KR20110043277A - A method of failure prediction at power plant using bayesian statics - Google Patents

Abstract

PURPOSE: A method of failure prediction at a power plant using Bayesian statistics is provided to efficiently maintain and manage a power facility by determining the efficient repair and maintenance period based on Bayesian statistics in consideration of a TVFR(Time-Varying Failure Rate) and the maintenance frequency. CONSTITUTION: The information on the failure history of each component is inputted to a database(201,202). Based on the information on the failure history, the years of failure and the number of failure times depending on each failure reason are analyzed(203). Based on the analyzed result, the rank and reverse rank are calculated(204). Through the calculated reverse rank, an adjusted rank and a media rank are calculated(205). In consideration of the calculated media rank and the failure years, regression analysis is performed.

Description

A Method of Failure Prediction at Power Plant Using Bayesian Statics

The present invention design and manufacture an algorithm and a development program for predicting the failure of each component of the power generation plant by using the Bayesian statistical method based on the data stored in the database of the failure history for each component of the power plant; The algorithm for predicting the failure of each component of the power plant is applied to the present invention the Weibull function, which is one of the Bayesian statistical techniques, for the equipment failure prediction. By predicting and diagnosing the service life, it is possible to prevent the economic loss caused by replacing the parts periodically and unnecessary in advance, thereby preventing the cost increase due to the purchase of parts and the economic loss caused by stopping the power generation, and improving the reliability of the power generation equipment. The present invention relates to a power plant failure prediction method using Bayesian statistical technique.

In general mechanical products, reliability and stability can be secured through functional design, strength design, and optimization design methods. However, in the case of products produced in small quantities, many products do not have enough data for statistical processing. There is a difficult problem to evaluate.

In the case of general electronic products, it is possible to predict by statistical technique with single product type and mass production. However, since both reliability and stability should be considered at the same time for mechanical parts, risk management criteria (Ri Or based on criteria such as Based Design) or Risk Based Maintenance.

Until now, maintenance of power generation facilities has been implemented by disassembling, replacing and maintaining the parts after a certain period of time according to TBM (Time Based Maintenance) .However, the environment around the rapidly changing power generation industry is how to maintain TBM. There is a problem that cannot guarantee the maximum efficiency and cost-benefit.

In the prior art related to the present invention, Korean Patent Publication No. 10-2004-0015011 discloses a technical configuration of Bayesian statistical inference used to achieve the network vascular health analysis evaluation system and method, and the Republic of Korea 10-2006-0044772 discloses a technical configuration of a Bayesian network using a table for tree learning, but the technical concept and configuration of Bayesian inference and Bayesian network used in each invention are described for each component of the power plant. There is a significant difference from statistical techniques for predicting failure.

The problem to be solved by the present invention is to maintain the power generation equipment according to the risk-based maintenance, such as maintenance of the power generation equipment of the TBM (Time Based Maintenance) method, the failure prediction of the power generation equipment components For this purpose, the failure history of each part is made into a database and the Bayesian statistical method, which is one of the statistical techniques, is used to predict the failure probability of the components of the power generation facilities to reduce the economic loss due to frequent replacement of parts. This is to reduce the economic loss of stopping the development.

Another problem to be solved by the present invention is to determine the effective maintenance interval in consideration of the Time-Varying Failure Rate (TVFR) of maintenance based on Bayesian statistical techniques and the number of maintenance, risk criteria By setting standards such as Risk Based Maintenance, it is possible to efficiently maintain power generation facilities to minimize economic losses and improve reliability.

The problem solving means of the present invention inputs the capacity, efficiency, operation information and operation time, failure and maintenance history of the power generation facility in the database, and prepare the probability distribution of parts or equipment, based on the wave distribution function Predict failure by the step, and analyze and analyze the failure and maintenance history of each part of the power generation equipment, and store the information (data) of the failure history for each part in the database, and use the data stored in the database to generate the power generation equipment parts. After analyzing each cycle, failure year, and number of failures of each cause, and determining the rank and reverse rank for each part of the power generation equipment by using the analysis result, REVERSE RANK is used to determine the ADJUSTED RANK and the MEDIAN RANK.The failure year and the MEDIAN Bayesian consisting of a step of calculating the wavelet distribution function and performing the failure prediction by performing the regression analysis using RANK) and determining the α, β values of the two factors of the wavelet distribution function. To implement a method for predicting failure of a power plant using statistical techniques.

Another problem solving means of the present invention is to determine the effective maintenance interval in consideration of the time-varying failure rate (TVFR) of maintenance and the number of maintenance based on Bayesian statistical techniques, risk-based maintenance It is to implement power generation equipment failure prediction method that can minimize the economic loss and improve the reliability by setting up the criteria such as Risk Based Maintenance.

The present invention is to maintain the maintenance of the power generation facilities on the basis of the risk-based maintenance management, such as maintenance of the TBM-type power generation facilities, database the failure history of each component to predict the failure of the components of the power generation facilities and statistics Bayesian statistical method, which is one of the techniques, predicts the failure probability of the components of power generation facilities, thereby reducing the economic loss due to frequent replacement of components and reducing the economic loss due to stopping the power generation for parts replacement.

Another effect of the present invention is to determine the effective maintenance cycle in consideration of the time-varying failure rate of maintenance and the number of maintenance based on Bayesian statistical techniques, and to set the criteria such as risk-based maintenance management, etc. To minimize economic losses and increase reliability.

Hereinafter, the present invention will be described in detail. The present invention through the step of inputting and storing the capacity, efficiency, operation information and operation time, failure and maintenance history information of the power plant in the database, using the data stored in the database failure cycle for each component of the power plant In this case, the system analyzes the failure year and the number of failures.

Analyze each failure cycle, failure year and number of failures of each power plant component by using the data stored in the database. Rank and reverse rank for each power plant component using the analyzed result Calculate the (REVERSE RANK).

Calculate the ADJUSTED RANK and the MEDIAN RANK using the REVERSE RANK, and perform the regression analysis using the failure year and the MEDIAN RANK. ) The fault prediction method of power generation facilities using Bayesian statistical technique is implemented through the step of calculating the value of α, β, which is two factors of the distribution function, and the fault prediction by calculating the WEIBULL distribution function.

The present invention can be applied not only to domestic power generation facilities, but also to various industrial facilities. Currently, about 50 domestic thermal power plants and industrial facilities and finished products are predicted for the probability of failure and defects throughout the country, so the demand and application of this technology are very wide. There are active demands and researches on power plant failure prediction, mainly from domestic and overseas manufacturers and research institutes.

Hereinafter, the configuration and operation of the embodiments of the present invention with reference to the accompanying drawings, the configuration and operation of the present invention shown and described in the drawings will be described by at least one embodiment, whereby the present invention described above The technical idea and its core composition and operation are not limited.

Look at the drawings to facilitate understanding of the present invention. FIG. 1 shows a method of constructing a failure DB, a method of calculating a probability of failure using this method, and a method of calculating a failure prediction using a wavelet distribution function, and FIG. 2 shows a failure of each component of a power plant. Each step required to perform the algorithm for prediction is shown. 3 shows a flowchart for performing a basic algorithm for fault prediction. 4 is a screen capture of one of the configurations of a database designed and manufactured according to the present invention, and FIG. 5 illustrates a method for obtaining a first-order function for deriving a wave factor through regression analysis. A specific embodiment according to the present invention will be described.

<Example>

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A preferred embodiment of the present invention will now be described with reference to the drawings. FIG. 4 is a view of capturing a screen of a database designed and manufactured according to the present invention, and shows a screen when a turbine is selected among power plant components. When turbine is selected, the left screen is divided into high pressure (HP), medium pressure (IP) and low pressure (LP), and the parts according to each pressure are listed, and on the right, the part code for each part listed on the left, It is configured to input part name, year of completion, year of failure, type of failure, cause of failure and corrective action in the database. The database may correspond to all components employed in a power generation facility, such as a boiler and a generator, in addition to a turbine.

Figure 2 shows a flow chart for each step required to perform the algorithm for failure prediction for each component of the power plant. Looking at each stage, it analyzes and analyzes the failure and maintenance history of each part of the power generation equipment, finds out the information on the failure history of each part, inputs it into the database, and develops based on the information about each part entered in the database. It analyzes each failure year and number of failures by cause of equipment parts.

Calculating the rank (RANK) and the reverse rank (REVERSE RANK) using the analyzed result, and calculates the adjusted rank (ADJUSTED RANK) and the median rank (MEDIAN RANK) using the determined reverse rank (REVERSE RANK) Go through the steps.

The regression analysis is performed using the determined median rank and failure year, and the step 206 of determining the scale parameter (α) and the shape parameter (β), which are two factors of the wavelet distribution function, is determined. Step 207 is performed by substituting the parameter α and the shape parameter β into the wavelet distribution function to perform failure prediction for each component.

The regression analysis is a method of predicting a dependent variable when a given independent variable is obtained by obtaining a linear relation, which is a mathematical model of a causal relationship between the independent variable and the dependent variable, for the continuous variables observed in statistics. It is also an analytical method that measures fitness to determine how well this mathematical model explains.

In the present invention, when the failure history and information of each component of the power generation equipment is inputted to the database, the power generation equipment is divided into large parts such as a turbine, a generator, and a boiler as shown in FIG. 4 (401), and the target equipment (402). After the selection, the part code 404 and the part name 405 are inputted, and the year of completion (406), year of occurrence of failure (407), type of failure (408), and cause of failure (409) of the power generation equipment employing the corresponding parts are entered. ) And measures (410) in sequence, but the correct input increases the reliability of failure prediction.

The failure prediction method according to the present invention determines an effective maintenance interval in consideration of the time-varying failure rate (TVFR) of maintenance based on Bayesian statistical technique and the number of maintenance, and risk-based maintenance management. It is designed to minimize the economic loss and improve the reliability of power generation facilities by setting the standards such as (Risk Based Maintenance).

The algorithm for predicting failure used in the present invention uses a Weibull distribution function {f (t), equation (1)}, which is one of the Bayesian statistical techniques, for predicting device failure.

Where α is the scale parameter and β is the shape parameter.

Using the failure database of each component, the scale parameter α and the shape parameter β can be obtained through the following steps.

a. (Failure Year-Completion Year) is calculated, and the number of failures occurring in the same year is calculated.

b. Rank is given by using the number of failures (n) calculated for each year. The rank is 1 when the number of failures n occurred per year is n = 1, and n / 2 when the number of failures n is greater than one.

c. The reverse rank is calculated using the rank.

  The reverse rank is an inverse of rank.

d. The adjusted rank is calculated by using reverse rank. More specifically, the adjusted rank divides the reverse rank by the previous adjustment rank, adds 1 to the number of failures, and then Calculate by dividing by the reverse rank plus one.

Expressed as a formula, the adjusted rank (Adjusted Rank) = [(Reverse Rank) (Previous Adjusted Rank) + (n + 1)] / (Reverse Rank + 1)

to be.

The initial value of the previous adjusted rank is zero.

e. The median rank is calculated using the adjusted rank, and more specifically, the median rank is calculated by dividing the value obtained by subtracting 0.3 from the control rank by the value of 0.4 added to the reverse rank. In formulas,

      Median Rank = [(Adjusted Rank-0.3)] / [Reverse Rank + 0.4]

to be.

f. Logarithm (ln) is calculated by subtracting the year of completion from the year of failure.

In formulas,

x = ln (failure year-completion year)

to be.

g. After subtracting the median rank from 1, take the reciprocal, take the logarithm of the value obtained by taking the reciprocal, and take the log to the value obtained by taking the log and then calculate it.

In formulas,

y = ln · ln (1 / (1-Median Rank))

to be.

A plurality of values calculated in step f are represented by the x-axis, and a plurality of values obtained by calculating corresponding to each x value in step g are represented by the y-axis as shown in FIG.

Through the regression analysis of the graph, it is possible to obtain a linear equation of y = αx + β (a straight line in FIG. 5), where the slope of the linear equation becomes the scale parameter α.

In addition, the shape parameter β, which is the second characteristic factor of the Weibull distribution function, can be obtained by calculating from Equation (2).

In Equation (2), the shape parameter β is obtained by calculating the exponential function by subtracting the average value of the plurality of y values from the average value of the plurality of y values divided by the α value.

Since the present invention uses an algorithm designed to perform failure prediction for each use year by using the Weibull distribution function, which is a statistical method, it is impossible to accurately predict the failure of each power generation component until now. Although there was a great economic loss from unilateral replacement of components at a certain period, failure prediction can be performed through this method, which prevents unnecessary replacement of components.

Through on-site application of the present invention can be used continuously to predict failure of each component of the power plant, the failure prediction method of the power plant components using statistics can secure reliable data through the continuous failure database update and continuously It can greatly contribute to the economic operation of power generation facilities.

Industrial Applicability

The present invention is to maintain the power generation facilities on the basis of the risk-based maintenance management, such as maintenance of the TBM-type power generation facilities, the database of failure history of each part to predict the failure of the parts of the power generation facility statistics Bayesian statistical method, which is one of the techniques, provides a method for predicting the failure of power plant equipment using Bayesian statistical technique that predicts the probability of failure of the components of power plant. Industrial applicability is very high as it can reduce the economic loss.

1 is a flowchart for calculating a failure prediction using a method of building a failure database and a wavelet distribution function.

Figure 2: A series of step-by-step flowcharts for performing fault prediction calculations

Figure 3: Basic algorithm for fault prediction

4: Screen capture showing an example of the configuration diagram of a database

Figure 5: Derivation of wavelet factor through regression analysis

Claims (6)

In the failure prediction method of power generation equipment using Bayesian statistical technique, Investigating and analyzing failure and maintenance history of each component of power generation equipment, finding information on failure history for each component, and inputting the same into a database; Analyzing each failure year and the number of failures for each failure cause of power generation equipment parts based on the information on the failure history of each component inputted in the database; Calculating a rank and an inverse rank using the analyzed result of the failure year and the number of failures; Calculating an adjustment rank and a median rank based on the calculated inverse rank; Performing regression analysis using the calculated median rank and the fixed year to calculate the values of the scale parameter α and the shape parameter β, which are two factors of the wavelet distribution function; And A method for predicting a failure of a power generation facility, comprising: performing a failure prediction for each part by substituting the calculated scale parameter and the shape parameter into a wavelet distribution function. The method according to claim 1, Breakdown and maintenance history for each part entered into the database is a generation equipment failure prediction method, characterized in that consisting of the year of completion of the power generation equipment employing the parts, the year of occurrence of the failure, the type of failure, the cause of the failure and measures. The method according to claim 1, The scale parameter (α) is obtained by subtracting the log year (ln) from the failure year minus the completion year and obtaining the x-axis value. Subtract the central rank from 1, take the inverse, take the logarithm of the value obtained by taking the inverse, take the logarithm of the value obtained by taking the logarithm, and find the y value, And calculating the slope of the first function obtained from the plurality of x values and the plurality of y values corresponding to x. The method according to claim 3, The shape parameter β is obtained by dividing an average value of y by an α value and taking an exponential function on a value obtained by subtracting an average value of x from this value. The method according to any one of claims 1 to 4, The rank is 1 when the number of failures (n) occurred in a year is n = 1, and when the number of failures (n) is greater than 1, it is calculated as n / 2. The method according to any one of claims 1 to 4, The control rank is calculated by dividing the reverse rank by the previous control rank, add 1 to the number of failures to the divided value, and calculates this value by dividing the value by adding 1 to the reverse rank.

Images