation), the lifespan prediction model is updated based on newly obtained time-series operation data and maintenance history data, and the model is updated. Furthermore, this example may be applied not only to machinery insurance, but also to insurance that covers fire, corrosion due to aging, rust, and other types of failures.

[0100] In this way, when applied to insurance, processing may be performed in a cloud system as shown in FIG. 12. FIG. 12 is a system configuration diagram of Example 3. In FIG. 12, the failure probability evaluation system 100, excluding the display unit 12, is connected to an insurance company system 101 and an asset owner system 102 via a network 1000. The insurance company system 101 and the asset owner system 102 have a network such as an intranet, and each information processing device (terminal or server) can access the failure probability evaluation system 100. Also, like the terminal 103, it may be connected to the network 1000 without going through an intranet or the like. Note that the display unit 12 is assumed to be included in each of these terminals. Note that even when failure probability evaluation is performed on a cloud system as described above, it is not necessarily necessary to constantly generate and update the damage model; it may be performed as appropriate. Note that each of these devices can be realized by a so-called computer. [Example]

[0101] The failure probability evaluation systems 100 of Examples 1 to 3 can each be realized by a computer. An example of this realization is shown in Fig. 13. Fig. 13 is a hardware configuration diagram of the failure probability evaluation system 100 in Example 4. As shown in Fig. 13, the failure probability evaluation system 100 has a processing device 111, a memory 112, a network interface 113, and a secondary storage device 114, which are connected to each other via a communication path such as a bus.

[0102] First, the processing device 111 is a so-called processor such as a CPU, and executes processing in accordance with a failure probability evaluation program 115 stored in a secondary storage device 114. This processing is the processing of each unit shown in Examples 1 to 3. Furthermore, the memory 112 expands the failure probability evaluation program 115 used for processing in the processing device 111 and each of the above-mentioned data.

[0103] Furthermore, the secondary storage device 114 stores the failure probability evaluation program 115 and the above-mentioned data, that is, the time-series operation database 2 and the maintenance history database 3. The secondary storage device 114 may be realized by various storage media such as an HDD (Hard Disk Drive), an SSD (Solid State Drive), or a memory card. Furthermore, it may be realized by a device separate from the failure probability evaluation system 100, such as a file server. [Explanation of symbols]

[0104] 1. Wind turbine 2...Time-series operational database 3...Maintenance history database 4...Worker 5...Time-series operation data 6...Maintenance history data 12...Display section 13...Time series operation data selection section 14...Feature calculation unit 15...Fault correlation calculation section 16...Selected time series operation data number determination section 17...Life Modeling Section 18...Life model recording section 19...Failure probability output section 20...Lifespan model 100...Failure probability evaluation system 101...Insurance Company System 102...Asset Owner System 103...Terminal 1000…Network

Claims

1. A failure probability evaluation system for evaluating the failure probability of a component that constitutes a mechanical system, a maintenance history database that stores maintenance history data of the mechanical system; a time-series operation database that stores a plurality of time-series operation data indicating the operation status of the part; a time-series operation data selection unit that selects time-series operation data having small variations in a failure probability function for calculating a failure probability for each of the time-series operation data; a failure probability evaluation unit that inputs the selected time-series operation data and the maintenance history data, identifies a correspondence relationship of learning data and a correspondence relationship of test data, which are correspondence relationships of failure probabilities with the input time-series operation data, and calculates a predictive performance index of a lifespan model from the correspondence relationship of the learning data and the correspondence relationship of the test data; a time-series operation data selection unit that selects time-series operation data in accordance with the calculated predicted performance index; A failure probability evaluation system that uses the selected time-series operation data preferentially to evaluate the failure probability.

2. 2. The failure probability evaluation system according to claim 1, The failure probability evaluation unit is a failure probability evaluation system that calculates a failure probability using the maintenance history data and the selected time-series operation data.

3. 3. The failure probability evaluation system according to claim 2, The failure probability evaluation unit includes a failure probability function identification unit that calculates a failure probability function of the mechanical system by statistical processing based on the maintenance history data; an explanatory variable generation and update unit having a function of generating explanatory variables for a failure probability function that minimizes the variation of the failure probability function using the time-series operation data; The failure probability function identification unit is a failure probability evaluation system that provides a failure probability function that minimizes the variation in lifespan.

4. 2. The failure probability evaluation system according to claim 1, The time-series operation data selection unit selects operation history data with the smallest degree of deviation, which indicates the degree of deviation between the correspondence relationship of the learning data and the correspondence relationship of the test data.

5. 5. The failure probability evaluation system according to claim 4, A failure probability evaluation system in which the correspondence relationship between the time-series operation data and the failure probability is indicated by a set of pairs of a predetermined number of top operation data items among the time-series operation data items that have a high correlation with the failure and the failure probability, or by a parametric or non-parametric failure probability function.

6. In the failure probability evaluation system according to claim 1, A failure probability evaluation system in which the prediction performance index is a dissimilarity between a set of pairs of correspondence relationships of failure probabilities with respect to the time-series operation data, a deviation rate between failure probability functions or Kaplan-Meier curves that define the input-output relationship of the correspondence relationships of failure probabilities with respect to the time-series operation data, or a difference in the variability itself of the correspondence relationships of failure probabilities with respect to the time-series operation data.

7. A failure probability evaluation method for evaluating the failure probability of components that make up a mechanical system. Leave, storing maintenance history data of the mechanical system in a maintenance history database; storing a plurality of time-series operation data indicating the operation status of the part in a time-series operation database; a time-series operation data selection unit selecting time-series operation data having a small variation in a failure probability function for calculating a failure probability for each of the time-series operation data; a failure probability evaluation unit inputs the selected time-series operation data and the maintenance history data, identifies a correspondence relationship of learning data and a correspondence relationship of test data, which are correspondence relationships of failure probabilities with the input time-series operation data, and calculates a prediction performance index of a life model from the correspondence relationship of the learning data and the correspondence relationship of the test data; a time-series operation data selection unit selecting time-series operation data in accordance with the calculated prediction performance index; A failure probability evaluation method for evaluating the failure probability by preferentially using the selected time-series operation data.

8. 8. The failure probability evaluation method according to claim 7, The failure probability evaluation method, wherein the failure probability evaluation unit calculates a failure probability using the maintenance history data and the selected time-series operation data.

9. 9. The failure probability evaluation method according to claim 8, The failure probability evaluation unit calculating a failure probability function of the mechanical system by statistical processing based on the maintenance history data; Using the time-series operation data, an explanatory variable of the failure probability function that minimizes the variance of the failure probability function is generated; A failure probability evaluation method that provides a failure probability function that minimizes the variation in life.

10. 8. The failure probability evaluation method according to claim 7, The failure probability evaluation method, wherein the time-series operation data selection unit selects operation history data with the smallest degree of deviation, which indicates the degree of deviation between the correspondence relationship of the learning data and the correspondence relationship of the test data.

11. 11. The failure probability evaluation method according to claim 10, A failure probability evaluation method in which the correspondence relationship between the time-series operation data and the failure probability is expressed by a set of pairs of a predetermined number of top operation data items with high correlation with failures among the time-series operation data and the failure probability, or by a parametric or non-parametric failure probability function.

12. The failure probability evaluation method according to claim 7, a failure probability evaluation method in which the prediction performance index is a dissimilarity between a set of pairs of correspondence relationships of failure probabilities with respect to the time-series operation data, a deviation rate between failure probability functions or Kaplan-Meier curves that define an input-output relationship of the correspondence relationships of failure probabilities with respect to the time-series operation data, or a difference in the variability itself of the correspondence relationships of failure probabilities with respect to the time-series operation data.

Images