Failure probability evaluation system and failure probability evaluation method

Abstract

To provide a fault probability evaluation system capable of effectively maintaining prediction precision of a life model in the future, while restraining a calculation load.SOLUTION: A fault probability evaluation system 100 for evaluating a fault probability of a fault in a component for composing a mechanical system includes: a fault event database DB1 for storing past fault event data D1 for the mechanical system; a time-series operation database DB2 for storing time-series operation data expressing an operation state of the mechanical system; a first life modeling unit 110 for estimating a fault probability function composing a life model of the mechanical system by using the time-series operation data and fault history data; and a second life modeling unit 120 for updating parameters of the fault probability function by using time-series operation data including the added time-series operation data after estimation of the fault probability function and fault history data including added fault history data.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] This invention relates to a technique for evaluating the failure probability of an object. In particular, it relates to failure diagnosis and prediction (prediction), including the calculation of failure probability. The objects covered include equipment, facilities, mechanical systems, groups of mechanical systems formed therefrom, and the components that constitute them. [Background technology]

[0002] In mechanical systems such as power generation equipment, transportation equipment, and various other industrial machines, it is desirable that the mechanical system functions normally. To achieve this, it is important to appropriately understand and manage the failure risk of each component and to carry out maintenance, such as repair or replacement of each component, at the appropriate time. When managing and operating multiple identical machines, statistical analysis of past failure and maintenance records makes it possible to predict the number of corrective and proactive maintenance sessions that may occur in the future, as well as the remaining lifespan of the target equipment. Here, failure records and maintenance records refer to data in which the content of the failure or maintenance and the time of occurrence are recorded in pairs.

[0003] In statistical analysis using maintenance records, it is necessary to estimate the failure probability density function f(t), failure probability function F(t), and failure rate function λ(t) required to predict the number of corrective and proactive maintenance cycles that may occur in the future, as well as the remaining lifespan of the target equipment. Techniques for this purpose are shown in Non-Patent Document 1, etc. Here, t is the operating time or running time of the equipment (hereinafter, operation and running will be referred to as running). 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] Furthermore, Non-Patent Document 1 assumes that the operating conditions of a mechanical system are static, that is, unchanging in space and time. However, in general, the operating conditions of mechanical systems are not constant in space and time. For example, the operating conditions of a wind turbine change moment by moment depending on wind conditions, and the load also differs depending on the site conditions. Similarly, the load on construction machinery changes daily due to different work items, the operator's driving characteristics, soil characteristics, and other working environment factors. Therefore, simply evaluating failure probability functions or failure rate functions based on operating time has limitations in estimating the number of maintenance cycles and remaining lifespan.

[0007] In recent years, various sensors have been attached to mechanical systems, making it easy to access data on how the mechanical systems are used and their operating environment via networks. Patent Document 1 discloses a maintenance parts life prediction system that enables highly accurate estimation of the remaining lifespan for individual devices by appropriately considering various reliability data such as usage / operating environment data and failure / maintenance data. Patent Document 2 discloses a failure probability evaluation system that enables more highly accurate estimation of the number of failures and lifespan by having a function to automatically generate a damage model that leads to the failure of a mechanical system based on time-series operation data and failure data of the mechanical system obtained from sensors. [Prior art documents] [Non-patent literature]

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

[0009] [Patent Document 1] International Publication No. 12 / 157040 [Patent Document 2] Japanese Patent Publication No. 2019-160128 [Overview of the Initiative] [Problems that the invention aims to solve]

[0010] However, in mechanical systems used over long periods, it is difficult to avoid changes in the system itself over time, and the possibility of degradation or drift of sensors attached to the mechanical system cannot be ruled out. Furthermore, since the control method of the mechanical system may change during operation, there is no guarantee that a lifetime model created at the present time based on past reliability data, as described in Patent Documents 1 and 2, will have the same level of predictive accuracy in the future. Therefore, in order to maintain predictive accuracy, it is desirable to periodically update the lifetime model. However, frequent lifetime modeling leads to increased computational costs and limits the number of computers with the computational power to perform it, which can hinder the use of cloud computing and other technologies. Therefore, the present invention aims to maintain the predictive accuracy of the lifetime model while suppressing the computational load. [Means for solving the problem]

[0011] To achieve the above objective, the present invention estimates the failure probability function of a machine system using time-series operation data and failure history data of the machine system, and updates the parameters of the estimated failure probability function using time-series operation data and failure history data, including additional time-series operation data and additional failure history data of the machine system after estimation.

[0012] Specifically, in a failure probability evaluation system for evaluating the failure probability of components constituting a mechanical system, the system includes a failure history database that stores past failure history data for the mechanical system, a time-series operation database that stores time-series operation data representing the operating status of the mechanical system, and a failure probability function that constitutes the life model of the mechanical system using the time-series operation data and failure history data. At a predetermined estimated frequency Using the first lifetime modeling unit that estimates the lifespan, and the time-series operation data including additional time-series operation data after the estimation of the failure probability function, and the failure history data including additional failure history data, With an update frequency higher than the estimation frequency of the failure probability function by the first lifetime modeling unit, A failure probability evaluation system having a second life modeling unit that updates parameters of the failure probability function. Furthermore, the present invention also includes a failure probability evaluation system for evaluating the failure probability of a component constituting a mechanical system, comprising: a failure history database for accumulating past failure history data for the mechanical system; a time-series operation database for storing time-series operation data representing the operating state of the mechanical system; a first life modeling unit for estimating a failure probability function constituting a life model of the mechanical system using the time-series operation data and failure history data; and a second life modeling unit for updating the parameters of the failure probability function using the additional time-series operation data and additional failure history data when information contributing to prediction accuracy is added to the time-series operation data and failure history data, including additional time-series operation data and additional failure history data, after the estimation of the failure probability function. The present invention also includes a failure probability evaluation using the failure probability evaluation system.

Advantages of the Invention

[0013] According to the present invention, by updating parameters, it becomes possible to maintain the prediction accuracy of the life model while suppressing the computational load. Problems, configurations, and effects other than those described above will be clarified by the description of the following embodiments.

Brief Description of the Drawings

[0014] [Figure 1] System configuration diagram when Example 1 is applied to a plurality of wind turbines 1. [Figure 2] Hardware configuration diagram of the failure probability evaluation system 100 in Example 1. [Figure 3] Diagram showing the failure event database DB1 in Example 1. [Figure 4] Diagram showing the data D31 of the failure time in Example 1. [Figure 5] Diagram showing the data D32 of the survival time in Example 1. [Figure 6] Diagram showing the survival analysis data D3 provided by the explanatory variable formula generation unit 111 in Example 1. [Figure 7] Diagram showing the survival analysis data D3 provided by the explanatory variable formula generation unit 111 in Example 1. [Figure 8] Diagram showing the change in the survival analysis data D4 in Example 1. [Figure 9] Diagram showing the update of the survival time analysis data in Example 1. [Figure 10] Diagram showing the configuration of a computer system that realizes the failure probability evaluation system 100 in Example 1. [Figure 11] Diagram showing an example of a graphical user interface (GUI) suitable for the display unit 2 in Example 1. [Figure 12]System configuration diagram when Example 2 is applied to multiple wind turbines 1. [Figure 13] This is a hardware configuration diagram of the failure probability evaluation system 100 when the failure probability evaluation system 100 is implemented in the cloud in Example 1. [Modes for carrying out the invention]

[0015] One embodiment of the present invention will be described below. In this embodiment, the failure probability function of a machine system to be managed is estimated using time-series operation data and failure history data of the machine system, and the parameters of the estimated failure probability function are updated using time-series operation data and failure history data, including additional time-series operation data and additional failure history data of the machine system after estimation. More specifically, the failure probability evaluation system 100 for evaluating the failure probability of failures in components constituting a machine system includes a failure event database DB1, which is a failure history database that stores failure event data D1, which is past failure history data of the machine system; a time-series operation database DB2, which stores time-series operation data representing the operating status of the machine system; a first life modeling unit 110 that estimates a failure probability function constituting a life model of the machine system using the time-series operation data and failure history data; and a second life modeling unit 120 that updates the parameters of the failure probability function using time-series operation data, including additional time-series operation data, and failure history data, including additional failure history data, after the estimation of the failure probability function. Furthermore, the present invention also includes a failure probability evaluation method using a failure probability evaluation system.

[0016] In the following sections, more specific examples of this embodiment will be described with reference to the drawings. The failure probability evaluation system 100 is implemented on a computer, and its functions are executed by a processing unit according to a program. However, its functions may also be implemented using dedicated hardware, and the present invention is not limited to the use of a program (software).

[0017] Furthermore, in each of the following embodiments, a wind turbine 1 is used as the mechanical system, and the failure probability of the wind turbine 1 and its components is evaluated. However, the application of the present invention is not limited to the wind turbine 1. [Examples]

[0018] In the following, we will describe Example 1 using a failure probability evaluation system 100 for the components of a wind turbine 1 as an example, with reference to the drawings. However, the application of the present invention is not limited to wind turbines.

[0019] Furthermore, as is clear from (Equation 1) and (Equation 2), if we can identify any of the failure probability density function f, failure probability function F, or failure rate function λ, we can calculate the other functions. Therefore, in the following examples, when identifying the probability distribution of failures, we use the failure probability density function f and the failure probability function F as needed.

[0020] Figure 1 is a system configuration diagram when Example 1 is applied to multiple wind turbines 1. In Figure 1, the failure probability evaluation system 100 includes a failure event database DB1, a time-series operation database DB2, a first life modeling unit 110, a second life modeling unit 120, and a failure probability evaluation unit 130.

[0021] First, the first lifetime modeling unit 110 uses the failure event data D1 and the time-series operation data D2 to estimate at least the failure probability function F1 that constitutes the lifetime model 113. Then, the first lifetime modeling unit 110 creates the lifetime model 113, which includes the failure probability function F1. To do this, the first lifetime modeling unit 110 takes the failure event data D1 and the time-series operation data D2 stored in the failure event database DB1 and the time-series operation database DB2 as input. Then, the first lifetime modeling unit 110 outputs the lifetime model 113 to the second lifetime modeling unit 120. Here, the lifetime model 113 includes the failure probability function F1 and the explanatory variable formula D(x) for generating its explanatory variables. In other words, the first lifetime modeling unit 110 estimates these failure probability function F1 and the explanatory variable formula D(x) for generating its explanatory variables. Note that the explanatory variable formula is written as D(x) to explicitly indicate that it is a function of the time-series operation data D2.

[0022] Furthermore, the second lifetime modeling unit 120 updates the parameters of the failure probability function F1 included in the lifetime model 113 based on at least the failure event data D1 and the time-series operation data D2. Here, the failure event data D1 used in the second lifetime modeling unit 120 includes additional failure event data D1 added since the estimation of the failure probability function F1 in the first lifetime modeling unit 110. Similarly, the time-series operation data D2 includes additional time-series operation data D2 added since the estimation of the failure probability function F1 in the first lifetime modeling unit 110.

[0023] More preferably, the second lifetime modeling unit 120 takes failure event data D1, time-series operation data D2, and lifetime model 113 as input. The second lifetime modeling unit 120 also outputs the explanatory variable equation D(x). Furthermore, the second lifetime modeling unit 120 outputs to the failure probability evaluation unit 130 the explanatory variables 124 and the failure probability function F2, which has been updated with the parameters of the failure probability function F1 since the estimation of the failure probability function F1 in the first lifetime modeling unit 110. Therefore, the second lifetime modeling unit 120 updates the parameters and estimates the failure probability function F2.

[0024] Furthermore, the failure probability evaluation unit 130 takes the time-series operating data D2, the explanatory variable equation D(x), the explanatory variable 124 at the present time mentioned above, and the failure probability function F2 as input and outputs the result. Based on the time-series operating data D2, which includes the additional time-series operating data, the failure probability evaluation unit 130 calculates the failure probability 133.

[0025] Here, the failure probability 133, which is the output of the failure probability evaluation system 100, is provided to a display unit 2 located inside or outside the failure probability evaluation system 100. The display unit 2 includes, for example, a smartphone, a PC, or a monitor. The output of the failure probability evaluation system 100 can also be provided to and utilized by a parts inventory management system 3 and an operation planning system 4 for multiple wind turbines. Furthermore, the recipients of the failure probability 133 are not limited to the owners or operating companies of the wind turbine 1. For example, the failure probability 133 can be provided to an insurance company for the purpose of designing insurance for the wind turbine 1. To this end, the failure probability evaluation system 100 is connected via a network to an insurance company system operated or used by the insurance company, and the failure probability 133 is transmitted to it. In addition, at least some of the functions of the failure probability evaluation system 100 may be provided to the insurance company system.

[0026] Next, Figure 2 is a hardware configuration diagram of the failure probability evaluation system 100 in Embodiment 1. As shown in Figure 2, the failure probability evaluation system 100 can be implemented in, for example, a general-purpose information processing device that includes a CPU 101, memory 102, storage device 103, input / output I / F 104, and communication I / F 105. Therefore, the functions of each part described using Figure 1 are realized by the CPU 101 loading a program previously stored in the storage device 103 into the memory 102 and executing it. The failure event database DB1, time-series operation database DB2, explanatory variable formula generation unit 111, failure probability function identification unit 112, lifetime model storage unit 121, explanatory variable calculation unit 122, failure probability function identification unit 123, operation status prediction unit 131, and failure probability calculation unit 132 are provided in the storage device 103 and memory 102.

[0027] Furthermore, the input / output interface 104 can be implemented, for example, by input devices such as a keyboard or mouse, display devices such as liquid crystal displays (LCDs) or organic EL displays, and printing output devices such as printers. In other words, the input / output interface 104 is used when a user of the failure probability evaluation system 100 makes any input or performs interactive processing regarding the failure probability evaluation system 100. The communication interface 105 is a LAN or satellite communication system and is used to communicate operating information of the wind turbine 1 to the failure probability evaluation system 100.

[0028] Furthermore, the information processing device that implements the failure probability evaluation system 100 does not need to be a single device; for example, the first life modeling unit 110 and the second life modeling unit 120 may be provided on different information processing devices.

[0029] The following sections will describe in detail the failure event database DB1, the time-series operation database DB2, the first life modeling unit 110, the second life modeling unit 120, and the failure probability evaluation unit 130.

[0030] First, the failure event database DB1 is stored in the storage device 103. The failure event database DB1 stores failure events of the components of the wind turbine 1. Here, a failure event is an event related to failure, such as "failure," "abnormality," or "parts replacement." If the target wind turbine 1 is equipped with a system for automatically detecting failure events, the automatic failure event detection system and the failure event database DB1 may be connected via a network to automatically store the data. Alternatively, the maintenance worker M may determine the failure event and register its details. With this configuration, failure events that have occurred in the past in multiple wind turbines 1 are stored in the failure event database DB1. Note that the maintenance worker M may be a user of the failure probability evaluation system 100, or the administrator may be a user.

[0031] Figure 3 shows the failure event database DB1 in Example 1. The failure event data D1 stored in the failure event database DB1 is information about malfunctions such as failures in the wind turbine generator 1, which is a mechanical system. Therefore, as illustrated in Figure 3, the failure event data D1 has the following items: data D11 of the individual unit that experienced the failure, time and date data D12, and failure details D13. Here, the data D11 of the individual unit that experienced the failure can be defined by the name of the site to which the wind turbine generator 1 in Figure 3 belongs and its unit number. The time and date data D12 indicates the date and time the failure occurred. Furthermore, the failure details D13 indicates the content of the failure and can be defined by the name of the part that experienced the failure, the location, and the event that caused it.

[0032] Furthermore, the time-series operation database DB2 is also stored in the storage device 103. The time-series operation database DB2 stores time-series operation data D2, which is time-series data related to the operation of the wind turbine 1. Large wind turbines 1 are usually equipped with a data collection system that collects wind condition data such as wind speed and wind direction, and operational data such as power generation amount and rotor rotation speed. In this embodiment, each of these data is stored as time-series operation data D2 in the time-series operation database DB2 via communication means such as a network.

[0033] In this case, there are no particular limitations on the data collection interval, but in this embodiment, since we are estimating the number of failures and remaining lifespan over a relatively long period of several months or several years, an interval of about one day is ideal. Furthermore, it is preferable to use statistical values ​​such as the maximum value, minimum value, mean value, and standard deviation within the collection interval for the time-series operational data D2, rather than measured values ​​sampled at arbitrary intervals. This makes it possible to maximize the use of information while significantly reducing the amount of data. In addition to simple statistical values ​​such as the mean value, a method may also be used in which the degree of fatigue damage is stored as time-series data as a result of fatigue damage analysis based on the linear cumulative damage law.

[0034] Here, fatigue damage level is a statistical quantity obtained by applying rain flow count processing or the like to the measured waveform. For example, if the target is equipped with sensors for acquiring load information, such as strain sensors or load cells, and it is clear that failure occurs due to the accumulation of fatigue damage, then adopting fatigue damage level as one of the time-series operational data is particularly effective. Thus, it is desirable that the data acquisition system on the wind turbine generator 1 side, or the time-series operational database DB2 on the failure probability evaluation system 100 side, be equipped with functions for performing preprocessing such as statistical processing and fatigue damage level analysis on the measured values.

[0035] Furthermore, the information stored in the time-series operation database DB2 is not limited to information obtained from the wind turbine 1 itself. For example, meteorological data such as temperature measured by weather observation equipment installed near the wind turbine 1 is also useful for evaluating the operating status of the wind turbine 1. In addition to simple time-series data, operational data that shows what kind of operations were performed is also useful information stored in the time-series operation database DB2. Operational data is, for example, data that shows how many times a certain operation was performed. For example, the number of times the energy storage system connected to the wind turbine 1 was charged falls into this category.

[0036] Furthermore, the first lifetime modeling unit 110 is mainly composed of an explanatory variable formula generation unit 111 and a failure probability function identification unit 112. First, the failure probability function identification unit 112 will be explained. For the sake of simplicity, the identification of the failure probability function using only failure event data D1 and without using time-series operation data D2 will be explained here. That is, in Figure 1, we assume a condition in which time-series operation data D2 is not input to the explanatory variable formula generation unit 111.

[0037] In this hypothetical scenario, first, the failure event data D1 stored in the failure event database DB1 is formatted into survival analysis data D3 for analysis by the explanatory variable formula generation unit 111 and provided to the failure probability function identification unit 112. Here, the survival analysis data D3 includes failure time data D31 and survival time data D32.

[0038] Here, Figure 4 shows the failure time data D31 in Example 1. Here, failure time refers to the operating time until the failure occurs. The operating time D31 is obtained from the failure event data D1. The failure event data D1 is data that associates failure events (data D11 of the individual that experienced the failure) with the time of occurrence D12, as shown in Figure 3. What is needed for identifying the failure probability function thereafter is the operating time until the failure occurs (hereinafter simply referred to as failure time). The failure time includes both the time since the previous failure event occurred (time between failure events) and the time from when the system started operating until the current failure event occurred.

[0039] Here, the failure event data D1 in Figure 3 records the time D12 when the previous failure occurred, so the failure time can be calculated from the difference between the time of the previous and current failures. Also, if the current failure is the first failure, the failure time can be calculated from the difference between the system start time and the time of the current failure. The failure time calculation process is performed by the explanatory variable expression generation unit 111, and the data is formatted into a format like the aggregated failure data D31 shown in Figure 4. Figure 4 is a diagram showing an example of the calculated failure time, for example, the failure time D31 was obtained from data of 200 wind turbines 1.

[0040] Figure 5 shows the survival time data D32 in Example 1. Here, failure time refers to the continuous operating time up to the present. When generating the aggregated failure data D31, the explanatory variable expression generation unit 111 formats it as aggregated survival data D32, as shown in Figure 5, which aggregates the time from the present to the previous failure event or the system operation start time.

[0041] Furthermore, the reason why the explanatory variable equation generation unit 111 generates survival analysis data D3, including failure time data D31 and survival time data D32, is as follows: In order for the failure probability function identification unit 112 to identify a more plausible failure probability function, it is necessary to reflect not only the failure event (failure time data D31 in Figure 4) but also the fact that the component remains in a healthy state even after continuous operation for a certain period of time. In fact, for example, it is necessary to reflect the survival time data D32 in Figure 5.

[0042] Normally, even if a failure occurs, the system is restored to a healthy state as quickly as possible through parts replacement or repair, and then restarted. Therefore, if we consider identifying the failure probability function at a certain time, most of the target parts at that time have continued to operate since the previous failure occurred, or since the system started operating. To reflect this fact, the explanatory variable formula generation unit 111 ultimately assigns a failure flag 22 to the aggregated failure data D31. Then, it assigns a survival flag 23 to the aggregated survival data D32, and integrates these to generate survival analysis data D3.

[0043] Figure 6 shows the survival analysis data D3 provided by the explanatory variable formula generation unit 111 in Example 1. As shown in Figure 6, the survival analysis data D3 consists of failure time data D31 with a failure flag 22 and survival time data D32 with a survival flag 23.

[0044] Furthermore, the failure probability function identification unit 112 identifies the failure probability function F1 by fitting a certain probability distribution to the survival analysis data D3 based on statistical methods. For simplicity, time-series operating data D2 is not considered here, and it is assumed that the explanatory variable leading to failure is simply represented by the cumulative operating time.

[0045] Furthermore, known techniques can be used to identify the failure probability function from data that includes failure data D31 and survival data D32 (censored data), such as survival analysis data D3, within the operating time, using methods such as maximum likelihood estimation or Bayesian estimation. Specifically, in the case of maximum likelihood estimation, parameters of the failure probability function are searched for in order to maximize the log-likelihood sum L defined by (Equation 3) for the assumed failure probability function F1 and the survival analysis data D3 (Patent Document 2).

[0046]

number

[0047] In (Equation 3), i and j are the numbers of the failed data D31 and the surviving data D32, respectively, and f1(t) is the failure probability density function of the failure probability function F1 obtained by (Equation 1). That is, the first term on the right-hand side of (Equation 3) represents the likelihood for the failed data D31, and the second term represents the likelihood for the surviving data D32. The failure probability function can be a normal distribution, log-normal distribution, gamma distribution, Weibull distribution, etc., but is not limited to those listed above.

[0048] Next, the explanatory variable formula generation unit 111 will be described. Simply evaluating the failure probability function based on operating time has limitations in estimating the number of maintenance cycles and remaining lifespan. Therefore, the explanatory variable formula generation unit 111 utilizes time-series operating data D2, which includes information on how the machine system is used and the operating environment. By converting this into explanatory variable D(D2) using the explanatory variable generation formula D(x) and using it as an explanatory variable for the failure probability function F1, it is possible to improve the prediction accuracy of the failure probability function F1. Here, "explanatory variable" is defined for each individual, similar to "operating days" in Figures 4 and 5. Therefore, even if "operating days" in Figures 4 and 5 is replaced with "explanatory variable," the above logic remains valid, and the survival analysis data D3 shown in Figure 7 is obtained. Figure 7 shows the survival analysis data D3 provided by the explanatory variable formula generation unit 111 in Example 1. In Figure 7, even when the variable of the failure probability function F1 is "explanatory variable D(D2)," the failure probability function identification method described above can still be applied.

[0049] In this case, when the components of a mechanical system such as a wind turbine 1 are complex and various load factors are involved in the failure, it is not easy for the user to quantify the influence of each factor based on physical mechanisms and define the explanatory variables that lead to component failure. In this case, a data-driven approach that automatically calculates the explanatory variable equation D(x) using failure event data D1 and time-series operation data D2 is effective. To achieve this, it is desirable to generate the explanatory variable equation D(x) multiple times, rather than generating it only once, and adopt the explanatory variable equation D(x) that best maximizes the prediction accuracy of the identified failure probability function. The following shows the configuration of the explanatory variable equation generation unit 111 that is desirable for this purpose.

[0050] First, it is desirable to use the variability V of the failure probability function as an evaluation metric for prediction accuracy (Patent Document 2). This is because a large variability V means that there is a range in the predicted failure time. Therefore, in order to minimize the prediction time range and estimate the next failure date and time within a short time range, it is necessary to reduce the variability V. Thus, the problem of determining the explanatory variable equation D(x) that maximizes prediction accuracy reduces to an optimization problem with the objective function being the variability V of the failure probability function. Furthermore, it is desirable to use the coefficient of variation obtained by dividing the standard deviation of the failure probability function F1 by the mean value as the variability V used as the objective function. This is because defining variability using standard deviation or variance does not allow for a unified evaluation of variability for different explanatory variables obtained from multiple trials. For this reason, the explanatory variable equation generation unit 111 calls the failure probability function identification unit 112 when generating the explanatory variable equation D(x) to evaluate the variability V of the failure probability function and repeatedly modifies the explanatory variable equation D(x) to minimize V.

[0051] The explanatory variable D(x) can be generated using a statistical model or a machine learning model, and the explanatory variable D can also be a multidimensional vector. However, since we are dealing with a mechanical system where failure is caused by the accumulation of damage, it is more desirable to use the cumulative damage model shown below. The cumulative damage model is a model that expresses the cumulative damage that leads to failure as a function of time-series operating data D2, and is represented by (Equation 4).

[0052]

number

[0053] Here, d(x) is the damage model per unit time, and xt is the operational data vector representing the t-th time-series operational dataset. Furthermore, the failures included in the malfunctions targeted by this embodiment are wear-out failures, among initial failures, random failures, and wear-out failures. Therefore, since we are dealing with the phenomenon in which failures are brought about by the accumulation of damage, we define the time integral of d(x) as the cumulative damage model D(x).

[0054] In this embodiment, the shape of the explanatory variable equation D(x) is not particularly limited. For example, the simplest way to express the explanatory variable D(D2) as a linear combination of time-series operational data D2, as shown in (Equation 5), is to use this method, which also requires relatively little computational cost for optimization.

[0055]

number

[0056] Here C=[c1, c 2、 ..., c N ] is a coefficient vector representing the weighting of the time-series operational data D2.

[0057] Alternatively, if the failure mechanism is known to some extent, a method may be adopted in which the user pre-defines only the shape of the equation according to the failure mechanism, and then searches for the coefficients. For example, it is empirically known that the temperature dependence of material degradation can be expressed by the Arrhenius equation shown in (Equation 6).

[0058]

number

[0059] Here, A is a constant, Ea is the activation energy, and T is the temperature data obtained as time-series operating data D2. Since k represents the rate of the degradation reaction, its value integrated over time can be considered equivalent to the accumulated damage. Therefore, when thermal load is the dominant factor in the degradation of the target, the temperature data obtained as time-series operating data D2 can be effectively considered by incorporating the Arrhenius equation shown in (Equation 6) into the damage model in (Equation 4).

[0060] In either method, the number of variables becomes the same as the number of undetermined coefficients in the damage model, resulting in a relatively large-scale optimization problem. Furthermore, since the objective function may be non-convex, it is desirable to use metaheuristics such as genetic algorithms or particle swarm optimization. Conversely, if the failure mechanism is completely unknown, a method that automatically searches for the equation shape itself using genetic programming (GP) may be employed.

[0061] However, adopting genetic programming (GP) increases the computational load. Therefore, the feasibility of adopting it requires careful consideration of the available computational resources. Regardless of the method adopted, the explanatory variable equation generation unit 111 repeatedly modifies the damage model and evaluates the variability V, and finally performs a convergence check to define a failure probability function that yields a smaller coefficient of variation. In the process of generating a damage model that minimizes the variability V, only the coefficient of variation is used as the objective function.

[0062] Therefore, in this process, only the smallness of the coefficient of variation may be prioritized, potentially leading to a smaller fit (likelihood) to the probability density function. To avoid this, one may set a constant value for the log-likelihood sum calculated by maximum likelihood estimation and use it as a constraint in updating the damage model, or solve the problem as a multi-objective optimization problem with both the coefficient of variation and the log-likelihood sum as objective functions.

[0063] Based on the above, it becomes possible to identify the explanatory variable equation D(x) and the failure probability function F1 that can minimize the variability V of the failure probability function for time-series operational data D2. When the explanatory variable D = D(D2) and the failure probability function is a Weibull distribution, the failure probability function F1 is given by (Equation 7).

[0064]

number

[0065] Here, α1 and β1 are the shape parameter and scale parameter of the Weibull distribution, respectively, when D is the explanatory variable.

[0066] The first lifetime modeling unit 110 outputs the explanatory variable equation D(x) and the failure probability function F1 (Equations 4, 5, and 7 in the case of the cumulative damage model and Weibull distribution) as a lifetime model 113 to the second lifetime modeling unit 120. The output lifetime model 113 is recorded in the lifetime model storage unit 121 of the second lifetime modeling unit 120.

[0067] As described above, generating the explanatory variable equation D(x) by the explanatory variable equation generation unit 111 requires a considerable computational load, so frequent generation of lifetime models leads to increased computational costs. Therefore, it is desirable to reduce the frequency of updating the explanatory variable equation D(x) and outputting a new lifetime model 113 to the second lifetime modeling unit 120, for example, at one-year intervals after the start of operation, except when the failure probability evaluation system 100 is put into operation. This will suppress the increase in computational costs, and if, for example, failure probabilities 133 are provided to insurance companies, it will be possible to provide a highly accurate lifetime model at the time of insurance renewal (when calculating premium rates).

[0068] Furthermore, the second lifetime modeling unit 120 is mainly composed of a lifetime model storage unit 121, an explanatory variable calculation unit 122, and a failure probability function identification unit 123. First, using Figure 8, we will explain the necessity of updating the lifetime model 113 during the operation of the failure probability evaluation system 100. Figure 8 is a diagram showing the changes in the survival analysis data D4 in Example 1. Here, for the sake of simplicity, Figure 8(a) is a reproduction of the survival time analysis data D3 from Figure 6, and Figure 8(b) is the survival time analysis data D4 180 days (about half a year) after Figure 8(a). Note that the individuals corresponding to each row in Figure 7 and Figure 8 are the same. For example, in the case of individual data 24, while it was operating for 30 days at the stage of Figure 8(a), a failure occurred at 190 days, so in Figure 8(b), the flag for survival time analysis has changed to "failed". Furthermore, for example, in the case of individual 25, the flag for survival time analysis remains "operational," but the number of days has increased to 355 days. Although not shown in Figure 8, data from newly operational individuals is also included in Figure 8(b).

[0069] As the failure probability evaluation system 100 is operated, new information is added to and modified in the failure event database DB1 and the time-series operation database DB2. In other words, additional failure event data and additional time-series operation data are stored. Therefore, the lifetime model 113 (explanatory variable formula D(x) and failure probability function F1) optimized for the failure event data D1 and time-series operation data D2 up to the point in Figure 8(a) is not guaranteed to be optimized at the point in Figure 8(b). For this reason, in order to maintain prediction accuracy, it is necessary to update the lifetime model 113 with the failure event data D1 and time-series operation data D2 obtained at the point in Figure 8(b) after the lifetime model 113 was generated by the first lifetime modeling unit 110.

[0070] When updating the lifetime model 113, the targets for updating are either the explanatory variable equation D(x) or the parameters of the failure probability function F1, or both. However, since the explanatory variable equation D(x) originally models the cumulative damage that leads to failure of the target mechanical system, it is thought to have a high correlation with the failure mechanism and not change significantly in the short term. On the other hand, even if the failure mechanism itself is system-specific and does not change significantly even with changes in the mechanical system over time, the following risks cannot be eliminated.

[0071] If sensors attached to a mechanical system deteriorate or drift, or if there is a sudden change in the control method of the mechanical system, the failure probability function F1, which was optimized for the operating state before such changes occurred, will no longer fit the failure probability of the mechanical system after the change. Therefore, in order to adapt to such risks, this embodiment suppresses the updating of the explanatory variable formula D(x), which is computationally expensive and is not expected to change significantly in the short term.

[0072] Then, based on the failure event data D1 and time-series operation data D2 obtained after generating the lifetime model 113, the parameters of the failure probability function F1 are updated periodically, for example, regularly, to match the failure probability of the actual machine. As a result, a failure probability function F2 used for failure probability evaluation is generated. Parameter identification of the failure probability function can be performed quickly using methods such as maximum likelihood estimation. Therefore, by updating only the parameters of the failure probability function F1 and increasing the update frequency compared to the modeling frequency of the lifetime model 113, it becomes possible to provide a failure probability evaluation system 100 that can maintain high prediction accuracy while keeping computation costs down.

[0073] The following describes the specific parameter update method in the second lifetime modeling unit 120. In the second lifetime modeling unit 120, the generation of the explanatory variable equation D(x) can be omitted. The explanatory variable calculation unit 122 retrieves the explanatory variable equation D(x) stored in the lifetime model storage unit 121. The explanatory variable calculation unit 122 then uses the time-series operation data D2 obtained after the lifetime model 113 was generated in the first lifetime modeling unit 110 to calculate the current explanatory variable 124 (i.e., D(D2)). The explanatory variable calculation unit 122 also adds the failure event data D1 obtained after the lifetime model 113 was generated in the first lifetime modeling unit 110 to the above explanatory variable 124.

[0074] As a result, the explanatory variable calculation unit 122 updates the survival time analysis data D3 used when creating the lifetime model 113 to survival time analysis data D4. Specifically, the survival time analysis data D3 used when creating the lifetime model 113 shown in Figure 9(a) is updated to the survival time analysis data D4 shown in Figure 9(b). Note that Figure 9(a) is the same as Figure 7. Thus, Figure 9 is a diagram showing the update of the survival time analysis data in Example 1.

[0075] Furthermore, the failure probability function identification unit 123 identifies the failure probability function F2 for the survival time analysis data D4. The failure probability function F1 and the failure probability function F2 are assumed to be of the same type. Therefore, if the failure probability function F1 is a Weibull distribution, then the failure probability function F2 will also be a Weibull distribution. When using maximum likelihood estimation during identification, the parameters of the failure probability function are searched for to maximize the log-likelihood sum L defined by (Equation 3) for the assumed failure probability function F2 and the survival analysis data D4. Also, if the failure probability function is a Weibull distribution, similar to (Equation 7), the failure probability function F2 is given by (Equation 8).

[0076]

number

[0077] Here, α2 and β2 are the shape parameter and scale parameter of the Weibull distribution, where D is the explanatory variable. The failure probability function F2 is stored in the lifetime model memory unit 121.

[0078] Through the above process, when the failure probability function F2 is identified for the first time by the failure probability function identification unit 123, the parameters of the failure probability function recorded in the lifetime model storage unit 121 are updated from (α1, β1) to (α2, β2). In subsequent identifications, the parameters of the failure probability function F2 recorded in the lifetime model storage unit 121 are also updated according to equation (8).

[0079] The second lifetime modeling unit 120 outputs the failure probability function F2, the explanatory variable generation formula D(x), and the current explanatory variables 124 to the failure probability evaluation unit 130. Note that immediately after the failure probability evaluation system 100 begins operation, until the parameters of the failure probability function F1 are updated, the failure probability function F1 is identical to the failure probability function F2.

[0080] Here, there is no restriction on the update frequency of the parameters, and for example, it may be performed every day. By increasing the update frequency in this way, when the failure probability function begins to deviate from the failure occurrence probability of the mechanical system, it becomes possible to promptly recover the prediction accuracy. Also, in the failure event data D1, it is also effective to update the parameters only at the timing when the failure flag 22 occurs. This is because when comparing the values inside the log functions of the first and second terms on the right side of (Equation 3) in the identification of the failure probability function, even if the probability density function f itself exceeds 1.0, its integral value is 1.0 or less. For this reason, it can also be seen that the failure data D31 has a greater impact. Thus, by updating the parameters when information contributing to the prediction accuracy is input, the prediction accuracy can be efficiently maintained while reducing the computational load. Note that "contributing to the prediction accuracy" includes cases where the prediction accuracy after update improves by a predetermined value or more, as well as cases where (when updated) a predetermined type and / or content of information is included.

[0081] Further, the failure probability evaluation unit 130 is mainly composed of an operation status prediction unit 131 and a failure probability calculation unit 132, and calculates the current failure probability F n and the failure probability F t after an arbitrary time in the future, respectively. When the user wants to evaluate the current failure probability, the failure probability evaluation unit 130 outputs F n as the failure probability 133, and when the failure probability F t after an arbitrary time in the future is desired, it outputs F t as the failure probability 133.

[0082] First, the method for evaluating the current failure probability F n will be described. The failure probability calculation unit 132 receives the current explanatory variable 124 for each individual from the explanatory variable calculation unit 122 of the second life modeling unit 120, and substitutes it into the failure probability function F2 (in the case of the Weibull distribution, (Equation 8)) to evaluate the current failure probability F n .

[0083] The failure probability F after an arbitrary timet This can be determined as follows. Here, the current time is t0 and an arbitrary time is Δt, and the explanatory variables 134 up to the time elapsed are estimated in advance by the operating status prediction unit 131.

[0084] While this embodiment does not impose any restrictions on the specific prediction method, it is necessary to make predictions individually for each individual. Specifically, estimates of the time-series operational data D2 up to a certain point in time for each individual are obtained by some method. The simplest method is to assume that the average value of the time-series operational data D2 recorded up to the present will continue in the future, which is equivalent to simply extrapolating the trend of the explanatory variable up to the present.

[0085] Furthermore, when using seasonally dependent operational data such as wind conditions and temperature, it is desirable to make estimations by referring to seasonal trends and forecasts from meteorological forecasting agencies. Alternatively, several future operational scenarios may be assumed, and time-series operational data D2 may be arbitrarily generated for each of them.

[0086] Furthermore, by substituting the estimated values ​​of the estimated or generated time-series operational data D2 into the explanatory variable formula D(x) defined by the explanatory variable formula generation unit 111, which is recorded in the lifetime model storage unit 121, D(t0+Δt) can be calculated as the future explanatory variable 134. Based on the future explanatory variable 134 and the current explanatory variable 124 (D(t0)), the probability F that an currently operational unit will fail after an arbitrary time (Δt) is calculated. t This can be calculated as a conditional probability P according to (Equation 9).

[0087]

number

[0088] Here, F2(D) is the failure probability function identified by the failure probability function identification unit 123. The failure probability P is the expected number of failure events after an arbitrary time (Δt) has elapsed. This value increases as Δt is taken to be longer, but it is usually desirable to set Δt based on the periodic inspection interval of the machine system, and within such a range of Δt settings, it is unlikely that the failure probability P of each individual will be close to 1.0. However, it is possible that the sum of the failure probabilities P for all individuals exceeds 1.0. This sum is the expected number of failure events for all individuals under consideration.

[0089] Therefore, by reflecting the total failure probability in, for example, the parts inventory management system 3, it becomes possible to optimize the inventory status of replacement parts. Alternatively, if the operation planning system 4 is connected to the wind turbine system, a method of changing the operation plan using the failure probability P may be adopted. For example, if the failure probability P until the scheduled periodic inspection is higher than expected, the operation plan can be changed to actively stop the wind turbine 1 or reduce its output, thereby extending the lifespan of the parts. Since this change in the operation plan will also change the future operating conditions and inevitably change the future cumulative damage, it is desirable to reflect the operation plan 5 in the calculation of future explanatory variables 134 in the operating conditions prediction unit 131 when adopting this method. With such a configuration, users can easily check the relationship between changes in the operation plan and changes in the failure probability.

[0090] The explanatory variable formula generation unit 111, failure probability function identification unit 112, lifetime model storage unit 121, explanatory variable calculation unit 122, failure probability function identification unit 123, operating status prediction unit 131, and failure probability calculation unit 132 described above are each implemented as computer programs. However, the above description does not limit the implementation form on a specific computer. Implementation examples will be explained later using Figures 10 and 13.

[0091] However, since the explanatory variable equation generation unit 111 needs to perform relatively computationally expensive calculations by repeatedly calling the failure probability function identification unit 112, it is ideal for both units to be implemented on the same computer.

[0092] Furthermore, the failure probability evaluation unit 130 evaluates the failure probability based on the failure probability function F2 obtained from the life model storage unit 121 of the second life modeling unit 120 and the current explanatory variables 124 calculated by the explanatory variable calculation unit 122. For this reason, it is desirable that both of these be implemented on the same computer.

[0093] Figure 10 shows the configuration of a computer system that implements the failure probability evaluation system 100 in Example 1. In this example, the failure probability of multiple wind turbine groups 1 is evaluated. For this purpose, a second life modeling unit 120, a failure probability evaluation unit 130, a failure event database DB1, and a time-series operation database DB2 are provided for each (corresponding to computers 200 and 210 in Figure 10). The second life modeling unit 120 and the failure probability evaluation unit 130 are implemented on the same computer for computers 200 and 210, respectively. In this example, the computational cost of updating the parameters of the failure probability function F2 by the second life modeling unit 120 is low. Therefore, the number of computers with the computational power to perform this is not limited, and for example, it is possible to select a low-cost cloud service with reduced computational power and memory capacity as the computer on which 200 and 210 are implemented.

[0094] Furthermore, in the first lifetime modeling unit, the explanatory variable equation generation unit 111 and the failure probability function identification unit 112 are implemented on the same computer 300 and connected to computers 200 and 210 via network 6. Computer 300 can be a local PC, a server, or a cloud service. The computer 300 on which the first lifetime modeling unit 110 is implemented requires computing power and memory capacity, resulting in a higher cost per unit. However, the modeling frequency of the lifetime model 113 is lower than the parameter update frequency of the second lifetime modeling unit. Therefore, it can be reused for modeling multiple groups of wind turbines. Consequently, even if the number of evaluation targets (groups of wind turbines) increases, the cost required for the first lifetime modeling can be substantially reduced, enabling the provision of a highly scalable service.

[0095] Here, Figure 13 is a hardware configuration diagram of the failure probability evaluation system 100 when the failure probability evaluation system 100 is implemented in the cloud in Example 1. In other words, in this figure, the aforementioned computers 200 and 210 are shown as the failure probability evaluation system 100. As shown in Figure 13, the failure probability evaluation system 100 has a processing unit 1001, a memory 1002, a network interface 1003, and a sub-storage device 1004, which are connected to each other via a communication path such as a bus.

[0096] First, the processing unit 1001 is a so-called processor, like the CPU 101 mentioned above, and executes processing according to the failure probability evaluation program 1005 stored in the sub-memory 1004. This processing is the processing of each part shown in Figure 1. The memory 1002 stores the failure probability evaluation program 1005 used for processing in the processing unit 1001 and the various data mentioned above.

[0097] Furthermore, the secondary storage device 1004 stores the failure probability evaluation program 1005 and the aforementioned data, namely the time-series operation database DB1 and the failure event database DB2. The secondary storage device 1004 may be implemented using various storage media such as an HDD (Hard Disk Drive), SSD (Solid State Drive), or memory card. Moreover, it may be implemented as a separate device from the failure probability evaluation system 100, such as a file server. The failure probability evaluation system 100 may also be connected to a terminal device used by users. In this case, the display unit 2 and input device described later are implemented on this terminal device.

[0098] Finally, an example of the display unit 2 will be described. Specifically, the display unit 2 consists of a computer and a display device that implement a screen drawing program, but the computer used here may be different from the first life modeling unit 110, the second life modeling unit 120, and the failure probability evaluation unit 130 described above.

[0099] Here, Figure 11 shows an example of a graphical user interface (GUI) suitable for the display unit 2 in Embodiment 1. In Figure 11, a component selection unit 26 is provided which displays a schematic diagram of the wind turbine 1, and the user selects any component from the schematic diagram using a pointer 27. Furthermore, when the user sets the period for calculating the failure probability in the failure probability calculation period setting unit 28, the current and after the specified period failure probability P for each machine for the component selected by the pointer 27 is displayed as a graph in the fleet failure probability display unit 29. This makes it easy for the user to check which individual components (selected by the pointer 27) have a higher failure risk. At the same time, the date and time when the explanatory variable formula D(x) was updated in the explanatory variable formula generation unit 111 and the date and time when the parameters of the failure probability function F2 were updated are displayed in the life model display unit 32. Here, the date and time of the update is displayed in the explanatory variable formula update date display unit 30. The date and time when the parameters were updated is displayed in the failure probability parameter update date display unit 31. Therefore, users can understand the update status of the model, and it becomes easier to decide when to update the explanatory variable equations based on the usage status of the computer on which the first lifetime modeling unit 110 is implemented and the update status of the model.

[0100] Furthermore, the present invention does not limit the platform for the program that implements the GUI. However, it is preferable that it be implemented as a web application that runs on a web browser and installed on the same computer as each part of the failure probability evaluation system 100 (first life modeling unit 110, second life modeling unit 120, failure probability evaluation unit 130). If it is possible to connect from a computer terminal used by a user (user terminal) via a communication means such as a network, the computing power and prerequisite software required of the user terminal can be kept to a minimum. This configuration is particularly effective when multiple users access the system simultaneously. [Examples]

[0101] Next, Example 2 will be described using Figure 12. Figure 12 is a system configuration diagram when Example 2 is applied to multiple wind turbines 1. In the following, the same configuration as in Example 1 will not be explained. The differences from the previous example are as follows. - The failure event data D1 is divided into failure event data D11, which is used when the first life modeling unit generates the life model 113, and failure event data D12, which is obtained thereafter. - The time-series operational data D2 is divided into time-series operational data D21, which is used when generating the lifetime model 113 by the first lifetime modeling unit, and time-series operational data D22, which is obtained thereafter.

[0102] The modeling process for the first lifetime modeling unit 110 is the same as in Example 1. In contrast, the second lifetime modeling unit 120 differs from Example 1 in that it updates the parameters of the failure probability function F2 using only the failure event data D12 and the time series D22. Thus, in this embodiment, the use of failure event data D11 and time series operation data D21, which are used when generating the lifetime model 113 by the first lifetime modeling unit, is omitted when updating the parameters. This reduces the memory capacity of the computer implementing the second lifetime modeling unit 120 and the storage capacity of the database, making it possible to provide a lower-cost failure probability evaluation system than in Example 1.

[0103] The following describes a method for updating the parameters of the failure probability function F2 using failure event data D12 and time-series operation data D22 in a limited manner. First, in order to calculate the explanatory variable 124 at the present time without using the time-series operation data D21, once the modeling in the first lifetime modeling unit 110 is completed, the explanatory variable for each individual is recorded in the lifetime model storage unit 121. In addition, the explanatory variable calculation unit 122 obtains the explanatory variable equation D(x) and the explanatory variable from the lifetime model storage unit 121 each time time time-series operation data D22 is input.

[0104] Here, the explanatory variable generated in the most recent step is added to the explanatory variable from the previous step using the most recent step data obtained from the time-series operational data D22 and the explanatory variable equation D(x). This calculates the current explanatory variable 124 and records it in the lifetime model storage unit 121. By repeating this process, it becomes possible to calculate the current explanatory variable each time new time-series operational data D22 is received.

[0105] Next, we will explain how to update the parameters of the failure probability function F2 using the failure event data D12 and time-series operation data D22 in a limited manner. If the failure event data D11 used when generating the lifetime model 113 is unavailable, the survival time analysis data D4, which includes all individuals, as shown in Figure 9(b), cannot be created. Therefore, maximum likelihood estimation cannot be applied when identifying the failure probability function. In this case, it is useful to estimate the probability density distribution π(θ|D) of the desired parameter from the product of the likelihood function L(D|θ) and the prior probability distribution π(θ) of the parameter, using the basic Bayesian statistical formula shown in equation (10).

[0106]

number

[0107] Here, D is the explanatory variable at the time a failure occurs or the explanatory variable at the present time. Also, θ is the parameter that determines the failure probability density function that the data D follows, and in the case of the Weibull distribution, it is the shape parameter α and the scale parameter β. In the case of the Weibull distribution, the likelihood function L(D|θ) is given by the following (Equation 11).

[0108]

number

[0109] The second lifetime modeling unit 120 utilizes the parameters identified by the failure probability function F1 (α1 and β1 in the case of a Weibull distribution) stored in the lifetime model storage unit 121, and also stores the prior probability distribution 126 of the parameters in the lifetime model storage unit 121. The failure probability function identification unit 123 estimates the probability density distribution π(θ|D) using the prior probability distribution 126 of the parameters and the likelihood function L(D|θ), and identifies the failure probability function F2. Furthermore, by using the probability density distribution π(θ|D) obtained at this time as the prior probability distribution when new failure event data is received, it becomes possible to update the parameters of the failure probability function F2 without using the failure event data D11.

[0110] This concludes the description of each embodiment, but the present invention is not limited to these. For example, it can be applied to plants other than the wind turbine generator 1 as the target mechanical system. It can also be applied to various malfunctions other than failure, such as deterioration. [Explanation of Symbols]

[0111] 100: Failure probability evaluation system 110: First Life Modeling Section 120: Second Life Modeling Section 130: Failure Probability Evaluation Department 200: A computer on which a second lifetime modeling unit and a failure probability calculation unit are implemented. 210: A computer on which a second lifetime modeling unit and a failure probability calculation unit are implemented. 300: Computer on which the first lifetime modeling unit is implemented 1: Wind turbine 2: Display section 3: Parts Inventory Management System 4: Operational Planning System 5: Operational Plan 5 6: Network 111: Explanatory variable formula generation unit 112: Failure probability function identification unit 113: Lifetime Model 121: Lifetime Model Memory Unit 122: Explanatory Variable Calculation Unit 123: Failure probability function identification unit 124: Explanatory variables at present 125: Current data for calculating explanatory variables 126: Prior probability distribution of parameters of failure probability function 131: Operation Status Prediction Unit 132: Failure Probability Calculation Unit 133: Failure probability 134: Future explanatory variables 22: Failure Flag 23: Survival Flag 26: Parts Selection Section 27: Pointer 28: Failure probability calculation period setting unit 29: Fleet failure probability display unit 30: Display section for explanatory variable formula update date 31: Display section for the update date of failure probability parameters D1: Failure event data D2: Time-series operational data D3: Data for survival analysis D4: Data for survival analysis D11: Failure event data used in the first life modeling unit D12: Failure event data obtained after modeling in the first life modeling section. D21: Time-series operational data used in the first life modeling unit. D22: Time-series operational data obtained after modeling in the first life modeling section. D31: Failure Data D32: Survival Data DB1: Failure Event Database DB2: Time-series operating database D(x): Explanatory variable formula F1: Failure probability function identified by the first lifetime modeling unit 110 F2: Failure probability function identified by the second lifetime modeling unit 120 M: Maintenance person V: Variation in the failure probability function 1001: Processing device 1002: Memory 1003: Network Interface 1004: Secondary storage device 1005: Failure probability evaluation program

Claims

1. In a failure probability evaluation system for evaluating the failure probability of components constituting a mechanical system, A failure history database that stores past failure history data for the aforementioned mechanical system, A time-series operation database for storing time-series operation data representing the operating status of the aforementioned machine system, A first life modeling unit estimates a failure probability function constituting the life model of the machine system at a predetermined estimated frequency using the aforementioned time-series operation data and failure history data. A failure probability evaluation system having a second lifetime modeling unit that updates the parameters of the failure probability function at a higher update frequency than the estimation frequency of the failure probability function by the first lifetime modeling unit, using time-series operation data including additional time-series operation data and failure history data including additional failure history data, after the estimation of the failure probability function.

2. In a failure probability evaluation system for evaluating the failure probability of components constituting a mechanical system, A failure history database that stores past failure history data for the aforementioned mechanical system, A time-series operation database for storing time-series operation data representing the operating status of the aforementioned machine system, A first life modeling unit estimates the failure probability function that constitutes the life model of the machine system using the aforementioned time-series operating data and failure history data, A failure probability evaluation system having a second lifetime modeling unit that updates the parameters of the failure probability function using the additional time-series operation data and the additional failure history data when information contributing to prediction accuracy is added to the time-series operation data and the additional failure history data after the estimation of the failure probability function.

3. In the failure probability evaluation system according to claim 1 or 2, The second life modeling unit is a failure probability evaluation system that updates the parameters using the additional time-series operation data and the additional failure history data itself.

4. In the failure probability evaluation system according to claim 1 or 2, The first life modeling unit is a failure probability evaluation system comprising: an explanatory variable formula generation unit that generates an explanatory variable formula for the failure probability function from the time-series operating data; and a failure probability function identification unit that identifies the failure probability function from the explanatory variable formula and the failure history data.

5. In the failure probability evaluation system according to claim 4, The explanatory variable formula generation unit is a failure probability evaluation system that generates an explanatory variable formula that minimizes the variation in the failure probability function identified by the failure probability function identification unit.

6. In the failure probability evaluation system according to claim 5, A failure probability evaluation system in which the variability of the failure probability function is represented by the coefficient of variation of the failure probability function.

7. In a failure probability evaluation method that evaluates the failure probability of components constituting a mechanical system using a failure probability evaluation system, The memory unit stores a failure history database for storing past failure history data for the machine system, and a time-series operation database for storing time-series operation data representing the operating status of the machine system. The first life modeling unit uses the time-series operation data and failure history data to estimate the failure probability function that constitutes the life model of the machine system at a predetermined estimated frequency. A failure probability evaluation method comprising a second lifetime modeling unit updating the parameters of the failure probability function at a higher update frequency than the estimation frequency of the failure probability function by the first lifetime modeling unit, using time-series operation data including additional time-series operation data and failure history data including additional failure history data after the estimation of the failure probability function.

8. In a failure probability evaluation method that evaluates the failure probability of components constituting a mechanical system using a failure probability evaluation system, The memory unit stores a failure history database for storing past failure history data for the machine system, and a time-series operation database for storing time-series operation data representing the operating status of the machine system. The first life modeling unit estimates the failure probability function that constitutes the life model of the machine system using the time-series operating data and failure history data. A failure probability evaluation method in which, when information contributing to prediction accuracy is added to time-series operation data including additional time-series operation data and failure history data including additional failure history data after estimation of the failure probability function by a second lifetime modeling unit, the parameters of the failure probability function are updated using the additional time-series operation data and the additional failure history data.

9. In the failure probability evaluation method according to claim 7 or 8, A failure probability evaluation method that updates the parameters using the additional time-series operation data and the additional failure history data itself, with respect to the second life modeling unit.

10. In the failure probability evaluation method according to claim 7 or 8, The first lifetime modeling unit includes an explanatory variable equation generation unit and a failure probability function identification unit. The explanatory variable equation generation unit generates the explanatory variable equation for the failure probability function from the time-series operating data. A failure probability evaluation method comprising identifying the failure probability function from the explanatory variable formula and the failure history data using the failure probability function identification unit.

11. In the failure probability evaluation method according to claim 10, A failure probability evaluation method wherein the explanatory variable formula generation unit generates an explanatory variable formula that minimizes the variation in the failure probability function identified by the failure probability function identification unit.

12. In the failure probability evaluation method according to claim 11, The variability of the failure probability function is expressed by the coefficient of variation of the failure probability function, which is used to evaluate the probability of failure.

Images