JP2012094016A - Production information management apparatus and production information management method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To process information to be managed in such a manner as to grasp abnormality or defect relating to manufacturing of a product even before the abnormality or the defect occurs.SOLUTION: It is first evaluated how much values of respective items such as parameters characterizing manufacturing steps such as an inspection device, a worker dealing with manufacturing, or a set temperature included in result data of a product and environmental conditions such as an air temperature, or a humidity are similar between different manufacturing lots (1021). Next, a plurality of representative points of the items such as the inspection device, or the worker are determined (1022). A degree of similarity between the plurality of representative points and the item values of the manufacturing lots is then evaluated (1023). Next, with a combination of representative points for the most similar items in the result data (representative point pattern), the result data are associated and the number of such combinations is counted (1024). Thereafter, a distance between the representative point patterns is calculated and on the basis of the distance, the representative point patterns are mapped and displayed two-dimensionally. When a search range is set on that map, corresponding result data can be extracted.

Description

  The present invention relates to a technique for managing production information stored in a system such as a manufacturing execution system or a process information management system.

  As a background art in this technical field, there is JP-A-2008-250910 (Patent Document 1). This technique is based on the premise of data mining using a method of finding a classification method that maximizes the variance of the objective variable, when it is “classified” and when it is not. The processing time including the waiting time is mechanically classified using a time classification method, and data mining is performed using the classified time group as an explanatory variable. By classifying using time classification methods into time groups, it becomes possible to handle time as the number of groups, and thus process management can be performed by applying the conventional data mining method as it is. " Are listed.

  Moreover, there exists Unexamined-Japanese-Patent No. 2005-327024 (patent document 2). In this publication, “In an assembly system device, a defect caused by the previous process flows into the main body and is detected by the main body inspection. In response to the problem that it was difficult to prepare linked data for performing data mining, “In the registration process, the manufacturing lot number is included in the part lot number and quality control information such as quality data is stored in the parent part lot number. Each part lot number is registered with a peg, and the process database manages quality control information such as quality data linked to the parent part lot number. In the data mining process, the parent part process managed in the process database is used. By “data mining all quality control information up to the child parts process and displaying the result”, “main unit inspection data registration” Sometimes, a defect found in the body inspection, to quality information terminal child part of can be a series in the data mining process, and is described as the instantaneous data mining system can be realized which performs the factor extraction to. "

  Moreover, there exists Unexamined-Japanese-Patent No. 2005-234979 (patent document 3). In this publication, in order to “provide a quality control system and a defect detection method for specifying a process and a manufacturing apparatus that cause a defect in a short time” of a product, “a plurality of manufacturing apparatuses having equivalent functions in a plurality of processes” Manufacturing data storage means for collecting and storing manufacturing information indicating an outline of processing by the manufacturing apparatus in a quality control system that manufactures products using, and inspects with a predetermined inspection apparatus to determine whether the product is acceptable or not, and inspection Inspection data storage means for collecting and storing inspection information indicating an outline of processing by the apparatus, data extraction means for extracting desired information from the manufacturing information and the inspection information, and displaying the extracted information, Alternatively, it comprises a data display means for statistically processing the extracted information and displaying desired data, and the manufacturing information and the inspection information are associated with each lot number, When a good phenomenon occurs, the data extraction means extracts the device processing history for each lot number and displays it on the data display means, thereby verifying the device processing history and identifying the manufacturing apparatus caused by the defective phenomenon "This makes it possible to reduce the time required for specifying the process and the manufacturing apparatus that cause the failure."

JP 2008-250910 A (paragraphs 0020 and 0021) Japanese Patent Laying-Open No. 2005-327024 (paragraphs 0003, 0005, and 0007) JP 2005-234799 A (paragraphs 0008, 0009)

  Patent Document 1 is a technique for classifying explanatory variables so that when there is an abnormality in an objective variable such as a yield, the explanatory variable can be best explained. In particular, in the explanatory variables, a time that is not suitable for classification processing is used. There is a feature in the device. Japanese Patent Application Laid-Open No. 2004-133867 is a technique that enables a quality information including parts that constitute a defect found during an inspection of a main body to be found in an assembly-type product. Similarly to Patent Document 2, Patent Document 3 is a technique that enables a process and a device used therein to be quickly searched for defects found in a product inspection process.

  As described above, conventionally, a user classifies or searches information based on an abnormality such as a yield abnormality or an inspection failure. For this reason, it was difficult to grasp knowledge such as the relationship between the production process and quality before an accident or failure occurred.

  In view of such circumstances, an object of the present invention is to process information to be managed so that signs or trends of abnormalities and defects can be grasped even before abnormalities and defects related to product manufacturing occur. It is in.

As means for solving the above-mentioned problems, in the present invention, firstly, the product data has equipment such as reaction vessels used for production, measuring instruments such as scales, inspection devices, workers engaged in production, set temperature, etc. It evaluates how similar the values of each item, such as parameters that characterize the manufacturing process, environmental conditions such as temperature and humidity, and the delivery source and manufacturing time of the raw materials used, between different production lots. Next, a plurality of representative points for each item such as equipment, measuring device, inspection device, and worker are determined. Next, the similarity between each representative value and each item value of the production lot is evaluated. Next, for the combination of representative points (representative point patterns) in which the items of the actual data are most similar, the actual data is linked and the number of the points is counted. After that, the distance between the representative point patterns is calculated, and the two-dimensional map is displayed based on the distance. If the search range is set on the map, the corresponding performance data can be extracted.
Details will be described later.

  According to the present invention, it is possible to process information to be managed so that signs or trends of an abnormality or failure can be grasped even before an abnormality or failure related to product manufacture occurs.

It is a figure which shows the software structure of the production information management apparatus of this embodiment. It is a figure which shows the detail of a clustering means. It is a figure which shows the detail of a map display means. It is a PAD figure which shows the process by an item similarity evaluation means. It is a PAD figure which shows the process by the item representative point determination means. It is a PAD figure which shows the process by the performance data representative point pattern similarity evaluation means. It is a conceptual diagram which shows the correspondence of a representative point pattern and a map. It is a PAD showing processing by representative point pattern mapping means 1032. It is a display example of a representative point pattern map. It is another example of a display of a representative point pattern map. It is a display example of a screen when performing a search or selection operation. It is a display example of a screen showing a state when the selection range is reduced in selecting a cluster. This is a display example 1100 in the case where the distance between the target result and the representative point of the cluster is expressed by two pieces of information, cosine and size. It is a display example 1200 when the distance between the target result and the representative point of the cluster is expressed by an inner product. This is a display example 1500 when the distance between the target result and the representative point of the cluster is expressed by another product in an inner product.

  Hereinafter, embodiments (hereinafter referred to as “embodiments”) according to the present invention will be specifically described with reference to the drawings.

≪Introduction≫
As described above, the production information management device according to the present embodiment processes information to be managed so that the abnormality or defect related to the production of the product can be grasped even before the abnormality or defect occurs. It solves the problem. However, the following problems arise in solving this problem.

  First, in product management, when a continuous quantity such as yield is used as an objective variable, a threshold value is set for the objective variable to determine whether or not it is abnormal, so it is necessary to divide the explanatory variable into two There is. However, for this purpose, it is necessary to first grasp the structural relationship between the objective variable and the explanatory variable. However, this is not easy, it is not easy to set a threshold value, and as a result, it is difficult to improve production.

  Also, without knowing the production lot number of the product that caused the accident and the name of the objective variable such as yield, it was not possible to search or drill down the data. That is, a search involving an operation requiring input of a character string or prior knowledge is complicated and the search range is substantially limited.

  In addition, as a more general clustering problem, a “curse of dimension” is known. The curse of dimension is described in, for example, “Ohm company Ishii et al.“ Easy-to-understand pattern analysis ””.

Here, a supersphere with a radius r on a d-dimensional superspace and a supersphere with a radius a · r on a d-dimensional superspace that is slightly smaller than that are considered. However, a satisfies the relationship 0 <a <1, and the center of the supersphere with the radius r coincides with the center of the supersphere with the radius a · r. At this time, the volume V1 of the hypersphere with radius r and the volume V2 of the hypersphere with radius a · r are respectively

V1 ｒ r ^ d
V2 ∝ (a ・ r) ^ d

Therefore, the ratio δV / V1 between the difference δV between V1 and V2 and the volume V1 is

δV / V1 = {r ^ d− (a · r) ^ d} / r ^ d
= 1-a ^ d

It becomes.

Here, it can be seen that when d → ∞, δV / V1 → 1. This indicates that as the dimension increases, the difference δV is almost concentrated on the surface of the hypersphere having the radius r. That is, the content (volume) is present at an approximately equal distance from the center. The fact that they are at substantially the same distance from the center means that the data cannot be distinguished from each other by distance. Therefore, in the clustering based on distance, there is a problem that the difference between the data is not known when the dimension is increased. When performing clustering, it is necessary to pay attention to such a curse of dimension when various objective variables and explanatory variables are set in information for managing products to be manufactured.
The production information management apparatus of this embodiment can solve these problems.

Note that the production information management apparatus of this embodiment is a computer mainly including hardware such as an input unit, a display unit, a storage unit, and a control unit.
The input unit is an input interface that receives input from the user. The input unit is, for example, a mouse or a keyboard.
The display unit is a display interface that displays a result of information processing by the control unit. The display unit is, for example, a display or a printer.

The storage unit stores data input from the input unit, data indicating an intermediate result or final result of information processing by the control unit, and a program for realizing software included in the production information management apparatus. The storage unit is, for example, a hard disk drive (HDD), a read only memory (also a storage medium), or a random access memory (RAM). This program also includes a production information management program for executing the production information management method.
The control unit mainly implements predetermined information processing related to production information management. The control unit is, for example, a CPU (Central Processing Unit). The control unit executes a program stored in the storage unit using data input from the input unit. As a result, cooperation between software and hardware is realized.

≪Configuration≫
FIG. 1 is a diagram illustrating a software configuration of the production information management apparatus according to the present embodiment. This production information management apparatus includes clustering means 102 and map display means 103 as software.

  The clustering unit 102 clusters production information acquired from the production process 101. Here, the production process 101 refers to each of the processes necessary for production of a product performed in a factory or the like. For example, the process is defined in a manual designed in advance, and an operator forms a production line according to the manual and produces products as much as necessary.

  The map display unit 103 displays the result of clustering by the clustering unit 102, for example, as a map in which clusters generated by clustering are arranged. For example, the cluster on the map is given a predetermined shading according to the amount and quality of information contained in the cluster.

  As a display other than the map, for example, a detailed analysis range candidate, a factor candidate, and a spillover range candidate to be searched by the user are displayed. The detailed analysis range candidate is a cluster candidate that the user wants to examine in detail. A factor candidate is a candidate for a factor when information belonging to a cluster whose user wants to examine details is set as a factor constituting the cluster. The spillover range candidate is a cluster candidate that has some relationship with the cluster that the user wants to examine in detail. The user selects a desired cluster or the like from these candidates by a search operation.

  The production information is, for example, information that characterizes the details related to the production of a product managed by a production lot. This production information includes (1) equipment such as reaction vessels used for manufacturing, (2) measuring equipment such as scales, (3) inspection equipment used for product quality inspection, (4) workers involved in manufacturing, 5) Multiple specific items such as parameters that characterize the manufacturing process such as set temperature and set pressure, (6) environmental conditions such as temperature and humidity, (7) supplier of used raw materials and (8) production time (Sometimes referred to as “attribute”). The user checks the production process 101 and specifies the values of these items one by one, and generates production information in which the specified values are collected in a predetermined format. The generated production information is stored in the storage unit of the production information management device. Basically, objective variables and explanatory variables are set in the items of production information. In the production information, a production lot is allocated (associated) for each product.

  FIG. 2 is a diagram showing details of the clustering means. The clustering unit 102 includes an item similarity evaluation unit 1021, an item representative point determination unit 1022, a result data representative point pattern similarity evaluation unit 1023, and a similar result occurrence frequency management unit 1024.

  The item similarity evaluation means 1021 evaluates how similar each production lot value is between different production lots. Specifically, for each item of production information such as equipment, measuring device, inspection device, worker, etc., the similarity between any two actual data is digitized as a distance. “Result data” refers to data indicating the value (actual value) of each item of the production information of the product of interest or search target. For convenience of explanation, the result data may be simply referred to as “result”.

  In FIG. 2, on the right side of the block of the item similarity evaluation means 1021, the digitized similarity is schematically indicated by a circle (schematic diagram with reference numeral 10211). In this schematic diagram, in order to make the image easy to understand in a small space, the circles are arranged in a one-dimensional shape. However, in general, data indicating the evaluation result of the distance between any two actual results is stored by using a data format such as a combination table.

  The item representative point determination unit 1022 clusters the results of the processing performed by the item similarity evaluation unit 1021 and determines one or more representative points for each item of production information. “Representative point” refers to a value representative of the performance of one or more products belonging to a cluster generated by clustering.

  In FIG. 2, on the right side of the block of the item representative point determining means 1022, representative points determined for each item are schematically indicated by ★ (schematic diagram with reference numeral 10221). For the determination of the representative points, for example, a clustering method based on a hierarchical method such as a k-means method or a dendrogram may be used. Further, based on the dispersion of the distance between the performance data, a unique method may be used in which a group of performance values in a closer range is grouped and an average value of the group is a representative point.

  The result data representative point pattern similarity evaluation unit 1023 evaluates which combination of the representative points each item is most similar to. Specifically, the representative point pattern that is a combination of representative points selected one by one from each item is compared with the actual data, and the representative point pattern closest to the actual data is extracted.

  In FIG. 2, the representative point pattern and the actual data are schematically shown on the right side of the block of the actual data representative point pattern similarity evaluation means 1023 (schematic diagram with reference numeral 10231). The representative point pattern is expressed by connecting the representative points indicated by ★ in each item with a solid line (symbol A or the like). The actual data is expressed by indicating the values taken for each item by ● and connecting the values indicated by ● with a solid line (symbol B). In this schematic diagram, the representative point pattern indicated by the reference symbol A is closest to the actual data indicated by the reference symbol B. Therefore, the representative point pattern indicated by the reference symbol A is associated with the actual result data indicated by the reference symbol B.

  Further, the result data representative point pattern similarity evaluation unit 1023 counts the frequency indicating the number of times of the representative point pattern associated with the result data. The counted frequency may be referred to as “frequency data” or “actual frequency”.

  The similar result occurrence frequency management unit 1024 manages the processing result by the result data representative point pattern similarity evaluation unit 1023. In FIG. 2, the results managed by the similar record occurrence frequency management unit 1024 are shown as a table 10241 on the right side of the block of the similar record occurrence frequency management unit 1024.

The table 10241 includes a “representative point pattern” field and a “frequency” field.
In the “representative point pattern” field, a part (or all) of representative point patterns determined in principle by determining representative points are registered.
In the “frequency” field, the number of times the representative point pattern registered in the “representative point pattern” field is associated with the actual data by the actual data representative point pattern similarity evaluation unit 1023 is registered as the frequency.

In principle, the representative point pattern obtained by the performance data representative point pattern similarity evaluation means 1023 is:

K1 x K2 x ... x Kp
(Ki is the number of representative points determined in the i-th item of production information (i = 1, 2,..., P. P is the number of items of production information).)

There are as many as indicated by. The representative point of each item with the most similar value of each item of the performance data is extracted and combined to be a candidate for the representative point pattern. If the candidate is a representative point pattern that has not yet occurred, it is treated as a new representative point pattern, the row data of the table 10241 is generated, and the frequency is set to 1.

  On the other hand, in the case of a representative point pattern that has already occurred, the row of the representative point pattern is extracted from the table 10241 (corresponding to internal data), and this frequency value is incremented. In this way, it is not necessary to prepare the table 10241 corresponding to all combinations of K1 × K2 ×... × Kp, and memory resources can be reduced.

  FIG. 3 is a diagram showing details of the map display means. The map display unit 103 includes a representative point pattern distance evaluation unit 1031, a representative point pattern map unit 1032, a representative point pattern map management unit 1033, and a search range setting unit 1034.

  The representative point pattern distance evaluation means 1031 evaluates the distance between representative point patterns. This evaluation of the distance is basically equal to the evaluation when the evaluation of the distance by the item similarity evaluation means 1021 is performed for each representative point constituting the representative point pattern. Therefore, the description regarding the process by the representative point pattern distance evaluation means 1031 is omitted.

  The representative point pattern mapping unit 1032 maps this distance on a predetermined map based on the distance between the representative point patterns obtained by the representative point pattern distance evaluating unit 1031. This map is, for example, a low-dimensional map that is easy to understand intuitively, such as one-dimensional, two-dimensional, or three-dimensional.

  The representative point pattern map managing unit 1033 manages the map determined by the representative point pattern map forming unit 1032 as a representative point pattern map. The representative point pattern map (sometimes simply referred to as “map”) managed by the representative point pattern map management unit 1033 is displayed in a predetermined display form on the display unit of the production information management apparatus. For example, when production information of a new product is input from the input unit of the production information management apparatus, the representative point pattern map is updated based on the production information.

  The search range setting means 1034 is realized as a scroll bar display and input operation processing function that accepts a data search operation on the representative point pattern map, or a range specifying operation processing function using a mouse or the like.

<< Process >>
Next, processing performed by the production information management apparatus of this embodiment will be described. The subject of this processing is the control unit of the production information management apparatus.
First, the process of digitizing the distance between the values of each item of arbitrary two record data in the item similarity evaluation means 1021 will be described.

  Among the production information obtained from the production process 101, the identification information (eg, identification number or identification) of equipment such as reaction tanks used for production, measuring instruments such as scales, inspection devices, workers engaged in production, etc. Code). Usually, these do not change extremely with respect to time, but since they are information on a nominal scale, the concept of distance cannot usually be applied.

  On the other hand, the raw material, the lot number of parts, etc., which are information on the same name scale, change drastically in the time direction. In other words, the former identification information often relates to multiple production lots, and those elements, for example, workers, may be increased or decreased, and may be introduced or discarded in the measuring instrument. There is little change in the whole. Even if there is a change, it will only be partially changed every six months. On the other hand, the latter often appears only in a specific production lot.

  On the other hand, there may be a continuous amount such as a manufacturing process parameter such as a set temperature or an air temperature or humidity included in environmental conditions. Strictly speaking, it is a nominal scale, such as the month, season, and weather (sunny, cloudy, rain, etc.), but it is easy to think of order, and it is an arbitrary scale that is based on certain rules. Some can be replaced relatively intuitively with numbers on a proportional scale.

  Particularly problematic here is information on the nominal scale, which is difficult to intuitively convert to an order scale, interval scale, and proportional scale. An example of such an operator will be described. There are the following methods for digitizing such items.

  For example, if multiple workers are involved in one production lot, the “worker” item of the production lot indicates whether or not a certain worker is involved by 1, 0 or the work time, and is the same as the total number of workers. It can be expressed as a vector with a number of dimensions. This vector is referred to as an “operator performance vector”. The distance in the “worker” item can be defined as a linear sum or inner product of the ratio of cosine and size of the worker performance vector. If the total number of workers, which is the number of dimensions of the worker performance vector, is at least the number of workers involved in the two production lots to be compared, the distance can be calculated by the method described later (see FIG. 4). .

  In addition, when the ratio of cosine and size of the operator result vector is a ratio of a small one to a large one, the closer the distance (the more similar), the closer to 1, and the farthest distance (completely dissimilar) When it becomes 0. This is different from the normal concept of distance, but here it is defined as such.

FIG. 4 is a PAD (Problem Analysis Diagram) diagram showing processing by the item similarity evaluation means.
First, in the production information, a first production lot is selected at random (STEP 401), and then a second production lot different from this is selected at random (STEP 402). The processing of STEP403 to STEP408 described below is performed for all first production lots and all second production lots different from the first production lot.

  Next, it is searched whether or not the distance has already been evaluated by selection (combination) in which the first production lot and the second production lot are reversed (STEP 403). If it has not been evaluated yet (No in STEP 403), related workers related to the first production lot and the second production lot are listed so as not to overlap, and the vector length of each worker result vector is calculated ( (STEP 404).

  Next, an operator result vector for each of the first production lot and the second production lot is generated (STEPs 405 and 406), and an inner product based on the worker result vector of the first production lot and the worker result vector of the second production lot is calculated. (STEP407).

Next, the calculated inner product is stored in the storage unit for the combination of the identification information indicated as the first manufacturing lot and the identification information indicated as the second manufacturing lot (STEP 408). A cosine / size ratio may be stored instead of the inner product. By storing in this way, the distance between the first production lot and the second production lot is defined.
The processing by the item similarity evaluation unit 1021 has been described above.

  The change described above, that is, the increase or decrease in the number of workers results in an increase or decrease in the dimension of the worker performance vector. However, if the distance is defined in this way, the increase / decrease of the dimension does not affect the past performance, so there is no problem even if there is a change. For example, assume that there are three workers A, B, and C at the beginning. Even if the number of workers D is newly increased here, the distance between the production lot performed with the addition of the worker D and the previous production lot is evaluated in a four-dimensional space. However, it does not affect the distance between production lots implemented by any combination of workers A, B, and C before worker D is added.

  On the other hand, when the presence / absence of use of a part number, a raw material lot number or the like is expressed as a vector, the length of the vector changes for each production lot, and the distance between arbitrary lots is always zero. Therefore, in the production information management apparatus according to the present embodiment, it is assumed that information such as a raw material lot number and a part number that changes rapidly is not used. However, the present invention does not exclude the use of such information. Such information that changes drastically may be dealt with by replacing it with information that does not change drastically, such as the delivery source code, product name (or product name), delivery season or month, or a combination thereof.

  Further, as a method of mapping such information on the nominal scale to a one-dimensional space similar to the schematic diagram shown by reference numeral 10211 in FIG. 2, there is a method such as a self-organizing map (Kohonen network). In particular, self-organizing map technology targeting documents calculates similarity between documents from words (information on nominal scales) that appear in the document, and the result of this calculation is relatively one-dimensional or two-dimensional. It maps onto a map with a small number of dimensions and can be used as a means for realizing the present invention.

  Needless to say, data (variables) on an interval scale or a proportional scale such as temperature and humidity can be mapped to a one-dimensional space similar to the schematic diagram indicated by reference numeral 10211 in FIG.

  Next, processing in the item representative point determination unit 1022 will be described. As described above, since the distance between any two results is known for each item, the representative point can be determined by applying the k-means method or the hierarchical method. Here, the hierarchical method will be described. To do. What is described here is a method called a cohesive type, particularly in a hierarchical method.

FIG. 5 is a PAD diagram showing processing by the item representative point determining means. This process is performed for the combination of the first production lot and the second production lot.
First, an item cluster including only one piece of the record data is generated using one piece of record data as a representative point (STEP 501). Here, the “item cluster” means a cluster generated by performing clustering while paying attention only to specific items constituting the performance data. The item cluster is generated, for example, by the number of actual data.

  Next, two item clusters are randomly selected from the generated item clusters, and all combinations of the two item clusters are specified. Then, for all the combinations, the distance between the representative points included in each of the two item clusters is calculated, and the item clusters are arranged in order of the closest distance (STEP 502). The distance between the representative points can also be referred to as a distance between item clusters, and can also be referred to as a distance between the result data when only one result data is included in the item cluster.

  Next, a threshold that can determine the roughness of the cluster is determined based on the distribution of the distance between the item clusters (STEP 503). The threshold is determined to be 3σ using, for example, a variance σ obtained from a set of target performance data. “Target performance data” is, for example, performance data of all or part of products stored in the storage unit of the production information management apparatus. The threshold is determined by, for example, an input from the user input unit.

  Next, based on the threshold value, two item clusters located within a predetermined distance from each other are specified, and the processing of STEP 505 to STEP 508 described below is repeated for these item clusters (STEP 504).

  First, as will be described later, if the distances between the two item clusters to be combined into one item and the other item clusters have been calculated, the data indicating the distance is deleted (STEP 505). .

  Next, a new item cluster including the two item clusters that are closest, that is, having the smallest distance is generated, and a representative point of the new item cluster is calculated (STEP 506). The representative point calculation method will be described later.

  Next, the distance between the representative point of the new item cluster and other item clusters, that is, the other item clusters other than the two item clusters related to the same item is calculated (STEP 507).

  Next, the distance obtained by the calculation in STEP 507 is stored in the storage unit for the combination of the first manufacturing lot and the second manufacturing lot (STEP 508).

  In this way, by creating the representative points of item clusters in which items that are close to each other are gathered in stages, one item cluster that ultimately falls within the threshold can be generated. As a result, it can be aggregated into several item clusters corresponding to the distribution of distances. Note that “contains within the threshold” means that the shortest distance between the representative points of the generated item clusters is smaller than this threshold. And, “the shortest distance is smaller than this threshold” means that the ratio of cosine and size becomes a large value (approaching 1), contrary to the normal distance, according to the definition described above. means.

  The generation of the representative point pattern that combines the representative points of the item clusters determined in this way may be performed each time a record of a certain production lot is added. Further, for example, the timing may be determined according to some standard, such as every month.

  In STEP 506, a method for determining a representative point of a new item cluster obtained by combining the two item clusters from the representative points of the two item clusters will be described. If the value of actual data is a scalar quantity or vector quantity on a proportional scale or interval scale, an average value (vector average) is generally used as a representative point.

  For ordinal scales, numerical values (domains) are assigned according to certain criteria and considered as interval scales or proportional scales, and the average value may be calculated to be the closest domain value, or the domain itself may be expanded. May be. In this way, the representative points for the order scale are determined.

In the case of a nominal scale, it can be treated as a vector on a proportional scale or interval scale by replacing it with the presence or absence and frequency of appearance. However, as described with reference to the example of the worker in FIG. 4, when taking the inner product (or the ratio of cosine to size) using only the nominal information of the specific item to be compared, the clusters are being joined together. This increases the possible nominal scale values. For this reason, it is necessary to manage not only a vector but also a vector of identifiers in the nominal space. In this way, a representative point for the nominal scale is determined.
The above is the processing by the item representative point determination unit 1022.

Next, processing in the performance data representative point pattern similarity evaluation unit 1023 will be described.
FIG. 6 is a PAD diagram showing processing by the performance data representative point pattern similarity evaluation means.

  First, the processing of STEP 6011 to STEP 6015 described below is repeatedly performed for each item of performance data (STEP 601). That is, first, representative point candidates that are candidates for representative points that are closest to the performance data in the item are initialized (STEP 6011). Next, the processing of STEP 6013 and 6014 described below is repeated for all representative points existing in the item (STEP 6012). That is, it is determined whether or not the representative point of interest is closer than the closest representative point candidate (STEP 6013). If it is close (Yes), the representative point of interest is updated to be the closest representative point candidate. (STEP 6014). The determination is performed by calculating the distance between the value of the result data and the representative point of interest. Thereby, the representative point in the item is determined (STEP 6015).

Next, a representative point pattern closest to the actual data is determined for the determined combination of representative points, and the presence / absence of frequency data for this representative point pattern (whether the frequency value is 1 or more) is determined (STEP 602). ). At that time, the table 10241 (see FIG. 2) managed by the similar performance occurrence frequency management means 1024 is referred to.
If there is frequency data (YES in STEP 602), the frequency is incremented in the table 10241 (STEP 6021). If there is no frequency data (No in STEP 602), it means a new representative point pattern, and the representative point pattern is added as row data in the table 10241, and the frequency is set to 1 (frequency initialization) ( (STEP 6022). In this way, the frequency of occurrence of the representative point pattern can be updated.

This process may be performed each time performance data is added, or may be performed by determining the timing on some basis, for example, every month.
When the representative point pattern is determined, the representative point pattern and the performance data determined to be closest to the representative point pattern are linked by the number of frequencies.
The above is the processing by the performance data representative point pattern similarity evaluation unit 1023.

  Since the processing by the representative point pattern distance evaluation means 1031 is basically the same as the evaluation performed for each representative point constituting the representative point pattern, the description thereof is omitted. As a result, the distance between the representative point patterns is calculated.

  Next, processing in the representative point pattern mapping unit 1032 will be described. As described above, the representative point pattern can take K1 × K2 ×... × Kp combinations, but can be ranked according to the frequency of the corresponding results.

First, the contents of assigning the representative point pattern to the map will be described.
FIG. 7 is a conceptual diagram showing the correspondence between the representative point pattern and the map.

  This map is mainly composed of nodes and links. The number N of nodes on the map is defined in advance by, for example, input from the input unit. From the ordered representative point patterns, N representative point patterns with a high actual frequency (the value of the frequency data is large) are extracted and assigned to the nodes of the map without duplication. The map shown in FIG. 7 is two-dimensional, and whether the map is one-dimensional, two-dimensional, or three-dimensional depends on the system design, but the distance between adjacent nodes on the map is Evaluation is based on the distance between the assigned representative point patterns. In this way, the distance between the corresponding representative point patterns is assigned to all the links connecting the nodes on the map. Therefore, the assignment of representative point patterns to nodes may be determined so as to minimize the total link length. Such a problem can be optimized by using a technique such as a genetic algorithm.

  In the two-dimensional case, the number of adjacent nodes for one node is usually four. However, considering the arrangement on the hexagonal lattice, the number of adjacent points may be six. Similarly, in the case of the three-dimensional case, the number of adjacent nodes is usually 6 points. However, if the network structure is a hexagonal close-packed lattice, 12 points can be used.

  In addition, it is desirable that the arrangement of the representative point pattern is a strict optimal solution, but since the purpose is to express the similarity of the representative point pattern to a smaller number of dimensions, It doesn't mean you don't have to.

When strict optimization is not required in this way, it can be mapped by a method as shown in FIG. 8, for example.
FIG. 8 is a PAD diagram showing processing by the representative point pattern mapping unit 1032.
First, for all combinations of selected representative point patterns, the distance between the representative point patterns is obtained, and a full mesh network relating to the representative point pattern having the distance as the link weight is constructed (STEP 801).

  Next, find the average distance, that is, the average weight of the links, and leave only the ones that are close to each other so that the number of links in the representative point pattern is at least the predetermined number or more for all nodes. Pruning is performed (STEP 802).

  Next, paying attention to the representative point pattern with the highest actual frequency, this is assigned as a selection node to the central node on the map (STEP 803).

  Next, an unassigned node that is not assigned a representative point pattern among nodes adjacent to the selected node is selected in a predetermined order (STEP 804). Next, the representative point pattern closest to the representative point pattern assigned to the selected node is assigned to the selected unassigned node (STEP 805). The processing in STEP 804 and STEP 805 is repeated in a predetermined order until there is no unassigned node. The predetermined order is, for example, the order in which the distance between the representative point pattern assigned to the selected node and the representative point pattern assigned to the unassigned node is small. In addition, as described above, there are usually four adjacent nodes in a two-dimensional full mesh network, but there are six adjacent nodes in consideration of an arrangement on a hexagonal lattice.

  Next, one node (which may be random) is selected from the nodes to which the representative point pattern is assigned according to a predetermined order (STEP 806). Next, among the nodes adjacent to the selected node, unassigned nodes to which no representative point pattern is assigned are selected in a predetermined order (STEP 807). The predetermined order is, for example, the order in which the distance between the representative point pattern assigned to the selected node and the representative point pattern assigned to the unassigned node is small. Next, the representative point pattern closest to the representative point pattern assigned to the selected node is assigned to the selected unassigned node (STEP 808). The processing of STEP806 to STEP808 is repeated in a predetermined order until there are no unassigned nodes. By repeating this, optimality in a strict sense is not guaranteed, but a map reflecting the distance between representative point patterns assigned to nodes can be constructed.

  Such an assignment is arbitrary (may be set clockwise in the upper left). For this reason, there is a possibility that the solution can be converged more quickly if the arbitraryness is determined by waving a random number and applying a genetic algorithm using it as an initial solution. May be.

  Also, instead of allocating the most frequent representative point pattern to the central node in the allocation to the node, increase the number of nodes on the map as appropriate, and stop the allocation at the stage of allocation to the predetermined node . By doing so, it can be expected to suppress the problem that the allocation of data to the map is biased and cannot be allocated at the boundary, and such a method may be adopted.

  Instead of mapping using all items, a representative point pattern may be assigned to a one-dimensional or two-dimensional map for each item. In this case, instead of performing distance evaluation for all items, distance evaluation is performed for each item (including calculation of a distance (similarity) between representative point patterns when the item is used). And the results should be managed for each item (including counting of frequency values). The evaluation, implementation, and management can also be performed when two or more items are used.

Next, processing in the representative point pattern map management unit 1033 will be described.
FIG. 9 is a display example of a representative point pattern map. On the display unit of the production information management apparatus, the representative point pattern map is displayed on the screen in a predetermined display mode by the dialog shown in FIG.

  The representative point pattern map emphasizes and displays a representative point pattern having a high frequency of similar results. For example, it may be possible to support the understanding of the distribution of the performance data by displaying in different colors according to the distance between the representative point patterns assigned to the nodes. In FIG. 9, for example, the distance between the representative point patterns is displayed by a black and white gradation. It is expressed that hexagonal clusters given the same color have similarities different from the similarities shown by the set of clusters given other colors.

  When a three-dimensional map is used, it can be displayed as a three-dimensional object that can be rotated by a screen operation from a user input unit.

  FIG. 10 is another display example of the representative point pattern map. This representative point pattern map is obtained by laying out a map of each item of production information by the representative point pattern map forming unit 1032 and displaying it in a two-dimensional table form (in a thumbnail format). The map is color-coded according to the distance between representative points for each item. In FIG. 10, for example, the distance between the representative points is displayed by black and white gradation. It is expressed that hexagonal clusters given the same color have similarities different from the similarities shown by the set of clusters given other colors. The mapping can also be performed when two or more items are used.

  Even when the display mode shown in FIG. 10 is used, when a three-dimensional map is used, the three-dimensional map can be displayed as a three-dimensional object that can be rotated by a screen operation from the input unit of the user.

Next, processing in the search range setting unit 1034 will be described.
FIG. 11 is a display example of a screen when a search or selection operation is performed. This screen is a screen displayed by the display unit of the production information management apparatus, and displays a screen 901, a search dialog 920, a calendar dialog 930, and a result selection dialog 940.

  On the screen 901, a cluster display unit 950 and a data display area 960 are mainly displayed. The cluster display unit 950 displays, for example, an equivalent map shown in FIG. In the data display area 960, the details of the cluster that the user pays attention to among the clusters displayed in the cluster display unit 950 are displayed (details will be described later).

  The search range setting unit 1034 supports search and comparison of similar data with as little character string input as possible by executing search processing and displaying the result based on user input as an operation required for the search. The user input includes, for example, a “finger” button 9021, a “latest” button 9022, a “most” button 9023, a “search” button 9024, a “calendar” button 9025, a “comparison” button 9026, and a scroll displayed on the screen 901. There are mouse 904 click operations and mouse wheel operations 90411 for the bars 951, 952, 961, and the like.

  First, by clicking the “Search” button 9024, the “Calendar” button 9025, the “Latest” button 9022, the “Most” button 9023, etc., the target, for example, a regular hexagonal cluster displayed on the cluster display unit 950 is displayed. select.

  When a “Search” button 9024 is clicked, a search dialog 920 for selecting actual data is displayed based on information such as product type, product name, and lot (referred to as a production lot). And the performance data used as object is selected by operation here. Hereinafter, the result data selected by the processing of the search range setting unit 1034 in this way is referred to as “target result”. In FIG. 11, in the search dialog 920, performance data belonging to the product type: soft drink, product name: Sawayaka ○○, and the production lot number L001110 is selected and highlighted by shading or the like.

  In the search dialog 920, a “date” column in which the date of manufacture of the product identified by the production lot number is listed is displayed on the right side of the “lot” column in which the production lot numbers are listed. Is displayed. Further, on the right side of the “date” column, a “size” column in which the sizes of products identified by the production lot number are displayed in a list is displayed.

  When the target record is selected, the cluster display unit 950 selects the cluster to which the target record belongs. The cluster selected by the processing of the search range setting unit 1034 in this way is called a “target cluster”. The “cluster to which the target achievement belongs” is a cluster to which the representative point pattern associated with the target achievement belongs.

  When a “calendar” button 9025 is clicked, a calendar dialog 930 for designating a date is displayed, and a date selection operation can be performed. When the date of the calendar displayed in the calendar dialog 930 is dragged with the mouse, the period indicated by the dragged date is selected. Thereby, the performance data of the product manufactured within this period is first selected. Note that the period is highlighted by shading.

  Then, when the “OK” button in the calendar dialog 930 is clicked, a result selection dialog 940 is displayed. The result selection dialog 940 displays information equivalent to the “lot” field, “date” field, and “size” field as the lot information. In the result selection dialog 940, by clicking one lot with a mouse, a search target is specified and a target result is selected. When the target record is selected, the cluster display unit 950 selects the cluster to which the target record belongs.

  When the “latest” button 9022 is clicked, the latest result data is selected as the target result. “Newest” may mean that the date of manufacture is the latest, or may mean that the registration and update of the performance data is the latest. When the record data is selected by the “latest” button 9022, the target cluster is the cluster to which the selected target record belongs.

  When the “most” button 9023 is clicked, the target achievement is not selected, and the cluster having the largest achievement included in the cluster characterized by the representative point pattern is selected as the target cluster.

  In addition, when the “finger” button 9021 is clicked, the mouse icon changes from a normal mouse icon (e.g., corresponding to a symbol indicated by reference numeral 9301) to a finger-like “finger” mouse icon 9502. Thereby, the cluster on the map displayed on the cluster display unit 950 can be selected. Then, by clicking on one of the clusters, the target cluster can be selected directly.

  Also, by turning the wheel 9041 of the mouse 904, the “area of interest” (corresponding to the cluster indicated by reference numeral 953 or reference numeral 954) that is a cluster around the target cluster can be expanded or contracted to the periphery. An arrow 9501 indicated by an alternate long and short dash line in the drawing schematically shows this state (a state in which the area of interest expands).

  For such a selection operation and range selection operation, the data display area 960 may display, for example, the actual production lot number and the actual data number closest to the representative point pattern of the target cluster or the cluster of the area of interest. Good. In addition, the color may be changed according to the selection range. In the “representative lot number” column of the data display area 960, a list of the actual data belonging to the corresponding cluster and the one closest to the representative point pattern belonging to the cluster is displayed. In the “number of achievements” column, a list of the number of achievement data belonging to the corresponding cluster is displayed. The color display applied to the cluster in the cluster display unit 950 and the color display in the data display area 960 correspond to each other. That is, the color applied to a cluster in the cluster display unit 950 is the same as the color displayed in the data display area 960 and applied to the description line of the record data belonging to the cluster.

  When the selection range is changed, the display content of the data display area 960 may be updated in conjunction with the change of the selection range. For example, in FIG. 11, 19 clusters including the target cluster selected by the “finger” mouse icon 9502 are selected, and the data display area 960 also includes information on the performance data belonging to the 19 clusters (representative lot number and actual performance). Number) is displayed.

On the other hand, the search operation can be performed so as to select a range smaller than the range when one cluster is targeted.
FIG. 12 is a display example of a screen showing a state when the selection range is reduced in selecting a cluster.

  FIG. 12A shows a state when the selection range is narrowed down to one by the mouse wheel operation 90411. In the cluster display unit 950 of FIG. 12A, a state where the selection range including the area of interest is narrowed down to one cluster is indicated by a dashed-dotted arrow 9501. At this time, the data display area 960 also displays only one cluster, that is, only the information for the cluster that is shaded in the cluster display unit 950.

  FIG. 12B shows a screen 901b when the area of interest is further narrowed down from the state of FIG. As shown in FIG. 12B, the selection range may be further narrowed down and displayed according to the distance from the record selected in the cluster (intracluster display). The “actual record selected within the cluster” means the target actual result. In FIG. 12B, the actual data identified by “L001010” in the “lot number” column of the data display area 970 is displayed. Point to. This result is selected by clicking the row data with the mouse.

  In the intra-cluster display state as shown on the screen 901b, the display content of the right data display area 970 is displayed with information on the distance between the identification code of the record data included in the area of interest and the selected data. It may be. The identification codes are listed in the “lot number” column of the data display area 970. The distances are listed in the “Distance” column of the data display area 970. “Selected data” is the above-mentioned “actual record selected within the cluster” (distance 0).

  In this case, the number of display data in the data display area 970 may be increased or decreased or the color may be changed according to the change of the range of the area of interest. FIG. 12B schematically shows the state of color change with a dashed-dotted arrow 9701.

In FIG. 12B, when the “detail” button 9503 is clicked, the distance between the target result and the representative point pattern of the cluster may be displayed for each attribute.
FIG. 13 shows a display example 1100 in the case where the distance between the target result and the representative point of the cluster is expressed by two pieces of information, cosine and size.
FIG. 14 shows a display example 1200 when the distance between the target performance and the representative point of the cluster is expressed by an inner product.
FIG. 15 shows a display example 1500 in the case where the distance between the target result and the representative point of the cluster is expressed by an inner product in another format.
Display examples 1100, 1200, and 1500 are screens displayed when a “details” button 9503 (FIG. 12B) is clicked.
In the display examples 1100, 1200, and 1500, the value of the target performance is also displayed in the “value” item at the same time.

Here, in the case of a scalar variable (explanatory variable or objective variable) as in Item 5, the cosine is always zero, but the scalar type attribute and the vector type attribute may be displayed separately. . In addition, in the display example 1200, since the target result is a comparison with the representative point pattern that is a representative of the target cluster, each item calculated based on the comparison between the representative point pattern and other result data included in the cluster is displayed. An example of normalizing and displaying based on the variance of the distance (inner product) was shown. However, the calculation result may be displayed as it is, as in a display example 1500 in FIG.
In this way, it is possible to easily grasp the peculiar points of the selected data in the cluster and the points that are not significantly different from others.

Similarly, in the in-cluster display state of the screen 901b of FIG. 12B, in addition to selecting the target results, one result ("comparison target result") from the result data in the area of interest displayed in the data display area 970 is displayed. It is assumed that “Details” button 9503 is clicked on. In this case, the distance between the two results data of the target result and the comparison target result may be calculated and displayed as shown in FIGS. By doing in this way, it is possible to easily grasp what is different and what is not different compared to other achievements in the same cluster.
Note that the operation for selecting the target results and the operation for selecting the results to be compared are the same, but the production information management apparatus performs processing separately from the selection of the target results and the selection of the results to be compared.

In addition, in the state where a plurality of clusters as shown in FIG. 11 and FIG. 12 are displayed, one cluster (referred to as “comparison target cluster”) other than the target cluster is selected by the “finger” mouse icon 9502 (see FIG. 11). Suppose that At this time, when the “comparison” button 9026 is clicked, the same diagram as FIG. 13 and FIG. 15 may be displayed. In this case, the representative point patterns of the two clusters of the target cluster and the comparison target cluster are used to compare the values. In this way, differences and features between clusters can be easily grasped.
Note that the target cluster selection operation and the comparison target cluster selection operation are the same, but the production information management apparatus performs the processing in distinction between the target cluster selection and the comparison target cluster selection.

  Similarly, when a target result is selected and another cluster is selected as a comparison target cluster and a “Compare” button 9026 is clicked, the selected target result and the representative point pattern of the selected comparison target cluster May be calculated and displayed as shown in FIGS. In this way, it is possible to easily grasp at what point the selected target performance is characteristic with respect to the selected comparison target cluster and at which point there is no significant difference. .

  Note that when displaying as shown in FIG. 14, the dispersion of the distance between the actual data included in the selected cluster and the representative point pattern of another cluster may be used. Further, the dispersion of the distance between the representative point pattern in the cluster including the selected result data and the result data in another cluster may be used. By using the display format as shown in FIG. 14, it is possible to determine whether it is relatively large or small with respect to the cluster, not the absolute value of the distance for each item. This has the effect of making it easier to grasp.

  When the target result is selected, the target cluster is also selected, and comparison between the result and the cluster and the comparison between the cluster and the cluster are possible. The former or the latter may be given priority, or a selection dialog may be displayed for selection.

≪Summary≫
From the above description, according to the present embodiment, it is possible to process the information to be managed so that the abnormality or defect can be grasped even before the abnormality or defect related to the manufacture of the product occurs.

  Similarity (distance) is calculated for various information such as used reaction tanks, measuring instruments, workers, set temperature, outside air temperature and humidity, and the similarity space spanned by all or some attributes (items) The above performance data is associated with a two-dimensional, three-dimensional or one-dimensional map. Therefore, it is possible to easily extract a similar record or a dissimilar record regardless of whether the yield is any value.

In addition, the performance values scattered on each attribute are expressed by a plurality of representative points, and among the combinations that can be taken by the representative points of each attribute, there are a plurality of representative point patterns that are combinations having a high degree of similarity. Thereby, the tendency of the record data which can be taken and the characteristic of each record data by comparison between representative point patterns can be grasped.
Clustering is basically a subjective analysis, and the determination of representative points is largely dependent on the subjectivity, but it can be said that the criteria for classifying products, that is, threshold values, are determined by the determination of representative points. This threshold value can also be used to determine whether an abnormality or accident has occurred in a product, and grasp the structural relationship between objective variables and explanatory variables, which is not easy. Useful for.

  In addition, the performance data is linked on a two-dimensional or one-dimensional or three-dimensional map. Thereby, the search range and selection operation of similar performance data can be implemented through a simple user interface such as a scroll bar. That is, it is not necessary to use troublesome operations such as character string input or operations requiring prior knowledge.

  In addition, the clustering is performed after individually clustering various information (items) such as used reaction tanks, measuring instruments, working personnel, set temperature, outside air temperature and humidity. For this reason, the number of dimensions can be suppressed, and the influence of the curse of dimensions can be suppressed.

≪Others≫
The above embodiment is suitable for carrying out the present invention, but the form of implementation is not limited thereto, and various modifications can be made without departing from the scope of the present invention. It is.

  For example, in this embodiment, a plurality of representative points are obtained for each item, and a cluster is configured using a representative point pattern that is a combination of these points. However, all data may be analyzed collectively to form a cluster.

  Specifically, a technique called a self-organizing map can be applied to all performance data. It is possible and useful to apply the self-organizing map to the production information described in the present embodiment. However, for example, when the amount of production information to be acquired or used is small and the influence of the curse of the dimension can be ignored, this form can be taken.

  Moreover, the display part of the production information management apparatus of this embodiment is mounted with a touch panel, and the touch panel has all or part of the functions of the input part of the production information management apparatus. Thereby, the function by the search range setting means 1034 is realizable by operation of a touch panel. For example, the target results and the target cluster may be selected by tapping the corresponding part of the map displayed on the cluster display unit 950 on the touch panel. The area of interest can be selected by tracing outward a predetermined distance from the location where the target cluster is displayed. Further, by double-tapping, pinching in, or pinching out the display area of the cluster display unit 950, the map displayed there can be enlarged or reduced.

  In addition, it is possible to realize a technique in which various techniques described in this embodiment are appropriately combined.

  In addition, specific configurations of hardware, software, flowcharts, and the like can be appropriately changed without departing from the spirit of the present invention.

DESCRIPTION OF SYMBOLS 101 Production process 102 Clustering means 103 Map display means 1021 Item similarity evaluation means 1022 Item representative point determination means 1023 Actual data representative point pattern similarity evaluation means 1024 Similar achievement occurrence frequency management means 1031 Representative point pattern distance evaluation means 1032 Representative points Pattern map forming means 1033 Representative point pattern map managing means 1034 Search range setting means

Claims (4)

In a production information management device that manages product production information by including one or more items that characterize the production process of the product,
The storage unit of the production information management device stores performance data summarizing values for each item of the production information for each product,
The control unit of the production information management device is
With reference to the storage unit, two different products among all target products are identified, and control for calculating the similarity between the performance data of the two products for each item;
Based on the similarity between the calculated performance data of the two products, the performance data of all the target products is clustered, and one or more values representing the clustered performance data for each item are set for each item. Control determined as a representative point of
By calculating the similarity between the representative point pattern combining the representative points determined for each item and the result data for each item, the representative point pattern and the result data most similar to the representative pattern; Control to link
For each of the representative point patterns, a control for counting the number of performance data linked to the representative point pattern as a frequency,
Control for identifying two different representative point patterns among all the representative point patterns to be processed, and calculating the similarity between the two representative point patterns for each item;
Performing control for arranging all or part of the representative point pattern on a map having a three-dimensional structure or less based on the counted frequency and the calculated similarity between the two representative point patterns. Production information management device characterized by The map is a map composed of a set of clusters generated by the clustering,
The control unit of the production information management device is
In a display mode based on at least the similarity between the two representative point patterns, control for displaying the map on the display unit of the production information management device;
Control that designates a search range for searching corresponding performance data for clusters in the map when the display area of the display unit including the map is operated by an input from the input unit of the production information management device The production information management device according to claim 1, wherein: The map is a map set using all or a part of the items,
The control unit of the production information management device is
Control for calculating the frequency and similarity between the two representative point patterns for the items used when setting the map;
And a control for arranging all or part of the representative point patterns on the set map based on the counted frequency and the calculated similarity between the two representative point patterns. The production information management device according to any one of claims 1 and 2. In a production information management method in a production information management apparatus for managing production information of a product by including one or more items characterizing the production process of the product,
The storage unit of the production information management device stores performance data summarizing values for each item of the production information for each product,
The control unit of the production information management device is
Referring to the storage unit, identifying two different products among all the target products, and calculating the similarity between the performance data of the two products for each item;
Based on the similarity between the calculated performance data of the two products, the performance data of all the target products is clustered, and one or more values representing the clustered performance data for each item are set for each item. Determining as a representative point of
By calculating the similarity between the representative point pattern combining the representative points determined for each item and the result data for each item, the representative point pattern and the result data most similar to the representative pattern; A step of linking
For each representative point pattern, counting the number of performance data associated with the representative point pattern as a frequency;
Identifying two different representative point patterns among all the representative point patterns to be processed, and calculating the similarity between the two representative point patterns for each item;
Placing all or part of the representative point pattern on a map having a three-dimensional structure or less based on the counted frequency and the calculated similarity between the two representative point patterns. A production information management method characterized by

Images