JP2007226639A - Multivariate data discrimination device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a multivariate data discrimination device easily specifying a factor by which an observation phenomenon is largely affected at high speed without supposing a distribution state of data on a factor item. <P>SOLUTION: This multivariate data distinction device has: data collection means 101-104 storing multivariate data added with attribute information; data group extraction means 105-108 set with an extraction condition designated with two factor items to one state item corresponding to a factor, extracting a multivariate data group corresponding to the factor items and the state item satisfying the extraction condition from the multivariate data; a discrimination possibility/impossibility decision means 109 deciding whether or not discrimination of the state item is possible in the factor item by minimizing a mixing area of the multivariate data group when two-dimensionally mapping and rotating the multivariate data group and projecting it to one factor item; and display means 111-113 calculating an influence degree of the factor item to the state item from a decision result of the discrimination possibility/impossibility decision means 109, and displaying it. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a multivariate data discriminating apparatus applied to device or facility failure prediction or human behavior prediction, and in particular, whether or not multivariate data collected from a device or facility can be discriminated by a specified item. The present invention relates to a multivariate data discriminating apparatus for determining whether or not.

In general, it is important to know factors related to a phenomenon such as a deterioration state when predicting a failure time of a device or facility.
At this time, as one criterion for specifying the factor for the state of the equipment or equipment, two states, a deteriorated state and an undegraded state, are set in advance, and these reference states and some collected data In the case where both are related, it is determined that those items are influencing factors.

  Further, as a determination means for determining whether or not the above relationship exists, two groups of data of items considered as influencing factors according to two states of a deteriorated state and a non-degraded state If the two factors can be discriminated, it is determined that the selected factor has an effect. On the other hand, if the two factors cannot be discriminated, it is determined that the selected factor has no effect.

That is, by identifying certain data according to several states, a factor that affects the deterioration is specified, and the deterioration time of the device or equipment is estimated.
As a determination method in such a multivariate data determination device, a method of plotting data of items considered as influencing factors and visually determining whether or not the plot result is divided into two groups by two states Is the most common.

However, when the number of collected data items is very large, the number of combinations of items that can be considered as a factor is enormous, and it is difficult to visually determine all the combinations.
In view of this problem, the application of discriminant analysis used in the field of statistics has been proposed.
In a conventional multivariate data discriminating apparatus that applies discriminant analysis, a technique is used to determine which is close to a newly collected sample for two known groups that have been discriminated in advance. The Mahalanobis distance is often used as a proximity criterion, and several methods have been proposed (see, for example, Patent Document 1 and Patent Document 2).

Patent Document 1 discloses a method for calculating a similarity using a discriminant function indicating a similarity measure of each class and selecting which class is discriminated for the purpose of pattern recognition. Yes.
Patent Document 2 discloses a method for classifying multivariate data in advance by classifying classes themselves into several larger categories and applying the above discriminant analysis technique to improve the discrimination accuracy. ing.

JP-A-8-106295 (pages 4-5, FIG. 2) JP 2004-213316 A (pages 5-7, FIG. 3)

  The conventional multivariate data discriminating apparatus aims at discriminating the deterioration state with higher accuracy in either case of Patent Document 1 or Patent Document 2, but the discrimination result corresponding to the state is known. Since the discriminant function is also known, the discriminant function for the actual data for which it is unknown whether it can be discriminated by a certain state cannot be derived.

  Further, the Mahalanobis distance used as a discrimination criterion in Patent Document 1 assumes that the distribution state of each group of data is a normal distribution, but the assumption of normality is not necessarily valid in actually collected data. In many cases, particularly when there is no correlation between two items that can be considered as influencing factors, there is a problem that the data is uniformly distributed and the Mahalanobis distance cannot be applied.

The present invention has been made to solve the above-mentioned problems, and in order to predict deterioration of facilities and equipment, human behavior prediction, etc., the observed phenomenon (factor) greatly affects which factor. It is an object of the present invention to provide a multivariate data discriminating apparatus that can easily and quickly specify whether or not the distribution of the data of factor items is assumed.
In order to realize the above purpose, the collected data of the influencing factor items are divided into multivariate data groups according to the state, and it is determined whether or not each data of the two groups can be discriminated by brute force. An object of the present invention is to obtain a multivariate data discriminating apparatus that can identify an item having the highest discriminating degree (the highest value that can be completely divided and the value expressed quantitatively) as an influencing factor.

  The multivariate data discriminating apparatus according to the present invention is a multivariate data discriminating apparatus that determines whether or not data collected by a sensor or manual measurement can be discriminated by an item designated as an influencing factor for one factor, Collecting data, adding attribute information including data ID information to the data, storing data in the database as multivariate data, and specifying two factor items for one status item corresponding to the factor When the extracted extraction conditions are set, the data group extraction means for extracting the multivariate data group corresponding to the state item and the factor item satisfying the extraction condition from the multivariate data, and the multivariate data group are two-dimensionally mapped. By rotating and minimizing the mixed area of multivariate data groups when projected onto one factor item, the state item can be identified by the factor item. And determining discrimination possibility determination means for determining capacity or not, the determination result of the determination permission determination unit, in which a display means for displaying by calculating the degree of influence of factors items for state item.

  According to the present invention, it is possible to easily and quickly specify which factor the observation phenomenon is greatly influenced without assuming the distribution state of the data of the factor item.

Embodiment 1 FIG.
FIG. 1 is a block diagram showing the overall configuration of the multivariate data discriminating apparatus according to Embodiment 1 of the present invention in the form of a flowchart. Show.
In FIG. 1, a plurality of measuring instruments 100 are provided in equipment or equipment (not shown) to be monitored. The measuring instrument 100 includes various sensors, measures state quantities related to each part of equipment and equipment, and outputs the data as data including the state to be monitored and influencing factors.

  A multivariate data discriminating apparatus according to Embodiment 1 of the present invention includes a collected data collection unit 101, an attribute information collection unit 102, a data combination unit 103, a multivariate data storage unit (database DB) 104, and a user operation. The data extraction means 105 that cooperates with the condition input means 105A, the extracted data storage means 107, the classification-by-state creation means 108, the discrimination availability determination means 109 that cooperates with the function input means 109A, and the discrimination availability judgment means 109. End determination block 110, factor influence calculation means 111, and factor influence display means 112 having a display 113.

Further, as indicated by a broken line arrow, a feature amount extraction unit 106 and a feature amount storage unit 106A are provided between the data extraction unit 105 and the state-specific classification creation unit 109 as necessary (described later).
The feature quantity extraction unit 106 and the feature quantity storage unit 106A constitute a variable compression unit. When the feature quantity extraction unit 106A includes multivariate items, only the feature quantity is extracted by decomposing the singular values of the multivariate data. The data is compressed into bivariate and sent to the state-specific classification creating means 109.

The collection data collection unit 101, the attribute information collection unit 102, the data combination unit 103, and the multivariate data storage unit 104 constitute a data collection unit that accumulates multivariate data to which attribute information (described later) is added as a database. Yes.
FIG. 2 is an explanatory diagram illustrating an example of a data format in each of the data collection units 101 and 102.
FIG. 2A shows an example of a recording format stored in the collection data collection unit 101. The collection data includes the failure occurrence rate, the shaft temperature, and the oil of the component associated with the equipment ID (or sensor ID). Including temperature and vibration values.
FIG. 2B shows an example of a recording format stored in advance in the attribute information collection unit 102, and the attribute information includes an installation location and an installation year associated with the equipment ID.

FIG. 3 is an explanatory view showing a table example of multivariate data in the multivariate data storage means 104 (database).
3, the multivariate data table includes the data of FIG. 2A (fault occurrence rate, shaft temperature, vibration value, oil temperature for each part ID “1, 2, 3,...”). Etc.) and attribute information (elapsed year etc.) in FIG. 2B are stored in association with each other.

Collection data (hereinafter simply referred to as “data”) detected (collected) by a plurality of measuring instruments 100 (or manual measurement) is stored in the collection data collection unit 101.
Here, the data includes the equipment ID (corresponding to the sampling location, etc.) and the state quantity of a certain part in the equipment (for example, failure occurrence rate, equipment operation, etc.) as shown in FIG. Influencing factors such as shaft temperature, oil temperature, shaft vibration value).
Further, the attribute information stored in advance in the accumulated data collection unit 102 includes influencing factors such as equipment ID (or sensor ID), installation location, and installation year, as shown in FIG.

The data combination means 103 reads the data in the collected data collection unit 101 and reads the attribute information in the attribute information collection unit 102 and adds it to the data, as shown in FIG. 3, as multivariate data for each equipment ID. The data is stored in the multivariate data storage means 104.
That is, the fixed attribute information for each equipment such as the installation location and the installation time of the equipment is stored in the attribute information collecting section 102 for each equipment ID, similarly to the data in the collected data collecting section 101. The two types of data in the sections 101 and 102 are merged by the equipment ID by the data combination means 103 and stored in the multivariate data DB 104.

  The data extraction unit 105, the feature amount extraction unit 106, the feature amount storage unit 106A, the extracted data storage unit 107, and the state classification creation unit 108 constitute a data group extraction unit, and a condition input unit 105A according to a user request. Is set in the multivariate data storage means 104 by setting the extraction information (measurement date, equipment ID or extraction target item, etc.) in which two factor items are specified for one state item. From the multivariate data, a multivariate data group corresponding to the state item and the factor item satisfying the extraction condition from the condition input means 105A is extracted.

Discriminability determination unit 109 (for example, a linear discriminability determination unit based on a linear discriminant function) rotates the group of multivariate data groups extracted by the data group extraction unit around the origin (the center of a discriminant line described later). Thus, by minimizing the mixed region (described later) of the multivariate data group when projected onto one factor item, it is determined whether or not the factor item can be linearly discriminated.
The factor influence degree calculating means 111, the factor influence degree displaying means 112 and the display 113 constitute display means for calculating and displaying the influence degree of the factor item with respect to the state item from the determination result of the discriminability determination means 109.

The multivariate data group extracted by the data extraction unit 105 from the multivariate data storage unit 104 based on the extraction condition from the condition input unit 105A is sent to the classification-by-state generation unit 108 via the extraction data storage unit 107. .
According to the specification of extraction conditions (items specified as state items, classification methods, etc.) from the condition input means 105A by the user operation, the state-specific classification creation means 108 converts the influence factor item data into a plurality of multivariate data groups. Classify into:

  For example, the state-specific classification creating means 108 designates “failure occurrence rate” of a certain part of equipment as a state, and designates this as high (60% or more), medium (20% to 60%), small (20% or less). ) Are classified into three multivariate data groups, such as installation location, shaft temperature, oil temperature, vibration value, and the like.

Subsequently, the state-specific classification creating unit 108 sends the classified multivariate data groups two by two to the discriminability determination unit 109.
The discriminability determination unit 109 performs a determination process based on a selection condition from the function input unit 109A according to a user request (whether the one that divides the multivariate data group is a linear discriminant function or a nonlinear discriminant function). . Here, a linear discriminant function is used as will be described later.
When the selection condition from the function input unit 109A is linear, no further user input operation is particularly necessary. However, when the selection condition from the function input unit 109A is non-linear, the function order and coefficient (for example, , Y = x3) and the like need to be further specified.

The discriminability determination unit 109 determines whether it is possible to discriminate according to user input for each of the two multivariate data groups from the state classification creation unit 108, and calculates a discriminating degree from the determination result.
Next, the end determination block 110 determines whether or not the determination process by the determination possibility determination unit 109 has been executed for all the groups (the determination process for brute force has been completed), and the determination has been completed for all the groups. If it is determined that there is no (that is, NO), the determination process by the determination possibility determination unit 109 is continued.

On the other hand, if it is determined in the end determination block 110 that the determination has been completed for all groups (that is, YES), the factor influence calculation means 111 is based on the determination degree calculated in the determination process for each multivariate data group. To calculate the factor influence on the selected state.
The calculated factor influence degree is displayed on the display 113 via the factor influence degree display means 112.

Next, with reference to FIG. 4, the determination processing operation by the determination possibility determination unit 109 in FIG. 1 will be specifically described.
FIG. 4 is an explanatory diagram showing a two-dimensional mapping state of the multivariate data group.
Here, when the cause of the failure occurrence rate of a component is a factor, FIG. 4A shows a case where the factor can be discriminated, and FIG. 4B shows a case where the factor cannot be discriminated.

4A, the horizontal axis (attribute information) is the number of years elapsed, and the vertical axis (state quantity) is the vibration value. In FIG. 4B, the horizontal axis (attribute information) is the installation location, and the vertical axis (state). Amount) is the oil temperature.
In FIGS. 4A and 4B, the plot point “□” corresponds to equipment with a low failure rate of parts, and the plot point “Δ” corresponds to equipment with a high failure rate.
For example, as shown in FIG. 4A, the detected vibration value tends to increase as the failure occurrence rate or the number of years elapsed.

First, the discriminability determination means 109 obtains data “vibration value” and “oil temperature” assumed to be influence factor data (state quantity) other than the “failure occurrence rate” selected as the state, as shown in FIG. Map in space as shown in b).
In this case, since two items are selected as the combination of the influencing factors, each becomes a two-dimensional mapping. That is, in FIG. 4A, the elapsed time is taken on the X axis, the vibration value is taken on the Y axis, and in FIG. 4B, the installation location (for example, elevation) is taken on the X axis, and the oil temperature is taken on the Y axis.
The two multivariate data groups to be compared are data “□” when the failure occurrence rate is low and data “Δ” when the failure occurrence rate is high.

At this time, in FIG. 4A, data “□” having a low failure rate and data “Δ” having a high failure rate are clearly distinguished from each other with a discrimination line 403 (see a two-dot chain line) as a boundary. In the case of FIG. 4A, it is determined that there are two multivariate data groups capable of linearly determining the failure occurrence rate (state item corresponding to the factor).
That is, since the combination of the items “elapsed years and vibration value” in FIG. 4A shows a different tendency depending on the “failure occurrence rate”, it is a factor that affects the level (factor) of the failure occurrence rate. Conceivable.

  On the other hand, in FIG. 4B, data “□” having a low failure rate and data “Δ” having a high failure rate are mixed and there is no discrimination line. Is determined to be two multivariate data groups that cannot be linearly determined as a factor (high or low failure rate).

  As shown in FIGS. 4 (a) and 4 (b), if the graph is divided into plot points “□” and “Δ” and displayed on the display 113 according to the difference in “failure occurrence rate”, linear discrimination is possible. It is possible for a person to visually determine whether or not the data group is a multivariate data group.

Next, referring to FIG. 5, a description will be given of the processing operation in the case where the determination is made automatically instead of being determined individually by visual observation.
5 (a) and 5 (b) show multivariate data groups related to the elapsed years, vibration values, installation location, and oil temperature, as in FIGS. 4 (a) and 4 (b). Each of FIGS. 5A and 5B shows a data group capable of linearly determining a factor (high or low failure rate).

As shown in FIG. 5A, if the discrimination line 501 is parallel to the vertical axis (Y axis) and the failure occurrence rate can be discriminated 100% by the discrimination line 501, the horizontal axis (X axis). It is possible to determine the failure occurrence rate only by the item corresponding to (years elapsed).
Now, if it is attempted to automatically determine the level of failure occurrence rate without visual inspection, data “□” corresponding to a low failure rate and data “△” corresponding to a low failure rate are obtained. Projecting on the X-axis as the region 503 and the region 504, respectively, it is only necessary to recognize whether the mixed region of each data on the X-axis is equal to or less than a predetermined value (a maximum value that can be considered to determine the factor). I understand.

On the other hand, as shown in FIG. 5B, if the discrimination line 502 is parallel to the X axis and the failure occurrence rate can be discriminated 100% by the discrimination line 502, an item corresponding to the Y axis (oil The failure rate can be determined only by (temperature).
Therefore, data “□” corresponding to the case where the failure occurrence rate is low and data “Δ” corresponding to the case where the failure occurrence rate is low are projected onto the Y axis as the region 505 and the region 506, respectively, on the Y axis. It can be seen that it is only necessary to recognize whether the mixed area of each data is equal to or less than a predetermined value.
In the examples of FIGS. 5A and 5B, since each data is completely classified by the discrimination lines 501 and 502, there is no mixed area, and therefore it can be considered that the factor can be discriminated. it can.

Next, with reference to FIG. 6, the rotation processing operation when the discrimination line 601 is not parallel to the X axis or the Y axis and the factor can be almost discriminated but cannot be discriminated 100% will be described.
FIG. 6 is an explanatory diagram showing a two-dimensional mapping state of the same multivariate data group as described above, and the X axis (elapsed years) and Y axis (vibration values) as factors when the failure occurrence rate is a factor. This shows the case where

FIG. 6A shows a state in which the discrimination line 601 is not parallel to the X axis or the Y axis, and the factor (the level of failure occurrence rate) can be almost discriminated, but is not 100% discriminable.
FIGS. 6B and 6C show a state in which the multivariate data group in FIG. 6A is sequentially rotated clockwise with the center of the discrimination line 601 as the origin.

In the state of FIG. 6A, the two multivariate data groups of data “□” having a low failure rate and data “Δ” having a high failure rate are distinguished by a discriminant line 601 composed of a negative linear function. .
Therefore, when each data corresponding to the multivariate data group is projected onto the X-axis, there are areas 602 and 603 where only one of the data exists, and an area 604 where the data of the two multivariate data groups are mixed. To do.

As shown in FIGS. 6B and 6C, the data group in FIG. 6A is set to the clock (right) as a group of the entire data little by little (for example, 10 degrees each) with the center of the discrimination line 601 as the origin. ) Rotate around and project data onto the X axis in each state.
In this case, as the rotation angle increases from FIG. 6A to FIG. 6B and FIG. 6C, the mixed regions 604, 614, and 624 gradually decrease in size, and FIG. As described above, when the discrimination line 601 is parallel to the Y axis, the mixed area 624 in the two groups of data areas projected onto the X axis is minimized.

As described above, when the data composed of the two groups is rotated and projected onto the X axis or the Y axis, a state in which the “ratio of the mixed area in the entire distribution area” becomes the smallest is obtained, and the influencing factors at that time Is regarded as the most discriminable state.
At this time, if the ratio of the mixed area is defined as “discrimination degree”, the discrimination degree is “0” when the mixed area is “0” and 100% can be discriminated as shown in FIGS. It becomes.
On the other hand, as shown in FIG. 4B, when the data of the two groups is completely mixed (each data area overlaps by 100%), the discrimination degree is “100”.

Therefore, as shown in FIG. 6C, if the discrimination degree (ratio) of the mixed region 624 set to the minimum is not more than a predetermined value (for example, 20 [%]), the factor is discriminated based on the extracted multivariate data. It is understood that is possible.
On the other hand, if the discriminating degree of the mixed region 624 set to the minimum is equal to or greater than a predetermined value, it can be understood that the factor cannot be discriminated by the extracted multivariate data.

In FIG. 6, the center of the discrimination line 601 is rotated as the origin, but the origin of the two-dimensional mapping (the intersection of the X axis and the Y axis) may be the center of rotation.
Further, the smaller the step size of the rotation angle of the multivariate data group is set, the higher the data correlation with the factor and the higher accuracy discrimination becomes possible.
Further, wherever the discrimination line exists, if it is rotated up to 90 degrees, it is always parallel to the X axis or Y axis, so it is sufficient to rotate it up to 90 degrees at the maximum.
For the data area of the influencing factors, the sample average ± 2σ is adopted in consideration of abnormal values and noise.

  Furthermore, although the case where the number of items of the factor that affects the level of the failure occurrence rate is “2” (elapsed years, vibration value) has been described here, the number of items of the influence factor is “3” (for example, elapsed years, If the value is greater than or equal to (vibration value, oil temperature, etc.), it automatically identifies the influencing factors in any number of multivariate items by calculating the discriminant degree (ratio of the mixed area) for every 2 items. It becomes possible.

  As described above, the multivariate data discriminating apparatus according to Embodiment 1 of the present invention is an item in which data collected by the measuring instrument 100 (or manual measurement) composed of sensors is designated as an influencing factor for one factor. Is a multivariate data discriminating apparatus that judges whether or not it can be discriminated, and collects data and adds attribute information corresponding to ID information included in the data to the data and accumulates it in the database as multivariate data Corresponding to the condition items and factor items satisfying the extraction condition from the multivariate data by setting the collection means 101-104 and the extraction condition in which two factor items are specified for one state item corresponding to the factor Data group extraction means 105-108 for extracting the multivariate data group, and two-dimensional mapping of the multivariate data group to rotate and one factor The determination result of the determination enable / disable determining means 109 for determining whether or not the state item can be determined by the factor item by minimizing the mixed region of the multivariate data group when projected to the eye, and the determination result of the determination enable / disable determination means 109 Display means 111 to 113 for calculating and displaying the influence degree of the factor item with respect to the state item.

The data collection means includes a collection data collection unit 101 that accumulates data, an attribute information collection unit 102 that stores attribute information, a data combination unit 103 that combines data and attribute information into multivariate data, And multivariate data storage means 104 serving as a database for storing variable data, and the attribute information includes data collection time or collection location.
Further, the discriminability determination means 109 calculates the ratio of the mixed area to the distribution area of the multivariate data group as the discriminating degree, and when the discriminating degree of the minimized mixed area indicates a predetermined value or less, the state is set as the factor item. It is determined that the item can be determined.
This eliminates the need to define a discriminant function. For example, as shown in FIGS. 4 (a) and 4 (b), multivariate data can be discriminated even when the data distribution is not normal and it is difficult to introduce the Mahalanobis distance. You can easily know whether it is possible.

Further, when the number of items that can be considered as factor items for the state item reaches three or more, the discrimination enable / disable determining unit 109 determines the factor items by two items.
Therefore, it is possible to know whether or not multivariate data can be discriminated by determining two items by two-dimensional mapping even in three or more multivariate items.

Embodiment 2. FIG.
In the first embodiment, the case where a linear discriminant function is selected by the function input unit 109A (see FIG. 1) is taken as an example, and it is determined that multivariate data can be discriminated when data can be distinguished by a discriminant line. However, even when a non-linear discriminant function (curve or the like) is selected by the function input unit 109A, the discriminability determination unit 109 can determine whether or not discrimination is possible.
Next, a second embodiment of the present invention in which a nonlinear curve or the like is selected as the discriminant function will be described with reference to FIG.

FIG. 7 is an explanatory diagram for explaining the discrimination method according to the second embodiment of the present invention, and shows the two-dimensional mapping state of the multivariate data group as described above. Here, a case where a cubic function is assumed as the nonlinear discriminant function is shown, but the same applies to the case where any other discriminant function is used.
When linear discrimination is performed as described above (see FIG. 6), the data group may be simply rotated little by little. However, when nonlinear discrimination is performed as in the second embodiment of the present invention, it is assumed. It is necessary to input the nonlinear discriminant function to be performed from the function input means 109A.

As shown in FIG. 7, when the discriminant function is a cubic function 702, the discriminant function is approximated by three discriminant lines 703, 704, and 705 individually corresponding to a plurality of regions 706, 707, and 708 (three regions).
In this case, the convex portion in the vertical direction on the cubic function 702 and the point where the differential coefficient is “0” are obtained and divided into three, and three discriminant straight lines 703, 704, 705 obtained by the respective differential coefficients are obtained. A plurality of regions 706, 707, and 708 obtained by each intersection are rotated.
Hereinafter, the method for determining the degree of discrimination by rotating is the same as in the first embodiment.

  As described above, according to the second embodiment of the present invention, the discriminability determination means 109 divides the discriminant function into a plurality of regions 706 to 708 that are finitely divided and performs linear approximation even when the discriminant function is nonlinear. Thus, it is possible to easily determine whether or not discrimination is possible even when a complicated nonlinear discrimination function is provided.

Embodiment 3 FIG.
In the first and second embodiments, raw multivariate data is handled as a determination target in the determination capability determination unit 109. However, there are very many items that can be considered as influencing factors, and two items are determined round-robin. If it is practically difficult to do so, the dimension is compressed by inserting the feature quantity extraction means 106 and the feature quantity storage means 106A using singular value decomposition, as shown by the broken line arrows in FIG. For example, the determination process may be executed after two-dimensionalization.
In this case, the influential factor may be easily found even in a large amount of multivariate data.

Further, if there are three or more items that can be considered as influencing factors (factor items for the state item), the feature amount extraction unit 106 may be used.
That is, the data group extraction unit includes a feature amount extraction unit 106 (variable compression unit), and the feature amount extraction unit 106 determines whether the factor item is a factor item when the number of items considered as a factor item for the state item reaches three or more. By decomposing the singular value, only the feature value is extracted and compressed into two variables.

It is a block diagram which shows in part a flowchart the whole structure of the multivariate data discrimination | determination apparatus which concerns on Embodiment 1-3 of this invention. It is explanatory drawing which shows the recording format of (a) collection data and (b) attribute information which concern on Embodiment 1-3 of this invention. It is explanatory drawing which shows the recording format of the multivariate data which concerns on Embodiment 1-3 of this invention. It is explanatory drawing which shows the two-dimensional mapping state of the multivariate data group which concerns on Embodiment 1 of this invention. It is explanatory drawing which shows the two-dimensional mapping state of the multivariate data group which concerns on Embodiment 1 of this invention. It is explanatory drawing which shows the rotation process of the two-dimensional mapping of the multivariate data group which concerns on Embodiment 1 of this invention. It is explanatory drawing which shows the nonlinear discriminant function on the two-dimensional mapping of the multivariate data group which concerns on Embodiment 2 of this invention.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 100 Measuring instrument 101 Collected data collection part 102 Attribute information collection part 103 Data combination means 104 Multivariate data accumulation means 105 Data extraction means 105A Condition input means 106 Feature quantity extraction means 106A Feature quantity accumulation means ( (Variable compression means), 107 extracted data storage means, 108 state-specific classification creation means, 109 discriminability determination means, 109A function input means, 111 factor influence calculation means, 112 factor influence display means, 113 display, 403, 501, 502, 601, 703, 704, 705 Discrimination line, 604, 614, 624 Mixed area, 706, 707, 708 Multiple areas.

Claims (7)

A multivariate data discriminating apparatus that judges whether data collected by a sensor or manual measurement can be discriminated by an item designated as an influencing factor for one factor,
Data collection means for collecting the data, adding attribute information including ID information of the data to the data, and accumulating in the database as multivariate data;
By setting an extraction condition in which two factor items are designated for one state item corresponding to the factor, the state item satisfying the extraction condition from the multivariate data and the multiple corresponding to the factor item are set. A data group extraction means for extracting a variable data group;
Is it possible to discriminate the state item by the factor item by minimizing the mixed area of the multivariate data group when the multivariate data group is two-dimensionally mapped and rotated and projected onto one factor item? Discriminability determination means for determining whether or not,
A multivariate data discriminating apparatus comprising: a display unit configured to calculate and display the degree of influence of the factor item on the state item from the determination result of the discriminability determination unit.   The discriminability determination means calculates the ratio of the mixed area to the distribution area of the multivariate data group as a discriminating degree, and when the minimized discriminating degree of the mixed area shows a predetermined value or less, the factor item The multivariate data discriminating apparatus according to claim 1, wherein the status item is discriminated to be discriminable.   3. The determination item according to claim 1, wherein when the number of items that can be considered as the factor items for the state item reaches three or more items, the determination possibility determination unit determines the factor items two by two. The multivariate data discrimination device described. The data group extraction means includes variable compression means,
When the number of items considered as the factor items for the state item reaches three or more items, the variable compression means extracts only the feature value by decomposing the singular value of the factor item and converts it into two variables. The multivariate data discriminating apparatus according to claim 3, wherein the data is compressed.   The discriminability determination unit includes a function input unit that inputs a linear discriminant function, and uses the discriminant function to determine whether the state item can be discriminated from the factor item. The multivariate data discriminating apparatus according to any one of claims 1 to 4.   The discriminability determination means has a function input means for inputting a nonlinear discriminant function, divides the discriminant function into a plurality of finitely divided regions and linearly approximates the state item by the factor item 5. The multivariate data discriminating apparatus according to claim 1, wherein the multivariate data discriminating apparatus according to claim 1 is determined. The data collection means includes
A collected data collection unit for accumulating the data;
An attribute information collection unit for storing the attribute information;
Data combining means for combining the data and the attribute information into multivariate data;
Including multivariate data storage means serving as a database for storing the multivariate data,
The multivariate data discriminating apparatus according to any one of claims 1 to 6, wherein the attribute information includes a collection time or a collection place of the data.

Images