KR101777387B1 - Apparatus for analyzing data based on failure period - Google Patents

Abstract

Disclosed is a device for analyzing data based on a failure occurrence period. The data related to a failure occurrence period is obtained from data stored in a subsequent logistical support database, and visualized and grouped by performing a failure occurrence period and correspondence analysis on the obtained data so as to accurately obtain data having the highest correlation with the failure occurrence period. The device for analyzing data comprises a data search unit, a correspondence analysis unit, and a pattern analysis unit.

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention [0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a data analysis apparatus, and more particularly, to a data analysis apparatus based on a fault occurrence period for improving reliability of a weapon system.

Due to the development of information technology, data in various fields are increasing rapidly. As a result, recent manufacturing industry has been interested in big data analysis as a technology that leads change of future manufacturing industry. Especially, improvement of process process using big data analysis has achieved excellent results in manufacturing quality improvement. For example, recent studies on quality improvement using Big Data have been carried out in order to solve the problems such as the accident of the vehicle suddenly. In response to these trends, various studies are being carried out in order to introduce big data analysis in the field of defense.

In particular, the logistics sector is recognized as one of the areas that can lead to breakthroughs through big data analysis. Currently, vast information is being acquired through various logistics information systems. However, various information is being used without being utilized effectively.

Although the flaws in the weapon systems can lead to fatal consequences such as loss of life and cost, the current weapon systems are carried out by different executing agencies, each stage from development and operation / maintenance to separate data collection and utilization, Since the development of the weapon system, the analysis of the cause of defects is insufficient, which is an obstacle to quality improvement.

The weapon system should increase its utilization rate through quality improvement, maintain military preparedness for combat, and demonstrate its performance in case of emergency. One of the key factors in combat readiness is the creation of high-quality weapons and the ability to perform the required functions without a fault during the intended period. For this, securing the reliability of the weapon system development stage should be considered first. Decreased reliability can lead to breakdown, short life span and increased maintenance costs, and it is not possible to maintain combat readiness efficiently.

Nevertheless, the current weapon system estimates the reliability by calculating the failure rate of the parts unit only at the development stage. This is because the weapon system is difficult to perform a lot of tests due to high unit cost, long development period, and limited space. In addition, there is a problem that it is difficult to collect a large amount of data with a small amount of accumulated data because the development process of the weapon system is progressively improved through the development process. In addition, it is difficult to obtain data at the development stage due to the importance of security. Therefore, in the present test evaluation, simple quantitative measure such as accuracy rate is the standard of evaluation. In other words, there is a limit to carry out research at the data level of the weapon system development stage.

Korean Registered Patent No. 10-1705347 (Registered on Feb. 23, 2013)

An object of the present invention is to extract reliability deteriorating factors by extracting variables affecting reliability at the time of development by using stored data in a subsequent logistical support database and grasping a correlation pattern between variables based on a failure occurrence period, And to provide a data analysis apparatus based on a fault occurrence period that can eliminate reliability deterioration factors in weapon development.

According to an aspect of the present invention, there is provided an apparatus for analyzing data on the basis of a fault occurrence period, comprising: a database for searching a subsequent logistical support database storing field operation and maintenance information, Setting a failure occurrence period as an objective variable in the acquired data, and setting a plurality of analysis variables from the remaining data, and if the objective variable and the plurality of the plurality of analysis variables are present, A data search unit for converting a continuous variable into a categorical variable; Analyzing the correlation between the target variable and each of the plurality of analysis variables by performing a corresponding analysis on each of the objective variable and the plurality of analysis variables to derive a pattern of the obtained data, A corresponding analysis unit; And grouping the categories of the analytical variables corresponding to each category of the objective variable from the visualized analysis results and calculating a relative frequency ratio of each category of the analytical variables grouped, A pattern analysis unit for extracting the highest group; .

Wherein the data searching unit searches for and obtains repair request information among information stored in the subsequent log support database, calculates the fault occurrence period from the repair request information, obtains the fault occurrence period as the target variable, A data extracting unit for selecting a plurality of the analysis variables based on the number of the missing values and the number of levels of categories for the data; And transforming the continuous variable into a categorical variable when the objective variable and the plurality of the analysis variables exist, and if the redundant data and the data in which the number of levels of each category is less than a predetermined reference number are removed by noise, part; And a control unit.

Wherein the correspondence analysis unit is configured to express the objective variable and the plurality of analysis variables as a categorical variable in the form of a crosstabulation table and to calculate a correlation coefficient indicating a dependency information of each of the plurality of analysis variables for the objective variable Calculating section; An inertial dimension analyzer for determining the number of dimensions for visually expressing the corresponding analysis result and calculating coordinates of each of the plurality of analysis variables on a space according to the determined number of dimensions; And a visualization unit for displaying each category of the target variable in the determined dimension space in a vector form and displaying a category of each of the plurality of analysis variables in the calculated coordinates to visualize the category; And a control unit.

Wherein the correlation coefficient calculation unit calculates the correlation coefficient using the Pearson residual obtained by performing a chi-square test on the target variable represented by the cross-tabular form and the categories of the plurality of analysis variables, And if the correlation coefficient calculated for the variable is less than a preset reference value, the corresponding analysis variable is excluded from the corresponding analysis target.

And the inertia dimension analyzer obtains the coordinates of the categories according to the singular value decomposition technique for the Pearson residual.

Wherein the correspondence analysis unit calculates a contribution indicating an inertia ratio of each of the categories for each axis in the determined dimension space and excludes an analysis variable having the calculated contribution less than the average contribution from the corresponding analysis target; And an expressiveness analyzer for calculating the expressive power of each of the categories for each axis in the determined dimension space and excluding analytic variables whose calculated expressive power is less than a predetermined reference expressive power from the corresponding analysis target; And further comprising:

Wherein the pattern analyzing unit classifies the category of each of the plurality of analysis variables displayed in the coordinates calculated in the determined dimension space into a group corresponding to a category of the object variable closest to the category of the objective variable indicated in a vector form, ; And extracting a group having the highest degree of correlation with respect to the objective variable by calculating a relative frequency ratio of each category of the analytical variables grouped and classifying the category of the objective variable and the category of the analysis variable corresponding to the extracted group A pattern extracting unit for sorting by categories for analysis; And a control unit.

Therefore, the fault analysis period based data analysis apparatus of the present invention obtains data related to the fault occurrence period from the data stored in the subsequent log support database, performs visualization and grouping of the obtained data by performing fault occurrence period and corresponding analysis, It is possible to accurately extract high-data having the highest correlation with the fault occurrence period. Therefore, it will be possible to develop efficient weapon systems by analyzing the important factors that should be considered first in the development of the weapon system in advance.

1 shows a data analysis apparatus according to an embodiment of the present invention.
2 is a graph showing the ratio of each dimension to the total inertia.
FIGS. 3 to 5 show the results of calculating the contribution of each of the analysis variables to the objective variable.
6 shows the result of calculating the expressive power of each analysis variable for the target variable.
Figures 7 to 9 show the corresponding analysis visualization results for the objective variable and the analysis variable, respectively.
10 illustrates a method of analyzing data based on a fault occurrence period according to an embodiment of the present invention.
11 is a detailed view of the corresponding analysis step of FIG.

In order to fully understand the present invention, operational advantages of the present invention, and objects achieved by the practice of the present invention, reference should be made to the accompanying drawings and the accompanying drawings which illustrate preferred embodiments of the present invention.

BEST MODE FOR CARRYING OUT THE INVENTION Hereinafter, the present invention will be described in detail with reference to the preferred embodiments of the present invention with reference to the accompanying drawings. However, the present invention can be implemented in various different forms, and is not limited to the embodiments described. In order to clearly describe the present invention, parts that are not related to the description are omitted, and the same reference numerals in the drawings denote the same members.

Throughout the specification, when an element is referred to as "including" an element, it does not exclude other elements unless specifically stated to the contrary. The terms "part", "unit", "module", "block", and the like described in the specification mean units for processing at least one function or operation, And a combination of software.

1 shows a data analysis apparatus according to an embodiment of the present invention.

As shown in FIG. 1, the failure analysis period-based data analysis apparatus 200 of the present invention performs data analysis based on a failure occurrence period using data stored in a subsequent log support database 100, By analyzing the defective factors of the weapon system by patterning, it provides information that can improve the reliability in the development of the weapon system in the future.

Subsequent logistic support means all logistical support processes for field operation and maintenance of equipment acquired after power up. Therefore, it is possible to obtain information such as maintenance time, parts replacement history, cause of defects, etc. of defects arising from field operation at the time of subsequent logistics support. Information on the current logistics support details is acquired in the form of data and stored in the subsequent log support database 100, which is an integrated database. That is, the subsequent logistical support database 100 records activities for efficiently operating the weapon system. Therefore, subsequent logistical support data provides state information of the weapon systems after power-up, and it is possible to obtain meaningful information.

Therefore, the stored data can be used for the analysis of tasks such as updating logistics support elements, deriving design change factors, and forecasting repair accessories. That is, the data obtained through the subsequent logistics support can be utilized to improve the problems of the above-mentioned weapon system development stage.

However, most of the data stored in the subsequent logistical support database 100 has characteristics of high dimensional categorical data. Therefore, analytical devices and analytical methods for the data of these characteristics are needed.

Referring to FIG. 1, the data analysis apparatus 200 includes a data search unit 210, a corresponding analysis unit 220, and a pattern analysis unit 230.

The data searching unit 210 includes a data extracting unit 211 and a preprocessing unit 212 to obtain data to be analyzed by searching for data stored in the subsequent logistical support database 100, Select the analysis variables to perform and preprocess the data corresponding to the selected variables.

The data extracting unit 211 searches for and extracts repair request information from the subsequent log support database 100, and selects an analysis variable from the repair request information. The repair request information includes information such as date, maintenance request unit, weapon system type, cause of defect, action item and amount, and is mostly composed of categorical variables. The data extracting unit 211 may select an analysis variable based on the number of missing data and the number of categorical levels.

In the present invention, for example, the data extracting unit 211 sets a categorical variable Military, Weapon System, Cause of Failure, and TUM as analysis variables as shown in Table 1 . Here, TUM means Top, Unit, Module, and indicates which sub-part structure of the weapon system has failed.

Table 1 shows the variables selected in the repair request information collected and stored in the subsequent logistical support database 100 for the predefined period.

The data extraction unit 211 selects a failure period, which is a categorical variable, as a target variable. The failure occurrence period is a period from the manufacturing date of the relevant weapon system to the failure acceptance date, and the data analysis apparatus of the present invention is a target variable to be selected as a variable as the base data for analyzing the defects of the weapon system. The data extraction unit 211 may calculate the period from the manufacturing date to the failure acceptance date to obtain the failure occurrence period since most of the failure occurrence periods are not separately stored in the subsequent logistical support database 100. [

The preprocessing unit 212 preprocesses the data corresponding to the obtained variables so that the correspondence analysis unit 220 can easily analyze the data. The correspondence analysis unit 220 performs correspondence analysis, which is one of exploratory data analysis techniques, and the correspondence analysis is a technique of performing analysis on two categorical variables. Therefore, the preprocessing unit 212 converts the failure occurrence period, which is a continuous variable, into a categorical variable. Since the fault occurrence period has a large number of outliers, the preprocessing unit 212 can perform the equal frequency discretization to convert the fault occurrence period into the categorical variable. In the present invention, for example, it is assumed that the preprocessing unit 212 divides the failure occurrence period into three groups FP1, FP2, and FP3 in the shortest period. The first group of failures (FP1) refers to the weapon system that failed within one year from the date of manufacture. The second group of failures (FP2) refers to the weapon system that failed within two years from the first year. System, and the third failure occurrence group (FP3) are assumed to include a weapon system for products with older manufacturing dates.

In addition, the preprocessing unit 212 removes redundant data having the same reception number and removes as the noise data the data of which the number of levels of the category is equal to or less than a predetermined reference number (for example, 40) in each variable, . In the present invention, the final analysis data in which redundancy and noise are removed through the preprocessing process by the preprocessing unit 212 are transmitted to the support units M1, M2, ..., M13, the weapon systems W1, W2, ..., W35, (Other, design failure, poor operation, material failure, work defect), TUM (Top, Unit, Module).

The correspondence analysis unit 220 receives the final analysis data obtained by the data search unit 210 and performs corresponding analysis. Correspondence analysis is an analytic technique that can explain not only the correlation of two categorical variables but also the patterns inherent in the data through low dimensional visualization of the analysis results. In addition, the counterpart analysis has an advantage in that it has more advantages in identifying the correlation characteristics than other techniques for exploratory data analysis, and has a strong point of knowing in which pattern the categories of the variables are related as well as the correlation information between the variables. Therefore, the correspondence analysis can be useful to identify patterns inherent in subsequent logistical support data with large quantities of categorical data characteristics

The correspondence analysis unit 220 includes a correlation coefficient calculation unit 221, an inertia dimension analysis unit 222, a contribution analysis unit 223, an expression power analysis unit 224, and a visualization unit 225.

The correlation coefficient calculation unit 221 calculates dependency information of each of the plurality of analysis variables (support unit, weapon system, defect cause, and TUM) for the failure occurrence period, which is the target variable, in the final analysis data applied by the data search unit 210 The correlation coefficient is calculated.

The correlation coefficient calculator 221 can obtain the correlation coefficient using a chi-square test. As described above, the fault occurrence period is a state converted to a categorical variable by the preprocessing unit 212. [ Thus, in the final analysis, the relationship between each of the multiple analytical variables for the time period of the fault, the target variable, can be expressed in the form of a cross table with the relationship between two categorical variables (failure duration and one analytical variable) .

The correlation coefficient calculator 221 performs a chi-square test to determine the correlation between the two categorical variables represented by the intersecting table, and the chi-square value of each cell is calculated through a chi-square test. This is called Pearson residuals. Pearson residuals represent the influence of each cell on the hypothesis of the chi-square test, and the calculation of Pearson residuals is described below.

The relative ratio matrix P can be defined as in Equation (1) with respect to the total frequency (n) of the cross table (F) in which the number of row categories is p and the number of column categories is q.

On the other hand, if the elements of the diagonal matrix (D _r ) and (j, j) in which the elements of (i, i) are in the row marginal proportion in the cross table (F) a diagonal matrix (D _c ) is calculated as shown in Equations (2) and (3).

Pearson's chi-square test is used to verify the uniformity of the distribution of rows and columns of the cross table (F). The Pearson chi-squared statistic and the Pearson residual are calculated as shown in equations (4) and (5).

The correlation coefficient represents total dependence of row and column as total inertia, and is calculated as shown in Equation (6).

The correlation coefficient, which is the total inertia, refers to the total amount of information described by the cross-tabulation table. The square root of the total inertia is used as a correlation coefficient of two variables. If the value is larger than a predetermined reference value (for example, 0.2), it can be interpreted that there is a significant correlation between rows and columns of the intersection table.

And can calculate the Chi square distance between the categories of each row and column variable in the cross table (F) by using the profile information as a mathematical expression. For example, the chi-square distance between the i-th row and the i * -th row in the row profile in the M-dimensional space can be calculated using Equation (7).

Table 2 shows an example of the correlation coefficient of each of the plurality of analysis variables (support unit and weapon system, cause of defect, and TUM) with respect to the target period of failure.

As shown in Table 2, three of the four analytical variables (support units and weapon system, cause of defects) have a significant correlation with the period of failure, while TUM shows that there is no significant correlation have. Therefore, additional analysis of the TUM need not be performed.

On the other hand, when the correlation coefficient calculation unit 221 calculates the correlation coefficient according to Equation (6), the inertia dimension analysis unit 222 determines the number of dimensions for expressing the corresponding analysis result, and determines the coordinates of each category . That is, the number of axes expressing the pattern of association is selected, and the coordinates of each category are calculated in each determined dimension.

The correspondence analysis has the advantage that it can be expressed as a low dimensional space graph using the distance information between the row variable and the column variable. However, for this purpose, it is necessary to calculate the coordinates so that the categories of the intersection table can be expressed as points on the low-dimensional space, and it is necessary to determine the number of dimensions for expressing the result of the first correspondence analysis before calculating the coordinates. In this case, as you select many dimensions, the variability of the explained data increases, but it can complicate the interpretation. In the present invention, it is assumed that the inertia dimension analyzer 222 determines the number of dimensions for expressing the corresponding analysis result as two.

Each dimension can be explained by inertia, the squared singular value is principal inertia or eigenvalue, and the sum of principal component inertia is expressed as total inertia. Depending on the principal component inertia ratio of each dimension for total inertia, it can be shown how each dimension describes the data.

When the number of dimensions is determined, the inertia dimension analyzer 222 calculates coordinates for each category of the cross table. Coordinates for the respective categories can be obtained as in Equation (8) by applying Singular Value Decomposition for the Pearson residual (R) obtained in Equation (5).

Then, the main coordinates of the rows and columns are calculated as shown in Equations (9) and (10), and the standard coordinates of the rows and columns can be calculated according to Equations (11) and (12).

Where D _λ is a diagonal matrix of singular values (Δ) sorted in descending order and can be obtained from eigenvalues (λ = D _λ ² ).

2 is a graph showing the ratio of each dimension to the total inertia.

In FIG. 2, (a) to (c) are graphs showing the correlation coefficients calculated by the correlation calculation unit 221, and the three analysis variables (the support unit and the weapon system , Cause of defect). As described above, in the present invention, since it is decided to use two dimensions, the horizontal axis represents one dimension (Dim1) and two dimensions (Dim2), and the vertical axis represents inertia of each dimension. In the bar graph, numerical values of the inertia ratios of the respective dimensions (Dim1, Dim2) are shown, and the sum of the inertia ratios for each of (a) to (c) is 100%. In other words, it can be seen that the total inertia consists of one-dimensional and two-dimensional, respectively. Therefore, the results of the corresponding analysis can be sufficiently expressed by only two dimensions.

On the other hand, as described above, the correspondence analysis can be visualized by expressing the graph of the low dimensional space using the distance information between the row variable and the column variable. And visualization is used to extract, summarize, and explain various meaningful analysis results. Therefore, the visualization is performed to express the row category and the column category of the cross table in the coordinate space to help understand the corresponding analysis result.

 A matrix diagram (Biplot) that visualizes the rows and columns of the corresponding analysis is a symmetric plot and an asymmetric plot. The symmetric figure shows the categories visually through Principal coordinates. And the asymmetric figure is a figure representing the sectional information of the column (or row) represented by the main coordinates in the row (or column) vector space represented by the standard coordinates. In the symmetric figure, the information of the inter-row distance and the hot distance can be interpreted through the symmetric drawing, but there is a disadvantage that the distance between the row and the column can not be analyzed. On the other hand, since the asymmetric picture can visually analyze the row-column relationship, the fault-based data analysis apparatus of the present invention is configured to analyze the correspondence analysis result using the asymmetric picture.

However, it should be noted that the analysis of the corresponding analysis is that the distances of the categories reflect relative information. If the two categories represented by the main coordinates have the same direction as the direction of the vector, then the characteristics of that vector appear relatively better in the category farther from the origin. Also, the correspondence analysis results in loss of information because high-dimensional information is expressed in a low-dimensional space. For example, a 9 (n-1) dimension is required to completely represent the information for a 10 X 10 crossing table. Therefore, when analyzing the correspondence analysis, it is possible to make meaningful interpretation by using the information of quality of representation and contribution information.

The contribution analysis unit 223 and the expression power analysis unit 224 calculate the contribution and the expressive power of the categories described for each axis.

The contribution analyzer 223 analyzes the contributions indicating the inertia ratios of the categories described for each axis. For example, a category with a high contribution to a one-dimensional axis can be considered to have a high explanatory power on a one-dimensional axis. Particularly, the contribution can be useful when there are many categories. Therefore, we can remove attributes with low contribution (for example, categories below average contribution) and interpret the attributes of the coordinate space using only high explanatory categories.

The contribution analyzer 223 may calculate the row contribution of the cross table F according to Equation (13) and calculate the thermal contribution according to 14. [

FIGS. 3 to 5 show the results of calculating the contribution of each of the analysis variables to the objective variable.

FIG. 3 shows the contribution of each dimension of the failure occurrence period calculated through the analysis of the failure occurrence period and the support unit. FIG. 4 is a graph showing the contribution of each dimension of the failure occurrence period calculated through the correspondence analysis between the failure occurrence period and the weapon system. And FIG. 5 shows the contribution of each dimension of the failure occurrence period calculated through the correspondence analysis between the failure occurrence period and the failure cause. And the dashed lines in Figs. 3 to 5 indicate the average contribution.

It can be seen that the category with high explanatory power differs for each dimension, but all the categories have more than average contribution in the two-dimensional space. Therefore, three categories of fault occurrence period (FP1, FP2, FP3) can be used to identify properties of low dimensional space.

On the other hand, since the correspondence analysis maps the high dimensional data to the low dimensional space, not all the information can be expressed significantly. Each category has a well-described category in the selected low-dimensional space and a well-described category in the other. Therefore, it is necessary to interpret the results in the low dimensional space centered on categories with excellent expressive power.

The expression power analysis unit 224 analyzes the expressive power of the categories described for each axis. The expression power analyzing unit 224 may calculate the row expressing power of the intersecting table F according to Equation 15 and calculate the column expressing power according to 14. [

The present invention can extract meaningful results in each category by selecting categories having an average contribution of more than average and expressiveness in a two-dimensional space equal to or greater than predetermined reference expressive power and performing the result analysis.

6 shows the result of calculating the expressive power of each analysis variable for the target variable.

FIG. 6 shows that all the categories of each variable have 100% expressive power in two dimensions. This can be interpreted as a result of the one-dimensional and two-dimensional axes explaining 100% inertia in FIG. Therefore, it can be understood that the response analysis results can be interpreted by using the support units, the weapon systems, and all categories of the cause of failure.

The visualization unit 225 visualizes each category by using the coordinates of each category calculated by the inertia dimension analysis unit 222. The visualization is used to extract, summarize, and explain various meaningful analysis results. As described above, in the present invention, the visualization unit 225 visualizes the asymmetric pictures for analysis of the categories included in the two variables . The asymmetric figure can represent scatter plot for two variables and it has an advantage that correlation pattern between two variables can be grasped.

In this case, the visualization unit 225 can visualize the category in which the contribution is calculated to be less than the average in the contribution analysis unit 223 and the categories in which the expression power is less than the standard expression power in the expression power analysis unit 224.

Figures 7 to 9 show the corresponding analysis visualization results for the objective variable and the analysis variable, respectively.

FIG. 7 shows the result of visualization of the correspondence analysis between the failure occurrence period and the support unit, FIG. 8 shows the correspondence analysis visualization result of the failure occurrence period and the weapon system, and FIG. 9 shows the correspondence analysis visualization result of the failure occurrence period and the failure cause .

In Figs. 7 to 9, arrows indicate vector directions for three categories FP1, FP2 and FP3 of the fault occurrence period. The direction of the vector represents the property for each quadrant, and it can be confirmed that the property having the short duration (FP1 or FP2) and the property having the long duration (FP2 or FP3) are divided mainly on the basis of the one-dimensional axis.

The categories (M1 to M13), (W1 to W32), (design defect, operation defect, material defect, work defect, etc.) around the vectors shown in FIGS. 7 to 9 are respectively set as support units, It indicates categories of causes. The distributed categories are close to the arrows in the failure occurrence period category, and the higher the relative frequency of the information indicated by the arrows is, the farther away from the origin.

The pattern analysis unit 230 analyzes characteristics based on the failure occurrence period of the weapon system from the visualized correspondence analysis result. The pattern analyzing unit 230 includes a grouping unit 231 and a pattern extracting unit 232.

The grouping unit 231 groups each category of analysis variables in the visualized correspondence analysis result by using the similarity with the vectors corresponding to the three categories (FP1, FP2, FP3) of the failure occurrence period, which is the target variable. In other words, each of the categories (M1 to M13), (W1 to W32), (design failure, poor operation, material failure, As the group corresponding to the closest vector among the groups.

The grouping results of the grouping unit 231 are shown in FIG. 7 to FIG. 9 as three groups G1, G2, and G3.

The pattern extracting unit 232 calculates and outputs the relative frequency ratios of the respective categories FP1, FP2 and FP3 of the failure occurrence period, which is an objective variable for each of the groups G1, G2 and G3, And generates characteristic information.

Table 3 shows an example of the result of calculating the relative frequency ratio of the category (FP1, FP2, FP3) of the failure occurrence period to each group (G1, G2, G3).

The values entered in Table 3 represent the relative frequency of the three categories of fault occurrence periods (FP1, FP2, FP3) for each of the groups (G1, G2, G3). In the analysis of the failure variables, the values are not calculated because there are no categories related to the second category (FP2). However, we can see that the relative frequency of categories (FP1, FP2, FP3) in the remaining groups is high for each group (G1, G2, G3).

Also, the correspondence analysis between the failure period and the weapon system showed the highest relative frequency ratio among the analysis of the three analysis variables, and the correspondence analysis between failure occurrence period and failure cause showed the lowest relative frequency ratio. This means that the higher the correlation coefficient of the two variables to be analyzed, the more significant the corresponding analysis results are.

In other words, it can be seen that the most correlated pattern exists in the failure period and the weapon system. Therefore, it is possible to perform an analysis to select weapons systems having short failures and to grasp attributes.

For the analysis, weapon systems with a failure duration of less than 2 years were selected. As a result, weapon system categories distributed between the first category (FP1) vector and the second category (FP2) vector of FIG. 8 were selected. The characteristics of each weapon system were identified through frequency analysis of failure cause, support unit, and TUM.

Table 4 shows additional analysis results for weapon systems with a failure duration of less than two years.

Table 4 shows the frequency of each variable in a specific category due to the characteristics of data. The cause of defect was material failure, TUM showed high frequency in Module, but support group showed almost even distribution. Therefore, the characteristics of each weapon system were identified by considering the second highest relative frequency of each attribute.

In Table 4, some weapons systems (W23, W28, W24, W2, W21) show the characteristics of defective material failure and module failure, though the characteristics of the support units are different. In addition, the weapon system (W5) has a high failure rate in TOP, unlike the above weapon system, and the second cause of failure is the nature of operational failure. On the other hand, weapons systems (W31, W8) show different characteristics in the cause of failure. The weapon system (W31) is the cause of the other defects, and the material defects are the second cause of the defects. In the weapon system (W8), the design failure and other causes of failure were high, and it was seen that failure occurred at the TOP. By using the above analysis results, it is possible to grasp the properties of the weapon system having a short period of failure and to utilize it as useful information.

As a result, by using the fault analysis period-based data analysis apparatus of the present invention, it is possible to identify the most relevant factor in the fault occurrence period which is the target variable. Further analysis of the other factors using the identified factors can be used as useful information for future development of the weapon because it can easily grasp the characteristics of each factor in the occurrence period of the failure.

FIG. 10 illustrates a method of analyzing data based on a fault occurrence period according to an embodiment of the present invention, and FIG. 11 is a detailed view of the corresponding analysis step of FIG.

Referring to FIGS. 1 to 9, the data search unit 210 of the data analysis apparatus 200 searches data stored in the logistics support database 100 to acquire data to be analyzed (S10). The data search unit 210 can search for and obtain repair request information as data to be analyzed.

Then, the data search unit 210 sets a variable in the acquired data and preprocesses data corresponding to the variable (S20). The data search unit 210 may select an analysis variable to be analyzed and an objective variable to be an analysis criterion. At this time, the data search unit 210 can select an analysis variable based on the number of missing values and the number of levels of categories. In addition, the data search unit 210 converts the categorical variable into a categorical variable when the variable is not a categorical variable but a continuous variable, and performs a preprocessing to remove redundant data and noise data. In some cases, the preprocessing can be omitted.

Thereafter, the correspondence analysis unit 220 performs a correspondence analysis between the objective variable and the analysis variable using the acquired data (S30). Correspondence analysis analyzes the correlation between the objective variable and the two categorical variables of the analytical variable and visualizes the analysis result to derive the data pattern.

Referring to FIG. 11, in the corresponding analysis step S30, the correlation calculation unit 221 of the correspondence analysis unit 220 first analyzes the dependency of the analysis variable on the target variable (S31). Dependency analysis is based on the relationship between two categorical variables (fault duration and one analytical variable) expressed in a cross-tabular form and a number of analytical variables And TUM) to calculate the dependency information. The correlation calculating unit 221 may perform caracomarge checking to obtain the Pearson residual and calculate it.

At this time, the correlation calculating unit 221 may exclude the variable whose calculated correlation coefficient is less than the preset reference value from the analysis target.

Then, the inertia dimension analyzer 222 determines the number of dimensions for expressing the corresponding analysis result, and calculates the coordinates of each variable category according to the determined dimension (S32). The inertia dimension analyzer 222 may apply the singular value decomposition on the Pearson residual R to obtain the main coordinates of the rows and columns and the standard coordinates of the rows and columns.

Then, the contribution analysis unit 223 calculates and confirms the contribution indicating the inertia ratio of the categories described for each axis (S33), and the expressiveness analysis unit 224 calculates the expressive power of the categories described for each axis and confirms (S34). As a result, categories with a contribution less than the average contribution and categories with less expressive power than expressiveness are excluded from the analysis.

Thereafter, the visualization unit 225 visualizes the corresponding analysis result as a graph of the determined dimension using the remaining categories except for the excluded categories (S35). The visualization unit 225 visualizes each category of the failure occurrence period, which is an objective variable, in the form of a vector, and displays each category of the analysis variable (support unit and weapon system, defect cause) on the coordinate.

Referring again to FIG. 10, when the corresponding analysis is performed, the pattern analyzing unit 230 groups each category of the analysis variables indicated in the coordinates on the basis of the category of the objective variable represented by the vectors in the visualized correspondence analysis result (S40 ). That is, a plurality of categories of analytical variables are grouped into a group of one of categories of multiple variables.

Then, the pattern analyzer 230 calculates the relative frequency ratio of each category of the objective variable in each group, and derives a pattern having no correlation with the pattern most correlated to the objective variable from the group having a relatively high frequency ratio.

In some cases, the pattern analyzer 230 may derive a detailed analysis result by interpreting additional analysis variable categories for the group with the highest correlation.

The method according to the present invention can be implemented as a computer program stored in a medium for execution in a computer. Where the computer-readable medium can be any available media that can be accessed by a computer, and can also include both computer storage media. Computer storage media includes both volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data, (Digital Versatile Disk) -ROM, a magnetic tape, a floppy disk, an optical data storage device, and the like.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is evident that many alternatives, modifications and variations will be apparent to those skilled in the art.

Accordingly, the true scope of the present invention should be determined by the technical idea of the appended claims.

Claims (7)

Searching the subsequent logistical support database storing the field operation and maintenance information to acquire data related to the failure occurrence period indicating the period from the manufacturing date to the failure acceptance date and setting the failure occurrence period as the objective variable in the obtained data A data search unit for setting a plurality of analysis variables from the remaining data, and converting the continuous variable to a categorical variable when the objective variable and the plurality of the analysis variables among the plurality of analysis variables exist;
Analyzing the correlation between the target variable and each of the plurality of analysis variables by performing a corresponding analysis on each of the objective variable and the plurality of analysis variables to derive a pattern of the obtained data, A corresponding analysis unit; And
Grouping the categories of the analytical variables corresponding to the respective categories of the objective variable from the result of the visualized correspondence analysis and calculating a relative frequency ratio of each category of the analytical variables grouped, And a pattern analysis unit for extracting the highest group,
Wherein the data searching unit searches for and obtains repair request information among the information stored in the subsequent log support database, calculates the fault occurrence period from the repair request information, obtains the fault occurrence period as the target variable, A data extracting unit for selecting a plurality of the analysis variables based on the number of missing values and the number of levels of categories for the data, and, if there is a continuous variable among the objective variable and the plurality of analysis variables, And a preprocessing unit for removing redundant data and data having a number of levels lower than a predetermined reference number for each category by noise,
Wherein the correspondence analyzing unit is configured to express the objective variable as a categorical variable and each of the plurality of analysis variables in the form of a crosstabulation table and to calculate a correlation coefficient for calculating a correlation coefficient indicating dependency information of each of the plurality of analysis variables for the objective variable An inertia dimension analyzing unit for determining the number of dimensions for visually expressing the result of the corresponding analysis and calculating coordinates of each of the plurality of analysis variables on a space in accordance with the determined number of dimensions, Further comprising a visualization unit for displaying each category of the objective variable in a space of a determined dimension in a vector form and displaying a category of each of the plurality of analysis variables on the calculated coordinates to visualize the category. delete delete The apparatus of claim 1, wherein the correlation coefficient calculation unit
Calculating the correlation coefficient using Pearson residuals obtained by performing a chi-square test on the target variable expressed in the cross-tabular form and the categories of each of the plurality of analysis variables, And if the correlation coefficient is less than a preset reference value, excludes the corresponding analysis variable from the object of the corresponding analysis. The apparatus as claimed in claim 4, wherein the inertia dimension analyzing unit
Wherein the coordinates of each of the categories are obtained according to a singular value decomposition method for the Pearson residual. The apparatus of claim 1, wherein the counterpart analyzing unit
A contribution analyzer for calculating a contribution indicative of an inertia ratio of each of the categories for each axis in the space of the determined dimension and excluding an analytical variable for which the calculated contribution is less than the average contribution from the object of the corresponding analysis; And
An expression analyzing unit for calculating the expressive power of each of the categories for each axis in the determined dimension space and excluding analytic variables whose calculated expressive power is less than a predetermined reference expressive power from objects of the corresponding analysis; Further comprising: a data analyzer for analyzing the data. The apparatus of claim 1, wherein the pattern analyzer
Grouping the category of each of the plurality of analysis variables displayed in the coordinates calculated in the space of the determined dimension into a group corresponding to a category of the object variable closest to the category of the object variable indicated in a vector form; And
Calculating a ratio of the relative frequency of each category of the analytical variables grouped to extract the group having the highest degree of correlation with respect to the objective variable and analyzing the category of the objective variable and the category of the analytical variable corresponding to the extracted group A pattern extracting unit for sorting by a category for; And a data analyzing unit for analyzing the data.

Images