JP5831363B2 - Manufacturing lead time prediction device, manufacturing lead time prediction method, and computer program - Google Patents

Description

  The present invention relates to a manufacturing lead time prediction apparatus, a manufacturing lead time prediction method, and a computer program, and in particular, a manufacturing lead time that is a period required for manufacturing a plurality of products having different manufacturing specifications through a plurality of steps. It is suitable for use for prediction.

  Product manufacturing processes in many industries, including the steel industry, require that products be manufactured appropriately according to the contents of orders from customers. For example, in the steel industry, it is required to deliver the required amount of products that meet the product order information including the product specifications (size, strength, surface coating, etc.) required by customers in the required amount. ing.

  In addition, with the diversification of customer needs, product order information, order quantity, and delivery date differ for each order, and the order quantity of each order has become small, so-called multi-product small-volume production is required. The tendency to become stronger. Therefore, in the product manufacturing process, it is required to manufacture a product that satisfies all product order information at a low cost. Furthermore, recently, it has become significant as one of the added value of products to be able to deliver products in a so-called “short construction period” in which the period from order to delivery is shorter than before.

  Under such circumstances, the manufacturing industry is required to have a manufacturing capability capable of withstanding these requirements without causing an increase in manufacturing costs. When drafting a production plan that meets these requirements, the desired date for starting the upstream process is usually determined by going back from the delivery date to the standard work period (standard lead time) for each order. While satisfying as much as possible, the manufacturing plan is drawn up in consideration of other planning conditions such as manufacturing lot and production capacity.

  However, in actual manufacturing factories, there are restrictions on the processing capacity of the processing process, so there are various operational factors such as waiting for processing due to the in-process of each process and failure of manufacturing equipment in the processing process. Therefore, even when manufacturing orders with the same product order information, the period required for actual manufacturing often varies. In this way, due to fluctuations in the work schedule, even in the above-mentioned standard work schedule, the production start target time of the upstream process is set by going back from the delivery time, and even if the production in the upstream process is started at the production start target time, the delivery time is set. Sometimes you can't protect.

  Further, in order to prevent such a delay in delivery time, in calculating the standard work schedule, in addition to the work schedule required for manufacturing, a surplus work schedule for absorbing the above-described fluctuations in the work schedule may be added. When the desired work date in the most upstream product manufacturing process is calculated using the standard work schedule including the spare work schedule in this way, the manufacturing is started at a timing considerably earlier than the actually required manufacturing work schedule. The early start of production results in in-process work in progress and inventory in the warehouse, and these in-process and stock increase lead time. In addition, an increase in handling load accompanying the occupation of the semi-finished product in the product storage area causes a hindrance to the manufacturing capability, resulting in a vicious cycle of extending the construction period. Therefore, it is desirable to accurately predict the manufacturing period.

Therefore, in Patent Document 1, the manufacturing apparatus is clustered based on the similarity of the pre-processing waiting time distribution of each manufacturing apparatus, the pre-processing waiting time representative model distribution of each cluster is calculated, and the sum of the representative model distribution is obtained. Therefore, a method for predicting the expected value and variance of the entire construction period has been proposed. In this patent document 1, the expected value and dispersion | distribution of the whole work period are estimated by superimposing the work period distribution for every manufacturing equipment using equipment use frequency information.
However, in a complex manufacturing factory, a process may be added after product inspection, and the number of times of equipment use is often unknown before manufacturing. Therefore, with the technique described in Patent Document 1, it is difficult to accurately predict the manufacturing period in such a case. Further, in a product manufacturing process with various types of products, manufacturing equipment that becomes a bottleneck varies depending on the product. Therefore, in the technique described in Patent Document 1, in such a case, there is a problem that the work period distribution for each manufacturing facility varies depending on the order configuration.

  In order to solve such a problem, Patent Document 2 uses a decision tree creation logic to create a passing process pattern (a combination of 0-1 information indicating whether each manufacturing process has passed) for each product, Based on the work schedule distribution calculated for each passing process pattern from past performance data, a work schedule capable of obtaining the target delivery date achievement rate is obtained, and this is set as the manufacturing work schedule.

Japanese Patent Laid-Open No. 2004-30088 JP 2010-128679 A

Kazunori Yamaguchi and two others, “Basics and Mechanism of Multivariate Analysis that can be understood well”, Hidekazu System, Inc., May 25, 2004, p.143-168

  However, in the technique described in Patent Document 2, an increase in the objective variable of the decision tree increases the structure of the model for determining the passing process pattern, which makes it difficult to manage. In the technique described in Patent Document 2, the passing process pattern is not considered at all without considering the distance between bits of the passing process pattern (for example, 010000 and 010001 are close, but 111011 and 000000 are distant). Because of the prediction, there is a problem that the prediction accuracy of the manufacturing construction period is deteriorated. Such a problem can be solved by Japanese Patent Application No. 2009-295661 filed by the present applicant. That is, as described in this patent document, such a problem can be solved by creating a decision tree that predicts the presence or absence of passage for each manufacturing process.

  However, in the technique described in Patent Document 2, since the manufacturing time period is calculated based on the work time distribution for each passing process pattern, there is a further problem different from the above-described problems. That is, in the technique described in Patent Document 2, when only a small number of products belonging to each passing process pattern can be obtained from the manufacturing performance data, the correct manufacturing time distribution cannot be expressed. There is a further problem that it varies greatly depending on how to take. This problem is still unsolved by the technique described in Japanese Patent Application No. 2009-295661.

  The present invention has been made in view of the problems as described above, and makes it possible to accurately predict the manufacturing period even if only a small number of products belonging to each passing process pattern are obtained from the manufacturing performance data. For the purpose.

Manufacturing period predicted equipment of the present invention, the order data containing the order-specific manufacturing specifications, a manufacturing period predicting device for predicting the construction period of the order, for each product to be manufactured through a plurality of manufacturing steps It is data indicating the manufacturing results, including the manufacturing specifications of the products, the presence / absence of passing results, which are information indicating the actual values of the passing / non-passing of each manufacturing process, and the information indicating the results of the construction period required to manufacture the products. Process-by-process schedule for deriving parameters that define the distribution of work schedules by manufacturing process using the manufacturing performance data acquisition means for acquiring manufacturing performance data including a certain actual work period, the presence / absence of actual results and the actual work period for each product. Using the parameters for defining the distribution of the construction period for each manufacturing process derived by the distribution deriving means, the construction period distribution deriving means for each process, and the presence / absence of the passing record for each product, the value of the passing record presence / absence is calculated. Each production process of the product using the production specification of the product included in the production performance data and the production schedule distribution by means of the passage process pattern for deriving the work schedule distribution by the passage process pattern obtained by arranging in a fixed order First type determining means for determining a product obtained by arranging predicted values of whether or not to pass in the predetermined order for each of the products included in the manufacturing performance data, and a product of each passing process pattern Using the number and the number of products by type, the type-specific passing process pattern generation rate deriving means for deriving the probability of occurrence of the passing process pattern by type, the passing process pattern period distribution, The production period model deriving means for deriving the period distribution for each type using the occurrence probability of the passing process pattern for each type, and the production period model included in the order data Using the specification, the predicted value of whether or not the order included in the order data passes through each manufacturing process is obtained, and the predicted value of whether or not the order passes through each manufacturing process is arranged in the predetermined order. Second type determining means for determining the type of each order included in the order data, the type determined by the second type determining means, and the type derived by the type-specific manufacturing period model deriving unit Using a construction period distribution of a type, and a manufacturing period derivation means for each type of input order for deriving a manufacturing period for each order, and the passage period derivation means for each passing process pattern passed in the passing process pattern As the probability density function determined based on the parameters obtained by adding the parameters defining the distribution of the construction period of the manufacturing process shown in FIG. Deriving the work period distribution for each over-process pattern , the first product type determination means receives the manufacturing specifications of the product, and outputs a predicted value of whether or not the product passes the manufacturing process. Applying the process pass determination logic creating means for creating the logic for each of the manufacturing processes, and the manufacturing specifications of the product to the process pass determination logic, the predicted value of the presence or absence of the passing of each manufacturing process of the product A first type creating unit that creates the type of the product by arranging the acquired predicted values in the predetermined order, and the second type determining unit includes the order type data in the order data Applying the included manufacturing specifications to the process passage determination logic, obtaining a predicted value of whether or not the order has passed, and arranging the obtained predicted values in the predetermined order to create the type of the order Characterized in that it further have a second breed creation means.

Manufacturing period predicted how the present invention, the order data containing the order-specific manufacturing specifications, a manufacturing period prediction method for predicting the construction period of the order, for each product to be manufactured through a plurality of manufacturing steps It is data indicating the manufacturing results, including the manufacturing specifications of the products, the presence / absence of passing results, which are information indicating the actual values of the passing / non-passing of each manufacturing process, and the information indicating the results of the construction period required to manufacture the products. A manufacturing result data acquisition process for acquiring manufacturing result data including a certain actual work period, and a process time period for deriving parameters defining the distribution of the work period for each manufacturing process using the presence / absence of actual results and the actual work period for each product. Using the parameters for defining the distribution of the construction period for each manufacturing process derived by the distribution derivation process, the construction period distribution derivation process for each process, and the presence / absence of the past performance for each product, Each production process of the product using the production specification of the product included in the production result data included in the production performance data, and the production schedule distribution derivation process by the production process data for deriving the work schedule distribution by the passage process pattern obtained by arranging in a fixed order A product type obtained by arranging predicted values of the presence or absence of passing in the predetermined order for each product included in the manufacturing performance data, and a product type for each passing process pattern Using the number and the number of products for each product type, a product-specific pass process pattern generation rate deriving step for deriving the probability of occurrence of the pass process pattern for each product type, the construction period distribution by the pass process pattern, and Using the occurrence probability of the passing process pattern for each product type, a process time model derivation process for each product type for deriving a work time distribution for each product type, and the production process included in the order data Using the specification, the predicted value of whether or not the order included in the order data passes through each manufacturing process is obtained, and the predicted value of whether or not the order passes through each manufacturing process is arranged in the predetermined order. A second type determining step for determining the type of each order included in the order data, a type determined by the second type determining step, and a type derived by the type-specific manufacturing period model deriving step A manufacturing time derivation process for each type of input order that derives a manufacturing time period for each order using a type of work time distribution, and the process time distribution derivation process for each passing process pattern has passed in the passing process pattern As the probability density function determined based on the parameters obtained by adding the parameters defining the distribution of the construction period of the manufacturing process shown in FIG. Deriving the work period distribution for each over-process pattern , and the first product type determination step receives the manufacturing specifications of the product, and outputs a predicted value of whether or not the product passes the manufacturing process. Applying a process pass determination logic creation process for creating the logic for each of the manufacturing processes and the manufacturing specifications of the product to the process pass determination logic, and predicting the presence or absence of passing of each manufacturing process of the product A first product type creation step of creating a product type by arranging the obtained predicted values in the predetermined order, and the second product type determination step includes adding the order type data to the order data. Applying the included manufacturing specifications to the process passage determination logic, obtaining a predicted value of whether or not the order has passed, and arranging the obtained predicted values in the predetermined order to create the type of the order Characterized in that it further have a second breed creation process.

  The computer program of the present invention causes a computer to execute each step of the manufacturing period prediction method.

According to the present invention, the parameters of the distribution of the manufacturing process indicated to have passed in the passing process pattern obtained by arranging the values of the passing results included in the manufacturing result data in a predetermined order are added together. As a probability density function determined on the basis of the parameters obtained in this way, a work period distribution by passing process pattern is derived.
Then, passing the varieties obtained by arranging the predictive value of the presence or absence of the passage of the manufacturing process of the products included in the manufacturing performance data in the predetermined order of specific breeds determined for each of the product step pattern The generation period distribution for each product type is derived using the execution period distribution for each passing process pattern and the generation probability of the passing process pattern for each product type. Then, an order type included in the order data is determined for each order, and the manufacturing period for each order is derived using the determined type and the distribution of the period for each type.
Therefore, it is possible to calculate the passing step pattern another implementation schedule distribution without relying largely on the number of products in each passage step pattern. Therefore, even if only a small number of products belonging to each passing process pattern are obtained from the manufacturing performance data, the manufacturing work period can be accurately predicted.

It is a figure which shows the 1st Embodiment of this invention and shows an example of a functional structure of a manufacturing construction period prediction apparatus. It is a figure which shows the 1st Embodiment of this invention and shows an example of manufacture performance data. It is a figure which shows the 1st Embodiment of this invention and shows an example of the process passage determination logic of a certain process. It is a figure which shows the 1st Embodiment of this invention and shows an example of the passage presence variable according to manufacture number, and a kind. It is a figure which shows the 1st Embodiment of this invention and shows an example of a passage process pattern generation rate. It is a figure which shows the 1st Embodiment of this invention and shows an example of the work period distribution according to a process. It is a figure which shows the 1st Embodiment of this invention and shows an example of construction period distribution according to passage process pattern. It is a figure which shows the 1st Embodiment of this invention and shows an example of the manufacturing period model according to a kind. It is a figure which shows the 1st Embodiment of this invention and shows an example of order data. It is a figure which shows the 1st Embodiment of this invention and shows an example of the passage presence-absence variable and kind according to order. It is a figure which shows the 1st Embodiment of this invention and shows an example of a target assortment achievement rate. It is a figure which shows the 1st Embodiment of this invention and shows an example of the manufacturing period according to an order. It is a flowchart which shows the 1st Embodiment of this invention and demonstrates an example of a process of a manufacturing construction period prediction apparatus. FIG. 14 is a flowchart illustrating the first embodiment of the present invention and continuing from FIG. 13. It is a figure which shows the 2nd Embodiment of this invention and shows an example of a functional structure of a manufacturing construction period prediction apparatus. It is a figure which shows the 2nd Embodiment of this invention and shows an example of the kind determination logic. It is a figure which shows the 2nd Embodiment of this invention and shows an example of the kind according to manufacture number. It is a figure which shows the 2nd Embodiment of this invention and shows an example of the kind according to order. It is a figure which shows a 1st comparative example and shows the functional structure of a manufacturing construction period prediction apparatus. It is a figure which shows a 1st comparative example and shows the number of products according to the results construction period according to passage process pattern. It is a figure which shows a 1st comparative example and shows the construction period distribution according to passage process pattern. It is a figure which shows a 2nd comparative example and shows the functional structure of a manufacturing construction period prediction apparatus. It is a figure which compares and shows the result by embodiment and a comparative example by a table type. It is a figure which shows an example of the relationship between the value which subtracted the actual work period from the manufacture work period of evaluation data, and the weight of a product. It is a figure which shows the modification of the 1st, 2nd embodiment of this invention, and shows an example of a functional structure of a manufacturing construction period prediction apparatus. It is a figure which shows an example of the work period distribution according to process calculated | required with the method of 1st Embodiment. It is a figure which shows an example of the work period distribution according to the process calculated | required with the method of 4th Embodiment. It is a figure which shows an example of the work period distribution according to process calculated | required with the method of 5th Embodiment. It is a figure which shows an example of the work period distribution according to process calculated | required with the method of 6th Embodiment. It is a figure which shows an example of the determination coefficient of a single regression equation. It is a figure which shows an example of the coefficient of a single regression equation.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
<< First Embodiment >>
FIG. 1 is a diagram illustrating an example of a functional configuration of the manufacturing work period prediction apparatus 100. The hardware of the manufacturing period prediction device 100 can be realized by using, for example, a known information processing device including a CPU, a ROM, a RAM, an HDD, and various interfaces, and a detailed description thereof will be given here. Omitted.
In the present embodiment, a case where a manufacturing work period of a steel product manufactured at a manufacturing factory (for example, a thick plate manufacturing factory) in the steel industry is predicted will be described as an example.

[Manufacturing result data acquisition unit 111]
The manufacturing performance data acquisition unit 111 acquires manufacturing performance data from the outside and stores it. For example, the manufacturing performance data acquisition unit 111 acquires manufacturing performance data according to a user interface input operation by a user, reads manufacturing performance data stored in a removable storage medium, or manufactures data from an external device via a communication line. By receiving data, manufacturing performance data can be acquired.

FIG. 2 is a diagram illustrating an example of the manufacturing result data 200.
In FIG. 2, the manufacturing performance data 200 includes a product number (product No.), manufacturing specifications, presence / absence of passing results, and actual work period.
The product number is a number for identifying a product.
The manufacturing specification indicates the attribute of the product. For example, the product specification of the product with the product number “0001” has a product (steel plate) thickness of 30 mm, a product plate width of 600 mm, and a product weight of 3.5 ton. Is A, and the cutting method when manufacturing the product is a. Here, the case where the plate thickness, the plate width, the weight, the manufacturing method, and the cutting method are manufacturing specifications is shown as an example, but the contents of the manufacturing specification differ depending on the product.

The presence / absence of passing results is for identifying the manufacturing process through which the product identified by the product number has passed and the manufacturing process that has not passed through for each of the target manufacturing processes (for example, included in the manufacturing factory). Is. In the example shown in FIG. 2, “1” is attached to the manufacturing process through which the product has passed, and “0” is attached to the manufacturing process that has not passed through. For example, it is determined from the presence / absence of passing results that the product number “0002” is manufactured through the manufacturing process of step 10 without passing through the manufacturing processes of process 1 and process 2. The Here, since the process 10 is the receipt of the product to the warehouse, the value of the presence / absence of passing results for the process 10 is always “1” (a bit is set).
Here, the product that passes through the manufacturing process is not limited to the product itself, but includes a raw material of the product and a semi-finished product in the middle stage before the product is completed. In addition, passing through a manufacturing process means that processing (for example, heat treatment, processing, storage) to be performed in the manufacturing process is performed.
Although the case where the number of manufacturing steps is “10” is shown as an example, the number of manufacturing steps is not limited to “10”. In the following description, “manufacturing process” is abbreviated as “process” as necessary.

The actual work period is the entire work period required to manufacture the product identified by the product number. For example, the actual construction period of the product with the product number “0001” is 5 days.
The manufacturing performance data acquisition unit 111 can be realized by using, for example, a CPU, ROM, RAM, HDD, and various interfaces.

[Process Pass Determination Logic Creation Unit 112]
The process passage determination logic creation unit 112 creates process passage determination logic that predicts, for each process, a process through which a product passes and a process through which the product does not pass based on a manufacturing specification included in the manufacturing result data 200. To do. In the present embodiment, a case where a decision tree is employed as the process passage determination logic will be described as an example.
FIG. 3 is a diagram illustrating an example of a process pass determination logic in a certain process (for example, process 1). As shown in FIG. 3, in the present embodiment, the process passage determination logic creation unit 112 uses a manufacturing specification included in the manufacturing performance data 200 as an explanatory variable, and shows a predicted value of whether or not each process passes. Create a decision tree with existence variables as objective variables. Here, it means that a product whose pass / fail variable in a certain process is “1” is predicted to pass the process (the product is likely to pass through the process), and a product that is “0” It means that it is predicted that the process will not be passed (the product is difficult to pass through the process).

  A decision tree is one of data analysis methods, and is an analysis method for classifying data like branches and leaves of trees according to various conditions. The decision tree is used for identifying the cause of manufacturing defects, classifying market information, and the like (see, for example, Non-Patent Document 1). The decision tree is composed of multiple nodes that are a cluster of data, and starts from the root node (root node) that represents the entire data, and the proportion of data that has specific attributes at the end nodes (leaf nodes, leaf nodes) In order to increase the number of nodes, that is, in order to include biased data, the nodes are created while branching one after another. The decision tree can be used for various predictions by using the obtained branch condition to the leaf node and past data (learning data) belonging to the leaf node. Here, the attribute to be predicted is referred to as “object variable”, and the attribute describing the data branch condition is referred to as “explanatory variable”. When creating a decision tree, design parameter settings such as how to define objective variables and explanatory variables and how to determine the size of the decision tree (number of nodes and depth) are obtained. This is extremely important because it is deeply related to the prediction accuracy and ease of handling of the decision tree.

Here, as design parameters, the number of leaf nodes of the decision tree to be created, the depth of the tree structure, and the like are set as parameters relating to the structure of the decision tree. Specifically, an upper limit value of the number of data held by one leaf node is given. If this upper limit value is reduced, the number of leaf nodes increases, that is, the decision tree becomes larger and deeper. More specifically, since the prediction accuracy was not improved even when the upper limit value of the number of data was increased to 100 or more, the upper limit value of the number of data was set to 100.
The process passage determination logic creation unit 112 determines a decision tree 300 as shown in FIG. 3 based on the design parameters thus obtained, the manufacturing specifications included in the manufacturing result data 200, explanatory variables, and objective variables. Created for each process by tree creation algorithm. In this embodiment, since the number of processes is “10”, ten decision trees 300 as shown in FIG. 3 are created.

If the decision tree 300 shown in FIG. 3 is a decision tree for the process 1, for example, the manufacturing method is B, the weight is 4.0 ton (≧ 3.2 ton), and the plate width is 700 mm (<1500 mm). The passing / non-passing variable of the product of the manufacturing specification is classified as “1”, so that it is predicted that the product passes the process 1 (it is easy to pass the process 1).
The process passage determination logic creation unit 112 can be realized by using, for example, a CPU, a ROM, and a RAM.

[Process passing logic storage unit 113]
The process passage logic storage unit 113 stores the decision tree 300 created by the process passage determination logic creation unit 112.
The process passing logic storage unit 113 can be realized by using, for example, an HDD.

[First kind creation unit 114]
The first product creation unit 114 applies the manufacturing specifications included in the manufacturing performance data 200 to the decision tree 300 stored in the process pass logic storage unit 113, and each of the products included in the manufacturing performance data 200. , The predicted value (passage presence / absence variable (= “1” or “0”)) of each process is obtained, and information obtained by combining the passage presence / absence variables is created as the product type information.

FIG. 4 is a diagram showing an example of the passage presence / absence variable and product type for each production number obtained from the production performance data 200.
In FIG. 4, by applying the manufacturing specifications included in the manufacturing performance data 200 to the decision tree 300 as shown in FIG. 3, a pass / fail variable 401 for each process is obtained for each product. The product 402 is created for each product by arranging the values of the pass / fail variable 401 for each process so that the higher-order bit is for the pass / non-pass variable 401 for the process with a smaller number. Comparing the result shown in FIG. 4 with the result shown in FIG. 2, for example, the result of passing through the process 1 of the product with the production number “0001” is “1” (see FIG. 2), but the same The pass prediction of the product with the production number “0001” in the process 1 is “0” (see FIG. 4). Thus, a prediction error may occur in the statistical method.
The first product type creation unit 114 can be realized by using, for example, a CPU, a ROM, and a RAM.

[Passage pattern generation rate creation unit 115 by product type]
The type-specific passing process pattern occurrence rate creation unit 115 calculates the passing process pattern occurrence rate for each type based on the manufacturing performance data 200.
The passing process pattern occurrence rate refers to the number of products belonging to the passing process pattern (actual value thereof) in which the presence / absence of passing results in the manufacturing result data 200 is arranged in order from the process having the smallest number, Divided by the number of products belonging to. For example, suppose that there are 9500 products with a passing process pattern (actual value) of “0101010000” and 10,000 products with a product type determined to be “0101010000”. In this case, the occurrence rate of the passing process pattern “0101010000” of the product having the product type “0101010000” is 0.95 (= 9500/10000).
FIG. 5 is a diagram illustrating an example of a passing process pattern occurrence rate 500 for each product type. In FIG. 5, the sum of the passing process pattern occurrence rates (sum of passing process pattern occurrence rates in each row in FIG. 5) for a certain product type is “1”.
The product-specific pass process pattern generation rate creation unit 115 can be realized by using, for example, a CPU, a ROM, and a RAM.

[Schedule distribution creation unit by process 116]
The process-specific work period distribution creating unit 116 calculates the work period-specific work period distribution using the manufacturing result data 200 (passing result presence / absence and result work period). The work period distribution is represented by a histogram or an approximation function (distribution function). In the present embodiment, a case where the work period distribution is represented by a normal distribution will be described as an example. Below, an example of the preparation method of distribution schedule according to a process is demonstrated.

First, symbols are defined as follows.
(A) The actual construction period of the product i is t _i and the presence / absence of the passing process of the process j is p _ij .
(B) The passing process pattern (the actual value) of the product i is described as p _i = (p _i1 , p _i2 ,..., P _iM ). Here, M is the number of steps, and in the example shown in FIG.
(C) The actual construction period of all products is t = (t ₁ , t ₂ ,..., T _N ) ^T. Here, N represents the total number of products, and T represents a transposed matrix.
(D) Let P = {p _ij } be the passing process pattern (the actual value) of all products.
(E) The work period distribution of the process j is assumed to be a normal distribution having an average μ _j and a variance v _j (= σ _j ² ). Here, the average μ _j and the variance v _j are the determining variables of the work period distribution by process.
(F) Actual construction period t _i shall be in accordance with the normal distribution of each passage step pattern (actual value).
(G) The average value of the construction period distribution of all processes is μ = (μ ₁ , μ ₂ ,..., Μ _M ), and the variance is v = (v ₁ , v ₂ ,..., V _M ). .
Also here
(H) The work schedule distribution of each process shall be independent of the work schedule distribution of other processes (that is, covariance shall not be considered).

Then, likelihood p About product i proven construction period t _i obtained (t _i | p _i, mu, v) is expressed by the following equation (1) and (2) below.

(1) of the left side, p _i as the passing step pattern (actual value) of the product i is, mu as the average construction period distribution of all process, v is given respectively as a dispersion construction period distribution of the entire process It represents the likelihood that actual construction period t _i is obtained when. Meanwhile, (1) the right side of the equation, the actual construction period t _i of the product i represents the average occurs according mu _i ~, dispersion is the v _i ~ normal distribution N. Thus, equation (1), likelihood p About product i proven construction period t _i obtained (t _i | p _i, mu, v) of the distribution, average mu _i ~, a variance v _i ~ normal It represents following the distribution N (μ _i ˜, v _i ˜).
Here, the average μ _i ˜ represents the sum of the average μ _j of the work period distribution of the process j through which the product i has passed. The dispersibility v _i ~ represents the sum of the variance v _j construction period distribution step j of product i has passed.

  From the equations (1) and (2), the likelihood p (t | P, μ, v) at which the actual work period t of all products is obtained is expressed by the following equation (3).

(3) formula, the likelihood p actual results construction period t of all products can be obtained (t | P, μ, v ) is, the actual construction period t ₁ of all products i, t _2, ···, likelihood of t _N It is expressed by the product of degrees.
From the equation (3), the log likelihood is expressed by the following equation (4).

In order to maximize the log likelihood of the equation (4), the average μ _j and the variance v _j that minimize the evaluation function J represented by the following equation (5) may be obtained.

(5) the evaluation function J of the equation is the evaluation function actual construction period t _1, t ₂ of all products i, · · ·, the higher the product of the likelihoods of t _N is greater value decreases. The average μ _j and the variance v _j may be calculated using the expression (5) as an evaluation function. However, if the value of N, which is the number of products, is large, the optimization calculation of the expression (5) becomes difficult. . Therefore, in this embodiment, the evaluation function of equation (5) is rewritten as follows.
Here, the following symbols are defined.
(I) Let K be the number of types of passing process patterns.
(J) Let the passing process pattern k (k = 1, 2,..., K) be q _k = (q _k1 , q _k2 ,..., Q _kM ).
(K) Let N _k ^ be the number of products belonging to the passing process pattern k.
(L) The mean μ _k ^ and variance v _k ^ of the work period distribution of the passing process pattern k are expressed by the following equation (6).

Here, the average μ _k ^ represents the sum of the average μ _j of the process distribution of the process j indicated as having passed in the passing process pattern k. The variance v _k ^ represents the sum of the variance v _j of the process distribution of the process j shown to have passed in the passing process pattern k.
From the above (i) to (l), the evaluation function J of the equation (5) can be rewritten to the evaluation function J represented by the following equation (7).

The evaluation function J shown in the equation (7) is expressed by the sum of the terms of the number K of types of passing process patterns. Further, in the equation (7), N _k ^ of the first term and the fourth term on the right side, and the integration (Σ (t _i ) ² , Σt _i ) of the second term and the fourth term on the right side, It can be calculated by counting. Therefore, it is easier to perform the optimization calculation using the evaluation function J shown in Expression (7) than to perform the optimization calculation using the evaluation function J of Expression (5). This optimization calculation can be executed by using a known nonlinear optimization calculation method such as downhill simplex method or quasi-Newton method. The process-by-process construction period distribution creating unit 116 gives the average μ _j and the variance v _j by performing the optimization calculation by giving the production result data 200 (passing result presence / absence and actual work period) to the expressions (6) and (7). Is calculated as an example of the parameter of the work period distribution by process.

FIG. 6 is a diagram showing an example of the work period distribution 600 calculated as described above. Here, optimization calculation was performed using the downhill simplex method of Nelder = Mean. In FIG. 6, the standard deviation σ _j is represented by the square root (= √v _j ) of the variance v _j .
The process-specific work schedule distribution creation unit 116 can be realized by using, for example, a CPU, a ROM, and a RAM.

[Construction distribution creation unit 117 by passing process pattern]
The passing process pattern-by-process period distribution creation unit 117 calculates the passing process pattern-by-process pattern distribution by using the process-by-process schedule distribution 600 created by the process.
First, the average μ _k ^ and standard deviation σ _k ^ of the process distribution by passage process pattern are calculated based on the process distribution 600 by process from the relationship of the expression (6) (the value 701 before conversion in FIG. 7). Column).
For example, since a product with a passing process pattern (actual value) of “0100010001” has passed through the process 2, the process 6, and the process 10, the average μ _k ^ and the standard deviation σ of the construction period distribution of the passing process pattern _k ^ is expressed by the following equations (8) and (9).

μ _k ^ = 25.5 + 20.1 + 0.4 = 46.0 (8)
σ _k ^ = sqrt (14.7 ² +8.0 ² +1.4 ² ) = 16.8 (9)
Next, the construction period distribution creation unit 117 by passing process pattern assumes that the construction period distribution is logarithmic normal distribution, logarithmic normal whose mean is μ _k ^ and variance is v _k ^ (= σ _k ^ ² ). Deriving the distribution. Here, the log-normal distribution parameters μ _k ^ ′ and v _k ^ ′ having an average μ _k ^ and variance v _k ^ are calculated using the following equations (10) and (11). The

The logarithmic normal distribution parameters μ _k ^ ′ and σ _k ^ ′ (= √v _k ^ ′) thus _obtained are shown in the column of the converted value 702 in FIG.
In the present embodiment, the construction period distribution by passing process pattern is expressed by the following equation (12).

  However, in the present embodiment, the formula (12) is discretized as shown in the formula (13), thereby obtaining a passing process pattern work period distribution PΔ (tΔ | k). That is, the passing process pattern-specific work period distribution creating unit 117 calculates the passing process pattern-specific work period distribution PΔ (tΔ | k) by calculating the following equation (13). The passage process pattern distribution period PΔ (tΔ | k) obtained in this way is shown in the column of the process period distribution 703 by pass process pattern in FIG. The construction period distribution PΔ (tΔ | k) by passing process pattern means the probability that the construction period of the product of the passing process pattern k is the construction period tΔ = (0, 1, 2,...). In addition, in the formula (13), the case where the probability density function of the formula (12) is discretized in units of one day is described as an example, but the discretization method is limited to such a method. It is not a thing.

  The passing process pattern-specific work period distribution creation unit 117 can be realized by using, for example, a CPU, a ROM, and a RAM.

[Production period model creation unit 118 by product type]
The production period model creation unit 118 for each product type includes a pass process pattern generation rate 500 (see FIG. 5) for each product type generated by the product-by-product pass process pattern generation rate creation unit 115 and a process term distribution creation unit 117 for each passing process pattern. The production period model for each product type is created using the process period distribution PΔ (tΔ | k) for each passing process pattern created in the above. In the present embodiment, the probability PΔ (tΔ | l) that the work period of a product having a product type l is tΔ is calculated as a production work model by product type according to the following equation (14).

In equation (14), rate _lk is the passing process pattern occurrence rate 500 of the passing process pattern k of the product type l.
Equation (14) is the probability (passage rate) of the passing process pattern k of the product type l calculated from the manufacturing result data 200 to the probability PΔ (tΔ | k) that the passing process pattern of the product of the product type 1 is k. The product obtained by multiplying the process pattern occurrence rate 500) for all the passing process patterns k indicates that it is a production period model for each product type.
FIG. 8 is a diagram showing an example of the production period model 800 classified by product type obtained as described above. The values shown in the respective columns in FIG. 8 are the values of the left side (PΔ (tΔ | l)) of the equation (14).
The type-specific manufacturing period model creation unit 118 can be realized by using, for example, a CPU, a ROM, and a RAM.

[Production period model storage unit 119 by product type]
The type-specific manufacturing period model storage unit 119 stores the type-specific manufacturing period model 800 created by the type-specific manufacturing period model creation unit 118.
The type-by-product manufacturing period model storage unit 119 can be realized by using, for example, an HDD.

[Order Data Acquisition Unit 120]
The order data acquisition unit 120 acquires and stores order data from the outside. For example, the order data acquisition unit 120 acquires order data according to a user interface input operation by a user, reads order data stored in a removable storage medium, or receives order data from an external device via a communication line. Order data can be acquired.

FIG. 9 is a diagram illustrating an example of order data 900.
In FIG. 9, order data 900 includes an order number (order No.) and manufacturing specifications.
The order number is a number for identifying an order.
The manufacturing specification indicates an attribute of the product, and has the same attribute (item) as the manufacturing specification in the manufacturing performance data 200 shown in FIG. Since the contents of each manufacturing specification shown in FIG. 9 are as described above, detailed description thereof is omitted here.
The order data acquisition unit 120 can be realized by using, for example, a CPU, ROM, RAM, HDD, and various interfaces.

[Process Pass Determination Unit 121]
The process pass determination unit 121 applies the manufacturing specifications included in the order data 900 to the decision tree 300 (see FIG. 3) stored in the process pass logic storage unit 113, and passes through each process. The ease (passing presence / absence variable (= “1” or “0”)) is determined for each order.
The process passage determination unit 121 can be realized by using, for example, a CPU, a ROM, and a RAM.

[Second type creation unit 122]
The 2nd kind creation part 122 arranges the value of the passage existence variable obtained by process passage judgment part 121 so that it may become an upper bit, so that it may be a passage existence variable of a process with a small number for every order. As a result, varieties for each order are created.
FIG. 10 is a diagram illustrating an example of the pass / non-passage variable 1001 and the product type 1002 for each order obtained from the order data 900. Note that the description of the method of creating the product has already been given in the description of the first product creation unit 114, and therefore the detailed description thereof is omitted here.
The 2nd kind creation part 122 is realizable by using CPU, ROM, and RAM, for example.

[Target assortment achievement rate setting unit 123]
The target assortment achievement rate setting unit 123 sets a target assortment achievement rate. The target assortment achievement rate is the ratio of the quantity of products that can be delivered (can be shipped) by the delivery date to the order quantity. In the present embodiment, since the product is a steel product, for example, the target assortment achievement rate can be determined by the ratio of the weight of the product that will be delivered by the delivery date with respect to the weight of the order. The value of the target assortment achievement rate is determined according to the user's desire. In the present embodiment, the target assortment achievement rate setting unit 123 identifies the value of the target assortment achievement rate based on the operation of the user interface by the user, and stores the value of the target assortment achievement rate. Set target assortment achievement rate.

FIG. 11 is a diagram illustrating an example of the target assortment achievement rate 1100. As shown in FIG. 11, here, 0.95 is uniformly set as the target assortment achievement rate regardless of the type or order. However, the target assortment achievement rate may be set for each of at least one of each product type and each order, for example.
The target assortment achievement rate setting unit 123 can be realized by using, for example, a CPU, a ROM, and a RAM.

[Production period calculation unit 124 by type input order]
The production period calculation unit 124 by type input order includes the production period model 800 (see FIG. 8) stored in the type production period model storage unit 119 and the order created by the second type creation unit 122. The construction period t is calculated for each order using another product type 1002 (see FIG. 10) and the target assortment achievement rate 1100 (see FIG. 11) set by the target assortment achievement rate setting unit 123.
In the present embodiment, the manufacturing time calculation unit 124 for each type input order calculates the time t satisfying the following expression (15) for each order (for each type corresponding to the order).

In the equation (15), α is a target assortment achievement rate [−]. Further, as described above, PΔ (tΔ | l) is the probability [−] that the work period of the product having the product type l is tΔ, and the value of each column of the product-specific manufacturing work model 800 shown in FIG. It is. Expression (15) represents that the minimum number of days required for assembling the target assortment achievement rate α is obtained as the construction period t.
As described above, in the present embodiment, the case where the construction period t for each order is calculated using the production period model 800 for each kind, the kind for each order, and the target assortment achievement rate 1100 will be described as an example. did. However, this is not necessarily required if the manufacturing period t for each order is calculated using the manufacturing period model 800 for each type and the types for each order. For example, assuming that PΔ (tΔ | l) is a normal distribution, using the average μ · standard deviation σ of PΔ (tΔ | l) and a constant x set by the user, the following (16) The work period t may be calculated for each order (for each type corresponding to the order) as in the equation.
PΔ (tΔ | l) = μ + x × σ (16)
FIG. 12 is a diagram illustrating an example of an order-by-order manufacturing period 1200 that is a period for each order.
The manufacturing period calculation unit 124 for each type of input order can be realized by using, for example, a CPU, a ROM, and a RAM. The value described in each column of FIG. 12 is the work period t for each order.

[Manufacturing period storage unit 125 by type input order]
The manufacturing period storage unit 125 by type input order stores the manufacturing period 1200 by order calculated by the manufacturing period calculation unit 124 by type input order (see FIG. 12).
The manufacturing period storage unit 125 according to product type input order can be realized by using, for example, an HDD.
[Production period output unit 126 by product type input order]
The production period output unit 126 according to type input order outputs information of the manufacturing period 1200 according to order stored in the manufacturing period storage unit 125 according to type input order. More specifically, for example, the type-by-order-by-order manufacturing period output unit 126 displays at least one of the information on the order-specific manufacturing period 1200, which is displayed on a display device, transmitted to an external device, and stored in a removable storage medium. Do one. The output of the order-specific manufacturing period 1200 may be performed based on a user interface operation by the user, or may be automatically performed at a preset timing.
The production time output unit 126 according to product type input order can be realized, for example, by using a CPU, ROM, RAM, HDD, and various interfaces.

[Operation flowchart]
Next, an example of processing of the manufacturing work period prediction apparatus 100 will be described with reference to the flowcharts of FIGS. 13 and 14.
First, in step S1301, the manufacturing performance data acquisition unit 111 stands by until the manufacturing performance data 200 is acquired. And if the manufacturing performance data 200 is acquired, it will progress to step S1302. In step S1302, the process passage determination logic creation unit 112 creates a decision tree 300 as shown in FIG. 3 for each process using a decision tree creation algorithm.

Next, in step S1303, the process passage logic storage unit 113 stores the decision tree 300 (for each process) created in step S1302.
Next, in step S1304, the first product type creation unit 114 applies the manufacturing specifications included in the manufacturing performance data 200 to the decision tree 300, and performs each process for each product included in the manufacturing performance data 200. A predicted value (passage presence / absence variable 401) of the presence / absence of passage is obtained, and information obtained by combining the passage presence / absence variables is created as information on the type 402 of the product.

Next, in step S1305, the product-specific pass process pattern occurrence rate creation unit 115 creates products that belong to the pass process pattern (actual value) in which the pass record presence / absence of the manufacturing record data 200 is arranged in order from the process with the smallest number. By dividing the number by the number of products belonging to the same type as the passing process pattern, the passing process pattern occurrence rate 500 for each type is calculated.
Next, in step S1306, the process-by-process construction period distribution creation unit 116 performs the optimization calculation by giving the production performance data 200 (passing performance presence / absence and performance construction period) to the expressions (6) and (7). A construction period distribution 600 (average μ _j , variance v _j ) by process is obtained.

Next, in step S 1307, the passing process pattern-specific work period distribution creating unit 117 calculates the average μ _k ^ and the standard deviation σ _k ^ of the passing process pattern-specific work distribution based on the process-specific work schedule distribution 600 (see FIG. Logarithm normal distribution parameters μ _k ^ ′, v _k ^ ′ (= σ _k ^ ′ ² ) expressed using mean μ _k ^ and variance v _k ^).
(Refer to the column of the converted value 702 in FIG. 7), the calculation of the formula (12) and the formula (13) is performed to calculate the work period distribution PΔ (tΔ | k) by the passing process pattern (see FIG. 7). 7 (refer to the column of construction period distribution 703 by passage process pattern 7).

Next, in step S1308, the type-specific manufacturing period model creation unit 118 uses the passing process pattern occurrence rate 500 for each type and the passing process pattern-specific period distribution PΔ (tΔ | k), The calculation is performed, and the probability PΔ (tΔ | l) that the construction period of the product having the product type l is tΔ is calculated as the production period model 800 for each product type.
Next, in step S1309, the type-specific manufacturing period model storage unit 119 stores the type-specific manufacturing period model 800 created in step S1308.

  Next, in step S1401 of FIG. 14, the order data acquisition unit 120 waits until the order data 900 is acquired. When the order data 900 is acquired, the process proceeds to step S402. In step S1402, the process passage determination unit 121 applies the manufacturing specification included in the order data 900 to the decision tree 300, and sets a predicted value (passage presence / absence variable) of each process for each order. Judgment.

Next, in step S1403, the second type creation unit 122 arranges the values of the passage presence / absence variables obtained in step S1402 so that the higher the number of the passage passage presence / absence variables of the process having a smaller number, the higher the bits. By performing for each order, a variety 1002 for each order is created.
Next, in step S1404, the target assortment achievement rate setting unit 123 sets a target assortment achievement rate 1100.
Next, in step S1405, the type-by-order-based manufacturing period calculation unit 124 satisfies the formula (15) using the type-specific manufacturing period model 800, the order-specific types, and the target assortment achievement rate 1100. The work period t is calculated for each order (for each type corresponding to the order). This calculation result is an order-specific manufacturing period 1200.

Next, in step S1406, the manufacturing input period for each type input order storage unit 125 stores the manufacturing period for each order 1200 calculated in step S1405.
Finally, in step S <b> 1407, the type input order manufacturing manufacturing period output unit 126 outputs information on the order manufacturing manufacturing period 1200. As described above, the processing in step S1407 may be performed when an instruction is given from the user.

<Second Embodiment>
Next, a second embodiment of the present invention will be described. The present embodiment and the first embodiment are mainly different in the type determination logic and the process of determining the type for each order using the type determination logic from the order data. Therefore, in the description of the present embodiment, the same parts as those in the first embodiment described above are denoted by the same reference numerals as those in FIGS.

FIG. 15 is a diagram illustrating a functional configuration of the manufacturing work period prediction apparatus 1500.
[Product type determination logic creation unit 1512]
The product type determination logic creating unit 1512 creates product type determination logic for predicting a passing process pattern (product type) from the manufacturing specifications included in the manufacturing performance data 200. The product type determination logic creation unit 1512 corresponds to the process passage determination logic creation unit 112.
In this embodiment, a decision tree is adopted as the kind determination logic.

FIG. 16 is a diagram illustrating an example of the product type determination logic. As shown in FIG. 16, the product type determination logic creating unit 1512 creates a decision tree 1600 having the manufacturing specifications included in the manufacturing performance data 200 as explanatory variables and the passing process pattern as an objective variable. In the present embodiment, the process passage determination logic creation unit 112 creates the decision tree 300 for each process. On the other hand, in this embodiment, the product type determination logic creation unit 1512 creates one decision tree 1600.
The product type determination logic creation unit 1512 can be realized by using, for example, a CPU, a ROM, and a RAM.

[Product type determination logic storage unit 1513]
The type determination logic storage unit 1513 stores the decision tree 1600 created by the type determination logic creation unit 1512.
The product type determination logic storage unit 1513 can be realized by using, for example, an HDD.
FIG. 17 is a diagram showing an example of a product type by production number obtained from the production performance data 200.
The product type 402 shown in FIG. 4 and the product type 1700 shown in FIG. 17 are obtained from the same manufacturing performance data 200, but have different values. This is because the logic for obtaining these varieties 402 and 1700 is different.

[Product type determination unit 1521]
The product type determination unit 1521 applies the manufacturing specifications included in the order data 900 to the decision tree 1600 (see FIG. 16) stored in the product type determination logic storage unit 1513 to create a product type for each order. .
FIG. 18 is a diagram illustrating an example of product types for each order obtained from the order data 900. In this embodiment, since the objective variable of the decision tree 1600 is a passing process pattern, the product type determination unit 1521 applies the manufacturing specifications included in the order data 900 to the decision tree 1600 to directly select the product type. Ask for.
The product type determination unit 1521 corresponds to the process pass determination unit 121 and the second product type creation unit 122 shown in FIG.
The product type determination unit 1521 can be realized by using, for example, a CPU, a ROM, and a RAM.

<< First Comparative Example >>
Next, a first comparative example will be described.
FIG. 19 is a diagram illustrating a functional configuration of a manufacturing work period prediction apparatus 1900 according to this comparative example. In FIG. 19, the same processing units as those shown in FIGS. 1 and 15 are denoted by the same reference numerals as those shown in FIGS. 1 and 15, and detailed description thereof is omitted.

  The second point of the manufacturing work period prediction apparatus 1900 of this comparative example is that, instead of the process-specific work period distribution creating part 116 and the process-by-process pattern work period distribution creating part 117, a process-by-process pattern work period distribution creating part 1916 is used. It differs from the manufacturing work period prediction apparatus 1500 of the embodiment.

[Transmission process pattern-specific work period distribution creation unit 1916]
The passage process pattern by construction period distribution creating unit 1916 calculates the number of products by passage process pattern by construction period according to the performance record data 200.
FIG. 20 is a diagram illustrating the number of products 2000 by actual construction period by pass process pattern.
In FIG. 20, for example, among the products whose passing process pattern is “0000000001”, the number of products whose actual work period is “0” is “422”.
Next, the passing process pattern-by-passing process pattern distribution distribution creation unit 1916 divides the passing process pattern-by-passing process pattern by product period 2000 product number by the passing process pattern-by-passing process pattern by product manufacturing number of products, thereby passing through process pattern-by-passing process pattern distribution period PΔ (tΔ | k) is calculated.

FIG. 21 is a diagram showing a work schedule distribution 2100 for each passing process pattern.
In FIG. 21, for example, when the actual number of manufactured products with the passing process pattern “0000000001” is 460, the number of products with the actual construction period “0” is 422 as shown in FIG. 20. By dividing 422 by 460, the probability that the construction period of the product whose passing process pattern is “0000000001” is “0” is 0.917 (= 422/460).
As described above, the passage process pattern-by-process period distribution creation unit 1916 corresponds to the process-by-step process period distribution creation unit 116 and the passage process pattern-by-pass process pattern distribution distribution unit 117.
The passage process pattern-by-process-period distribution creation unit 1916 can be realized by using, for example, a CPU, a ROM, and a RAM.

<< Second Comparative Example >>
Next, a second comparative example will be described.
FIG. 22 is a diagram illustrating a functional configuration of the manufacturing work period prediction device 2200 of this comparative example. The manufacturing work period prediction device 2200 of this comparative example is different from the manufacturing work time prediction apparatus 1900 of the first comparative example shown in FIG. The “process passage determination logic creation unit 112, process passage logic storage unit 113, and first product type creation unit 114” described in the embodiment are used, and the product type determination unit 1521 is used instead of the product type determination unit 1521. The process passage determination unit 121 and the second type creation unit 122 are used.

<Contrast between this embodiment and comparative example>
FIG. 23 is a diagram showing the results of the embodiment and the comparative example in a table format.
23, A1, A2, B1, and B2 indicate that the processing units A1, A2, B1, and B2 indicated by broken lines in FIGS. 1, 15, 19, and 22 are used. That is, in FIG. 22, the results described in the columns specified by A1 and B1 indicate the results obtained by the method of the first embodiment, and the results described in the columns specified by A2 and B1 are The result by the method of 2nd Embodiment is shown. The results described in the columns specified by A2 and B2 indicate the results of the method of the first comparative example, and the results described in the columns specified by A1 and B2 are the results of the second comparative example. The result by the method is shown.

  Here, manufacturing performance data is divided into learning data and evaluation data, and using the “process passage determination logic / product type determination logic” and “product type manufacturing time model” created using the learning data, the manufacturing time period of the evaluation data Evaluated.

Each evaluation index in FIG. 23 will be described.
First, the inventory amount will be described.
FIG. 24 is a diagram illustrating an example of a relationship between a value obtained by subtracting the actual work period from the production work period of the evaluation data (manufacturing work period−actual work period) and the weight of the product.
The graph 2401 shown in FIG. 24 is obtained by subtracting the actual work period from the production work period of the evaluation data for each of all the order data, and obtaining the sum of the weights of the orders in which the subtraction value becomes the same value. 24, in graph 2401, the value of “horizontal axis value” × “vertical value” in the region where the horizontal axis (manufacturing period-actual period) is larger than 0 (the hatched area in FIG. 24) The value obtained by dividing the total by the period (number of days) of the evaluation data is the amount of inventory per day. “Inventory quantity” in FIG. 23 indicates the daily inventory quantity calculated in this way.

Next, the assortment achievement rate is the weight ratio of an order in which the manufacturing period of the evaluation data is equal to or greater than the actual period (manufacturing period ≥ actual period) with respect to the total weight of the order (total weight of products belonging to the evaluation data). .
Next, the average work period is an average value of the production work period of the evaluation data.
Next, the robustness is a coefficient representing the correlation between the standard work period calculated using the learning data and the standard work period calculated using the evaluation data. Here, when obtaining these standard work schedules, the same decision tree created using the learning data is used. The standard work period is a production work period obtained by using a product-specific production process model calculated by setting a target assortment achievement rate for each product type.

  Next, the product match rate indicates how much the product types obtained when the decision tree created from the learning data and the decision tree created from the evaluation data are applied to the evaluation data match. The exact match is obtained by dividing the number of varieties obtained from the decision tree and the number of varieties that match in all bits with the passing process pattern determined from the evaluation data, by the total number of products included in the evaluation data. . The 1-bit difference is obtained by dividing the number of varieties obtained from the decision tree and having a value different by 1 bit with respect to the passing process pattern determined from the evaluation data, by the total number of products included in the evaluation data. The difference of 2 bits is obtained by dividing the number of varieties obtained from the decision tree that differ in value by 2 bits for the passing process pattern determined from the evaluation data, by the total number of products included in the evaluation data. It is done.

As shown in FIG. 23, it can be seen that the robustness improves when the production period model for each product type is created as in the first and second embodiments.
In the first and second embodiments, the likelihood p (t _i | _pi , μ, v) of the actual construction period t _i of each product i is the average μ of the construction period distribution (normal distribution) of the process that has actually passed. _It is assumed that a normal distribution N (μ _i ˜, v _i ˜) having an average μ _i ˜ and a variance v _i ˜ obtained by accumulating _j and variance v _j is assumed. Then, the likelihood p proven construction period t _i of each such product i | seeking process by implementation schedule distribution 600 when (t _i p _i, mu, v) the product of the maximum. Then, by adding the values of the work period distribution (average μ _j , variance v _j ) of the processes shown to be passing in the passing process pattern, the average μ _k ^ and the standard deviation of the work period distribution by the passing process pattern σ _k ^ is obtained, and using these, a work period distribution PΔ (tΔ | k) by passage process pattern is obtained (see FIG. 7).

On the other hand, in the first and second comparative examples, the actual work periods of products belonging to each passing process pattern are totaled to create a work period distribution 2100 by passing process pattern, so the number of products belonging to each passing process pattern If the amount is small, it is not possible to obtain a significant distribution of work schedules by passing process pattern, and it is difficult to increase the accuracy of the work schedule distribution by passing process patterns. As a result, the calculation result of the manufacturing period varies greatly depending on how the learning data is taken.
On the other hand, in the first and second embodiments, the average μ _k ^ and the standard deviation σ _k ^, which are examples of parameters of the passing schedule pattern-by-process schedule distribution, are obtained by adding the values of the routing schedule 600 by process. Therefore, the work period distribution for each passing process pattern can be calculated without depending on (largely) the number of products belonging to each passing process pattern (that is, the robustness can be improved). Therefore, the manufacturing period can be calculated stably. That is, even if only a small number of products belonging to each passing process pattern are obtained from the manufacturing performance data 200, the manufacturing work period can be accurately predicted.

Further, from FIG. 23, the decision tree 300 having the manufacturing specification as an explanatory variable and the passage presence / absence variable as an objective variable is created for each process (in the case of A1). Rather than creating one decision tree 1600 as an objective variable (as compared to A2), the inventory amount is small, the average construction period is short, and the type match rate is high.
In the decision tree 1600 having the manufacturing specification as an explanatory variable and the passing process pattern as an objective variable, in order to directly obtain the passing process pattern, the distance of each bit of each passing process pattern is considered when classifying each passing process pattern. Can not do it. Therefore, the classified passing process patterns include those that are greatly deviated from the passing process patterns that are correct. In contrast, in the first and second embodiments, since each bit (process) of each passing process pattern is obtained, it is difficult to obtain a passing process pattern that greatly deviates from the correct passing process pattern. Therefore, it is possible to accurately predict the manufacturing period.
In addition, the decision tree 1600 shown in FIG. 16 is a decision tree having a large size because the maximum number of objective variables is 2 (the number of steps). On the other hand, the decision tree 300 shown in FIG. 3 has two types of objective variables “0” and “1”, so that the decision tree has a small size.
From the above, it can be seen that creating the decision tree 300 as in the first embodiment is more desirable than creating the decision tree 1600 as in the second embodiment.

<Third Embodiment>
Next, a third embodiment of the present invention will be described. In the first and second embodiments, the passing process pattern of each order is obtained. On the other hand, in the present embodiment, a case will be described in which the order-specific manufacturing period is calculated and output when the passing process pattern of each order included in the order data is known in advance.
FIG. 25 is a diagram illustrating an example of a functional configuration of the manufacturing work period prediction apparatus 2500 in the present modification.

  The manufacturing lead time prediction apparatus 2500 of this embodiment differs from the manufacturing lead time prediction apparatus 100 of the first embodiment by the processing unit A1 (process passage determination logic creation unit 112, process passage logic storage unit 113, and first product type creation. 114), the production process pass pattern generation rate creation unit 115 by product type, the production period model creation unit 118 by product type, the process pass determination unit 121, and the second product creation unit 122 are deleted. Manufacturing period prediction model storage unit 119, order data acquisition unit 120, manufacturing period calculation unit 124 by type input order, manufacturing period storage unit 125 by type input order, manufacturing period output by type input order In place of the unit 126, the passing process pattern-specific work period distribution storage unit 2519, the order data acquisition unit 2520, the passing process input order-specific manufacturing work time calculation unit 2524 Over process input order by manufacturing period storage unit 2525, in which to use a passage step input order by manufacturing period output section 2526.

Similarly, the manufacturing work period prediction apparatus 2500 of the present embodiment is different from the manufacturing work period prediction apparatus 1500 of the second embodiment in that A2 processing units (product type determination logic creation unit 1512 and product type determination logic storage unit 1513) The passing process pattern generation rate creation unit 115, the product-specific manufacturing process model creation unit 118, and the product type determination unit 1521 are deleted, and the product-specific manufacturing process model storage unit 119 of the manufacturing process prediction apparatus 1500 according to the second embodiment, Instead of the order data acquisition unit 120, the manufacturing period calculation unit 124 by type input order, the manufacturing period storage unit 125 by type input order, and the manufacturing period output unit 126 by type input order, a period distribution storage unit 2519 by passing process pattern, Order data acquisition unit 2520, passage process input order-specific manufacturing period calculation unit 2524, passage process input order-specific manufacturing period storage unit 2525, It is obtained to use a step input orders by manufacturing period output section 2526.
Below, an example of a structure different from the manufacturing lead time prediction apparatuses 100 and 1500 of 1st and 2nd embodiment among the manufacturing lead time prediction apparatuses 2500 of this embodiment is demonstrated.

[Time distribution storage unit 2519 by passing process pattern]
The passing process pattern-specific work period distribution storage unit 2519 stores the passing process pattern-specific work period distribution generation part 117 created by the passing process pattern-specific work period distribution distribution PΔ (tΔ | k).
The passing process pattern-specific work period distribution storage unit 2519 can be realized by using, for example, an HDD.

[Order data acquisition unit 2520]
The order data acquisition unit 2520 acquires order data from the outside and stores it. The order data acquired by the order data acquisition unit 120 of the first and second embodiments does not include a passing process pattern, but the order data acquired by the order data acquisition unit 2520 of the present embodiment includes , The passing process pattern in each order is included. The order data acquisition method is the same as the order data acquisition unit 120 of the first and second embodiments, for example. Here, the passing process pattern in each order is obtained by including the passing process pattern in each order in the order data. However, the passing process pattern in each order is obtained separately from the order data. May be.
The order data acquisition unit 2520 can be realized by using, for example, a CPU, ROM, RAM, HDD, and various interfaces.

[Production period calculation unit 2524 for each passing process input order]
The passing process input order-by-order manufacturing work period calculating unit 2524 includes a passing process pattern-by-pass process pattern distribution distribution unit 2519 stored in the passing process pattern-by-process pattern distribution PΔ (tΔ | k) and “each order included in the order data”. And the target assortment achievement rate 1100 (see FIG. 11) set by the target assortment achievement rate setting unit 123, the work period t is calculated for each order.
Here, the manufacturing process period calculation unit 2524 for each passing process input order calculates a process period t that satisfies the following expression (17) for each order (for each type corresponding to the order).

It is assumed that PΔ (tΔ | k) is a normal distribution, and using the average μ and standard deviation σ of PΔ (tΔ | k) and a constant x set by the user, the following (18) The work period t may be calculated for each order (for each type corresponding to the order) as in the equation.
PΔ (tΔ | k) = μ + x × σ (18)
The passage process input order-specific manufacturing period calculation unit 2524 can be realized by using, for example, a CPU, a ROM, and a RAM.

[Manufacturing period storage unit 2525 according to passing process input order]
The passing process input order-by-order manufacturing manufacturing period storage unit 2525 stores the order period (order-by-order manufacturing period) calculated by the passing process input order-by-order manufacturing period calculation unit 2524.
The manufacturing process period storage unit 2525 for each passing process input order can be realized by using, for example, an HDD.
[Manufacturing period output unit 2526 for each passing process input order]
The passing process input order manufacturing manufacturing period output unit 2526 outputs the order specific manufacturing manufacturing period information stored in the passing process input order manufacturing manufacturing period storage unit 2525. The form of this output is the same as, for example, the production period output unit 126 by type input order in the first and second embodiments.
As described above, when the passing process pattern of each order is known in advance, it is possible to calculate (predict) an order-specific work period (order-specific manufacturing process period) without using a product type.

<Fourth Embodiment>
Next, a fourth embodiment of the present invention will be described. In the above-described first to third embodiments, the process-by-process distribution at the time of maximum likelihood estimation is a normal distribution (see (e) of paragraph [0033]), and the passing process pattern schedule distribution at the time of maximum likelihood estimation is It is a normal distribution (see paragraph [0033] (f)), and it is assumed that the passage process pattern construction distribution at the time of calculating the construction distribution by passage process pattern is a log normal distribution (see equation (12)). Explained with an example. On the other hand, in the present embodiment, the process-by-process distribution at the time of maximum likelihood estimation is a normal distribution, the passing process pattern schedule distribution at the time of maximum likelihood estimation is a lognormal distribution, A case where the passing process pattern work period distribution is assumed to be a lognormal distribution will be described. As described above, the present embodiment and the first to third embodiments mainly differ in part of the processes of the process-by-process construction period distribution creating unit 116 and the passing process pattern-by-process pattern construction period distribution creating unit 117. Therefore, in the present embodiment, only the processes of the process-specific work schedule distribution creating unit 116 and the passing process pattern-based work schedule distribution creating unit 117 will be described, and description of other parts will be omitted.

[Schedule distribution creation unit by process 116]
The process-specific work period distribution creating unit 116 calculates the work period-specific work period distribution using the manufacturing result data 200 (passing result presence / absence and result work period). The work period distribution is represented by a histogram or an approximation function (distribution function).

First, symbols are defined as follows.
(A) The actual construction period of the product i is t _i and the presence / absence of the passing process of the process j is p _ij .
(B) The passing process pattern (the actual value) of the product i is described as p _i = (p _i1 , p _i2 ,..., P _iM ). Here, M is the number of steps.
(C) The actual construction period of all products is t = (t ₁ , t ₂ ,..., T _N ) ^T. Here, N represents the total number of products, and T represents a transposed matrix.
(D) Let P = {p _ij } be the passing process pattern (the actual value) of all products.
(E) The work period distribution of the process j is assumed to be a normal distribution having an average μ _j and a variance v _j (= σ _j ² ). Here, the average μ _j and the variance v _j are the determining variables of the work period distribution by process.
(F) Actual construction period t _i shall lognormal distribution for each passage step pattern (actual value).
(G) The average value of the construction period distribution of all processes is μ = (μ ₁ , μ ₂ ,..., Μ _M ), and the variance is v = (v ₁ , v ₂ ,..., V _M ). .
Also here
(H) The work schedule distribution of each process shall be independent of the work schedule distribution of other processes (that is, covariance shall not be considered).

Then, likelihood p About product i proven construction period t _i obtained (t _i | p _i, mu, v) is expressed by the following equation (19) to (21) below.

(19) of the left side, p _i as the passing step pattern (actual value) of the product i is, mu as the average construction period distribution of all process, v is given respectively as a dispersion construction period distribution of the entire process It represents the likelihood that actual construction period t _i is obtained when. On the other hand, the right side of the equation (19) indicates that the actual work period t _i of the product i occurs according to a lognormal distribution LN having an average μ _i ˜ and a variance v _i ˜. Therefore, expression (19), the likelihood p About the product i track record construction period t _i obtained _{(t i | p i, μ} , v) of the distribution, the parameter μ _i ~ of the log-normal distribution, is a v _i ~ It represents following the lognormal distribution LN (μ _i ˜, v _i ˜).

Here, the average μ _i of the lognormal distribution represents the sum of the average μ _j of the work period distribution of the process j through which the product i has passed. Also, the logarithmic normal distribution variance v _i ⁻ represents the sum of the variance v _j of the work period distribution of the process j through which the product i has passed.
Further, the parameters μ _i ˜ and v _i ˜ of the lognormal distribution can be calculated using the average μ _i − and the variance v _i ⁻ of the log normal distribution from the equation (21).

  From the equations (19) to (21), the likelihood p (t | P, μ, v) at which the actual work period t of all products is obtained is expressed by the following equation (22).

(22) is, the likelihood p actual results construction period t of all products can be obtained (t | P, μ, v ) is, the actual construction period t ₁ of all products i, t _2, ···, likelihood of t _N It is expressed by the product of degrees.
From the equation (22), the log likelihood is expressed by the following equation (23).

In order to maximize the log likelihood of the equation (23), the average μ _j and the variance v _j that minimize the evaluation function J represented by the following equation (24) may be obtained.

(24) the evaluation function J of the equation is the evaluation function actual construction period t _1, t ₂ of all products i, · · ·, the higher the product of the likelihoods of t _N is greater value decreases. The average μ _j and the variance v _j may be calculated using the expression (24) as an evaluation function. However, if the value of N, which is the number of products, is large, optimization calculation of the expression (24) becomes difficult. . Therefore, in this embodiment, the evaluation function of equation (24) is rewritten as follows.
Here, the following symbols are defined.
(I) Let K be the number of types of passing process patterns.
(J) Let the passing process pattern k (k = 1, 2,..., K) be q _k = (q _k1 , q _k2 ,..., Q _kM ).
(K) Let N _k ^ be the number of products belonging to the passing process pattern k.
(L) It is assumed that the mean μ _k ^ and variance v _k ^ of the work period distribution of the passing process pattern k are expressed by the following equation (25).

Here, the average μ _k ^ represents the sum of the average μ _j of the process distribution of the process j indicated as having passed in the passing process pattern k. The variance v _k ^ represents the sum of the variance v _j of the process distribution of the process j shown to have passed in the passing process pattern k.
From the above (i) to (l), the evaluation function J of the equation (24) can be rewritten to the evaluation function J represented by the following equation (26).

The evaluation function J shown in the equation (26) is expressed by the sum of the terms of the number K of types of passing process patterns. In Equation (26), N _k ^ of the first term and the fourth term on the right side and the integration (Σ (lnt _i ) ² , Σlnt _i ) of the second term and the fourth term on the right side It can be calculated by counting. Therefore, it is easier to perform the optimization calculation using the evaluation function J shown in Expression (26) than to perform the optimization calculation using the evaluation function J of Expression (24). This optimization calculation can be executed by using a known nonlinear optimization calculation method such as downhill simplex method or quasi-Newton method. The process-by-process construction period distribution creating unit 116 performs optimization calculation by giving the production performance data 200 (passing performance presence / absence and performance construction period) to the expressions (25) and (26), thereby calculating the average μ _j and the variance v _j. Is calculated as an example of the parameter of the work period distribution by process.
The process-specific work schedule distribution creation unit 116 can be realized by using, for example, a CPU, a ROM, and a RAM.

[Construction distribution creation unit 117 by passing process pattern]
The passing process pattern-specific work period distribution creating unit 117 calculates the passing process pattern-based work period distribution using the process-specific work period distribution creating unit 116.
First, from the relationship of the equation (25), the average μ _k ^ and the variance v _k ^ of the passing schedule pattern-specific work period distribution are calculated based on the process-specific work period distribution (average μ _j , variance v _j ).
Next, the construction period distribution creation unit 117 for each passing process pattern derives a lognormal distribution whose average is μ _k ^ and variance is v _k ^. For this purpose, the passage process pattern-by-process-period distribution creation unit 117 calculates logarithmic normal distribution parameters μ _k ^ ′, v _k ^ ′ using the formula shown in the proviso of formula (26).

  In the present embodiment, the construction period distribution by passing process pattern is expressed by the following equation (27).

However, in the present embodiment as well, in the same way as in the first to third embodiments, the equation (27) is discretized so as to obtain a work period distribution PΔ (tΔ | k) by passing process pattern (( (See 13)).
The passing process pattern-specific work period distribution creation unit 117 can be realized by using, for example, a CPU, a ROM, and a RAM.

<Fifth Embodiment>
Next, a fifth embodiment of the present invention will be described. In the present embodiment, the process-specific work period distribution at the time of maximum likelihood estimation, the passing process pattern work period distribution at the time of maximum likelihood estimation, and the pass process pattern work period distribution at the time of calculating the work process time distribution by pass-through process pattern are all Poisson distributions. The assumed case will be described. As described above, the present embodiment and the first to fourth embodiments are mainly different in part of the processes of the process-specific work period distribution creating unit 116 and the passing process pattern-specific work period distribution creating part 117. Therefore, in the present embodiment, only the processes of the process-specific work schedule distribution creating unit 116 and the passing process pattern-based work schedule distribution creating unit 117 will be described, and description of other parts will be omitted.

[Schedule distribution creation by process 116]
The process-specific work period distribution creating unit 116 calculates the work period-specific work period distribution using the manufacturing result data 200 (passing result presence / absence and result work period). The work period distribution is represented by a histogram or an approximation function (distribution function). Below, an example of the preparation method of distribution schedule according to a process is demonstrated.

First, symbols are defined as follows.
(A) The actual construction period of the product i is t _i and the presence / absence of the passing process of the process j is p _ij .
(B) The passing process pattern (the actual value) of the product i is described as p _i = (p _i1 , p _i2 ,..., P _iM ).
(C) The actual construction period of all products is t = (t ₁ , t ₂ ,..., T _N ) ^T. Here, N represents the total number of products, and T represents a transposed matrix.
(D) Let P = {p _ij } be the passing process pattern (the actual value) of all products.
(E) It is assumed that the work period distribution of the process j is a Poisson distribution with an average λ _j . Here, the average λ _j becomes the determining variable of the work period distribution by process.
(F) Actual construction period t _i shall be in accordance with Poisson distribution for each passage step pattern (actual value).
(G) Let λ = (λ ₁ , λ ₂ ,..., Λ _M ) be the average value of the work schedule distribution in all processes.
Also here
(H) The work schedule distribution of each process shall be independent of the work schedule distribution of other processes (that is, covariance shall not be considered).

Then, the product i for actual construction period t likelihood _i is obtained p (t _i | p _i, lambda) is expressed by the following equation (28) and (29).

The left side of the equation (28) shows the likelihood that the actual work period t _i is obtained when p _i is given as the passing process pattern (its actual value) of the product i and λ is given as the average value of the work period distribution of all processes. Represents. On the other hand, the right side of the equation (28) indicates that the actual work period t _i of the product i occurs according to the Poisson distribution Poisson whose average is λ _i . Therefore, the distribution of the likelihood p (t _i | p _i , λ) from which the actual work period t _i is obtained for the product i follows the Poisson distribution Poisson (λ _i ˜) whose average is λ _i ˜. Represents that.
Here, the average λ _i represents the sum of the average λ _j of the work period distribution of the process j through which the product i has passed.

  From the equations (28) and (29), the likelihood p (t | P, λ) with which the actual work period t of all products is obtained is expressed by the following equation (30).

(30) equation, the likelihood p actual results construction period t of all products can be obtained (t | P, λ) is, the actual construction period t _1, t ₂ of all of the products i, ···, of the likelihood of t _N It is expressed by the product.
From the equation (30), the log likelihood is expressed by the following equation (31).

In order to maximize the log likelihood of the equation (31), an average λ _j that minimizes the evaluation function J represented by the following equation (32) may be obtained.

(32) the evaluation function J of the equation is the evaluation function actual construction period t _1, t ₂ of all products i, · · ·, the higher the product of the likelihoods of t _N is greater value decreases. The average λ _j may be calculated using the equation (32) as an evaluation function. However, if the value of N, which is the number of products, is large, the optimization calculation of the equation (32) becomes difficult. Therefore, in this embodiment, the evaluation function of equation (32) is rewritten as follows.
Here, the following symbols are defined.
(I) Let K be the number of types of passing process patterns.
(J) Let the passing process pattern k (k = 1, 2,..., K) be q _k = (q _k1 , q _k2 ,..., Q _kM ).
(K) Let N _k ^ be the number of products belonging to the passing process pattern k.
(L) The average λ _k ^ of the work period distribution of the passing process pattern k is represented by the following equation (33).

Here, the average λ _k ^ represents the sum of the average λ _j of the process distribution of the process j that is shown to have passed in the passing process pattern k.
From the above (i) to (l), the evaluation function J of the equation (32) can be rewritten to the evaluation function J represented by the following equation (34).

The evaluation function J shown in the equation (34) is expressed by the sum of the terms of the number K of types of passing process patterns. Further, in the equation (34), N _k ^ of the first term on the right side and the integration (Σt _i ) of the second term on the right side can be calculated by counting the manufacturing result data 200. Therefore, it is easier to perform the optimization calculation using the evaluation function J shown in the equation (34) than to perform the optimization calculation using the evaluation function J of the equation (32). This optimization calculation can be executed by using a known nonlinear optimization calculation method such as downhill simplex method or quasi-Newton method. The process-by-process construction period distribution creating unit 116 gives the production results data 200 (passage record presence / absence and actual construction period) to the formulas (33) and (34) and performs the optimization calculation, thereby calculating the average λ _j by the process. Calculate as an example of the parameters of the work schedule distribution.
The process-specific work schedule distribution creation unit 116 can be realized by using, for example, a CPU, a ROM, and a RAM.

[Construction distribution creation unit 117 by passing process pattern]
The passing process pattern-specific work period distribution creating unit 117 calculates the passing process pattern-based work period distribution using the process-specific work period distribution creating unit 116.
First, the average λ _k ^ of the process schedule distribution by passing process pattern is calculated based on the process schedule distribution (average λ _j ) from the relationship of the equation (33).
In the present embodiment, the construction period distribution by passing process pattern is expressed by the following equation (35).

However, in the present embodiment as well, in the same way as in the first to fourth embodiments, this formula (35) is discretized so as to obtain a construction period distribution PΔ (tΔ | k) by passing process pattern (( (See 13)).
The passing process pattern-specific work period distribution creation unit 117 can be realized by using, for example, a CPU, a ROM, and a RAM.

<Sixth Embodiment>
Next, a sixth embodiment of the present invention will be described. In the present embodiment, the process-by-process distribution at the time of maximum likelihood estimation, the passing process pattern schedule distribution at the time of maximum likelihood estimation, and the passing process pattern schedule distribution at the time of calculation of the process schedule distribution by the passing process pattern are all gamma distributions. The assumed case will be described. As described above, the present embodiment and the first to fifth embodiments are mainly different in part of the processes of the process-specific work period distribution creating unit 116 and the passing process pattern-based work period distribution creating part 117. Therefore, in the present embodiment, only the processes of the process-specific work schedule distribution creating unit 116 and the passing process pattern-based work schedule distribution creating unit 117 will be described, and description of other parts will be omitted.

[Schedule distribution creation by process 116]
The process-specific work period distribution creating unit 116 calculates the work period-specific work period distribution using the manufacturing result data 200 (passing result presence / absence and result work period). The work period distribution is represented by a histogram or an approximation function (distribution function). Below, an example of the preparation method of distribution schedule according to a process is demonstrated.

First, symbols are defined as follows.
(A) The actual construction period of the product i is t _i and the presence / absence of the passing process of the process j is p _ij .
(B) The passing process pattern (the actual value) of the product i is described as p _i = (p _i1 , p _i2 ,..., P _iM ).
(C) The actual construction period of all products is t = (t ₁ , t ₂ ,..., T _N ) ^T. Here, N represents the total number of products, and T represents a transposed matrix.
(D) Let P = {p _ij } be the passing process pattern (the actual value) of all products.
(E) It is assumed that the work period distribution of the process j is a gamma distribution having a shape parameter b _j and a scale parameter θ. Here, the shape parameter b _j and the scale parameter θ are the decision variables of the work period distribution. Here, in the present embodiment, the scale parameter θ is not a parameter for each process. This is because the reproducibility of the work schedule distribution cannot be obtained unless this is done.
(F) Actual construction period t _i shall be in accordance with the gamma distribution for each passage step pattern (actual value).
(G) The shape parameter of the work schedule distribution of all processes is set as b = (b ₁ , b ₂ ,..., B _M ).
Also here
(H) The work schedule distribution of each process shall be independent of the work schedule distribution of other processes (that is, covariance shall not be considered).

Then, likelihood p About product i proven construction period t _i obtained (t _i | p, b, theta) is expressed by the following equation (36) and (37) below.

(36) The left-hand side of the equation, p _i as the passing step pattern (actual value) of the product i is, b as shape parameter construction period distribution of all processes, theta each as scale parameter construction period distribution of all stroke actual construction period t _i when given represents the likelihood is obtained. On the other hand, (36) the right side of the equation, the actual construction period t _i of the product i represents the shape parameter is b _i ~, scale parameter is generated in accordance with a gamma distribution Gamma is theta. Thus, equation (36) is likelihood p About product i proven construction period t _i obtained (t _i | p _i, b, theta) distribution of the shape parameter is b _i ~, in scale parameter is theta It represents following a certain gamma distribution Gamma (b _i ˜, θ).
Here, the shape parameter b _i ˜ represents the sum of the shape parameters b _j of the work schedule distribution of the process j through which the product i has passed.

  From the equations (36) and (37), the likelihood p (t | P, b, θ) from which the actual work period t of all products is obtained is expressed by the following equation (38).

(38) equation, the actual construction period t likelihood is obtained p of all products (t | P, b, θ ) is, the actual construction period t ₁ of all products i, t _2, ···, likelihood of t _N It is expressed by the product of degrees.
From the equation (38), the log likelihood is expressed by the following equation (39). The gamma function Γ included in the equation (39) is a function defined by the integration of the following equation (40) for the complex number z whose real part is positive.

In order to maximize the log likelihood of the equation (39), the shape parameter b _j and the scale parameter θ that minimize the evaluation function J represented by the following equation (41) may be obtained.

(41) the evaluation function J of the equation is the evaluation function actual construction period t _1, t ₂ of all products i, · · ·, the higher the product of the likelihoods of t _N is greater value decreases. The shape parameter b _j and the scale parameter θ may be calculated using the equation (41) as an evaluation function. However, if the value of N as the number of products is large, the optimization calculation of the equation (41) is performed. It becomes difficult. Therefore, in this embodiment, the evaluation function of equation (41) is rewritten as follows.
Here, the following symbols are defined.
(I) Let K be the number of types of passing process patterns.
(J) Let the passing process pattern k (k = 1, 2,..., K) be q _k = (q _k1 , q _k2 ,..., Q _kM ).
(K) Let N _k ^ be the number of products belonging to the passing process pattern k.
(L) It is assumed that the shape parameter b _k ^ of the work schedule distribution of the passing process pattern k is expressed by the following equation (42).

Here, the shape parameter b _k ^ represents the sum of the shape parameters b _j of the process distribution of the process j shown to have passed in the passing process pattern k.
From the above (i) to (l), the evaluation function J of the equation (41) can be rewritten to the evaluation function J represented by the following equation (43).

The evaluation function J shown in the equation (43) is expressed by the sum of terms of the number K of types of passing process patterns. Further, in expression (43), and the hand side of the third term and the fourth term N _k ^, integration of the right side first term and the second term (Σlnt _i, Σt _i), by aggregating the manufacturing performance data 200 Can be calculated. Therefore, it is easier to perform the optimization calculation using the evaluation function J shown in the equation (43) than to perform the optimization calculation using the evaluation function J of the equation (41). This optimization calculation can be executed by using a known nonlinear optimization calculation method such as downhill simplex method or quasi-Newton method. The process-by-process construction period distribution creating unit 116 performs optimization calculation by giving the production performance data 200 (passing performance presence / absence and performance construction period) to the formulas (42) and (43), so that the shape parameter b _j , scale The parameter θ is calculated as an example of a parameter of the work period distribution by process.
The process-specific work schedule distribution creation unit 116 can be realized by using, for example, a CPU, a ROM, and a RAM.

[Construction distribution creation unit 117 by passing process pattern]
The passing process pattern-specific work period distribution creating unit 117 calculates the passing process pattern-based work period distribution using the process-specific work period distribution creating unit 116.
First, based on the relationship of the formula (42), the shape parameter b _k ^ of the process distribution by passage process pattern (shape parameter b _j ) is calculated based on the process distribution by shape (shape parameter b _j ).
In this embodiment, the work period distribution by passing process pattern is expressed by the following equation (44).

However, in the present embodiment as well, in the same way as in the first to fourth embodiments, this formula (44) is discretized so as to obtain the work process distribution PΔ (tΔ | k) by passing process pattern (( (See 13)).
The passing process pattern-specific work period distribution creation unit 117 can be realized by using, for example, a CPU, a ROM, and a RAM.

<Comparison by function assumed as work schedule distribution>
As described above, in the present specification, the cases assumed as the work schedule distribution are described as examples (A) to (D) below. The fourth to sixth embodiments first correspond to the configuration of the embodiment.
(A): Refer to the first embodiment for details. Construction period distribution by process at the time of maximum likelihood estimation; passing process pattern construction period distribution at the time of normal distribution maximum likelihood estimation; passing process pattern at the time of calculation of the distribution distribution by normal distribution passage process pattern Construction period distribution; log normal distribution (B); refer to the fourth embodiment for details. Construction period distribution by process at maximum likelihood estimation; passage process pattern construction distribution at normal distribution maximum likelihood estimation; construction period by log normal distribution passage process pattern Passing process pattern construction distribution at the time of distribution calculation; log normal distribution (C); see the fifth embodiment for details; construction distribution by process at the time of maximum likelihood estimation; passing process pattern construction distribution at the time of Poisson distribution maximum likelihood estimation; Poisson Distribution process pattern distribution distribution at the time of calculation of distribution distribution process pattern; Poisson distribution (D); see the sixth embodiment for details; Distribution passing step pattern implementation schedule distribution during MLE; Gaussian passing step pattern specific construction period distribution calculation time of the passing step pattern implementation schedule distribution; Gaussian distribution

Here, with respect to (same) manufacturing performance data (learning data) obtained in a certain period in the steel manufacturing process, it is assumed that the schedule distribution is as shown in (A) to (D), and the schedule according to the process. The distribution (parameters thereof) was calculated (see the description of the process-specific work period distribution creation unit 116 in each embodiment).
FIG. 26 shows an example of the work period distribution by process in the case of (A), FIG. 27 shows an example of the work period distribution by process in the case of (B), and FIG. 28 shows the process distribution in the case of (C). An example of the work schedule distribution is shown, and FIG. 29 shows an example of the work schedule distribution by process in the case of (D).

  Furthermore, by using the process-by-process distribution obtained as described above, the process-by-process pattern distribution PΔ (tΔ | k) was calculated for each of the cases (A) to (D) (each embodiment). (Refer to the description of the passing process pattern construction period distribution creation unit 117 in FIG. 1). For each order, the “passing process pattern in each order” included in the order data and the target assortment achievement rate 1100 (see FIG. 11) set by the target assortment achievement rate setting unit 123 are used. The work period t was calculated for each of the cases (A) to (D) (see the description of the manufacturing process calculation part 2524 for each passing process input order). 0.85 (= 85%), 0.90 (= 90%), and 0.95 (= 95%) are set as the target assortment achievement rate 1100. In each case, the construction period t for each order is calculated. did. Thereby, the work schedule t for each target assortment achievement rate 1100 and each passing process pattern is obtained. In the following explanation, “the construction period t in each passing process pattern when the target assortment achievement rate 1100 is 0.85 (= 85%), 0.90 (= 90%), and 0.95 (= 95%)” Are referred to as “85% construction period”, “90% construction period”, and “95% construction period”, respectively.

  On the other hand, from the manufacturing performance data (evaluation data) in the steel manufacturing process obtained in a period different from the above-described manufacturing performance data (learning data), the construction period distribution PΔ (tΔ | k) by passing process pattern was calculated (passing) (Refer to the description of the process pattern-specific work period distribution creation unit 1916). In the same manner as described above, the “85% work period”, “90% work period”, and “95% work period” were calculated. Here, the actual construction period (“85% construction period”, “90% construction period” and “95% construction period”) cannot be calculated correctly with a small number of passing process patterns. For this reason, here, a passing process pattern to which evaluation data of 500 sheets or more belongs is set as an evaluation target.

Then, the construction period obtained from the learning data and the construction period obtained from the evaluation data are applied to a single regression equation, and each of the “85% construction period”, “90% construction period” and “95% construction period” is single regression. The coefficient of determination of the equation and the coefficient of the single regression equation were calculated as a measure of the prediction accuracy evaluation.
FIG. 30 is a diagram illustrating determination coefficients of a single regression equation, and FIG. 31 is a diagram illustrating coefficients of a single regression equation. The coefficient of determination is a general index that represents how much the work period obtained from the learning data can explain the work period obtained from the evaluation data. Also, the coefficient of the single regression equation should be closer to 1, and if it is greater than 1, it means that the work period obtained from the evaluation data tends to be longer than the work period obtained from the learning data, and conversely, 1 If it is smaller, it means that the construction period obtained from the evaluation data tends to be shorter than the construction period obtained from the learning data.
As shown in FIGS. 30 and 31, when the above-described (C) is performed (assuming that the work period distribution is Poisson distribution), the value of the coefficient of determination increases, but the coefficient of the single regression equation is “1”. Greatly exceeded. That is, in the steel manufacturing process that is the subject of the current work schedule prediction, assuming that the work schedule distribution is Poisson distribution, there is a tendency to calculate the predicted work schedule shorter. On the other hand, when (B) mentioned above is assumed (assuming that the work period distribution by process is a normal distribution and the passing process pattern work period distribution is a log normal distribution), the coefficient of determination and the coefficient of the single regression equation are close to “1”. Become. In other words, in the steel manufacturing process that is the subject of the current work schedule prediction, assuming that the work schedule distribution by process is a normal distribution and the passing process pattern work schedule distribution is a lognormal distribution, it can be said that the total prediction accuracy of the work schedule is high. In this way, as in this case, when a manufacturing process in which work is always accumulated in each process is set as the target of construction period prediction, the average and variance differ greatly for each process. When a distribution including both the average value and the variance is used (assuming the work period distribution as in (A) and (B) described above), the work period can be predicted with high accuracy.

  However, depending on the manufacturing process for which the construction period is to be predicted, the construction period can be accurately predicted even if the construction period distribution is assumed as in the above-described (C) and (D). For example, in a manufacturing process with a high manufacturing capacity in each process (a manufacturing process in which devices are not always accumulated and there is free time in equipment), the process is likely to be queued. For such a manufacturing process, it is considered that the construction period can be accurately predicted if the construction period distribution is assumed to be a Poisson distribution as in (C) described above. In addition, although the average value of the work schedule distribution does not vary greatly for each process, for the manufacturing process in which the distribution (shape) is different, the work schedule distribution is a gamma distribution as described in (D) above. Assuming that the construction period can be predicted with high accuracy. The Poisson distribution and the gamma distribution are reproducible distributions. Reproducibility refers to the property that a probability distribution of the sum of two independent random variables having probability distributions included in the same distribution family is also included in the same family. Therefore, assuming the work schedule distribution as in (C) and (D) described above, the work schedule distribution by process at the time of maximum likelihood estimation, the process schedule distribution by passage process pattern at the time of maximum likelihood estimation, and the work schedule by pass process pattern Since the process schedule can be calculated under the same distribution assumption, and the passage schedule pattern schedule distribution at the time of distribution calculation can be consistently calculated with the same distribution, degradation of distribution estimation accuracy is minimized. Can be suppressed. If the process-by-process construction distribution at the time of maximum likelihood estimation is assumed to be a log normal distribution, the process-by-process construction distribution cannot be calculated analytically. For this reason, here, a work schedule distribution is assumed as in (A) and (B) described above.

<Other variations>
In the first embodiment described above, the case where the process pass determination logic is the decision tree 300 is taken as an example, and in the second embodiment, the case where the type determination logic is the decision tree 1600 is taken as an example. However, these are not limited to decision trees. That is, the process passage determination logic can use, for example, a statistical model or a table as long as it predicts, for each process, a process through which a product passes and a process that does not pass from the manufacturing specifications included in the manufacturing result data 200. Or use a model of the scheme. Similarly, if the product type determination logic predicts the product type from the manufacturing specifications included in the manufacturing performance data 200, a statistical model or a table type model can be used.

<< Correspondence with Claims >>
(Claims 1 and 15)
The manufacturing performance data acquisition means (process) is realized by, for example, the manufacturing performance data acquisition unit 111 (processing to be performed). Here, the manufacturing performance data is realized by the manufacturing performance data 200, for example.
The process-specific work schedule distribution deriving means (process) is realized by, for example, the process-specific work schedule distribution creating unit 116 (processing performed). Here, the parameters that define the distribution of the construction period for each manufacturing process are realized by, for example, the average μ _j and the variance v _j shown in FIG.
The passage process pattern-specific work period distribution deriving means (process) is realized by, for example, the passage process pattern-specific work period distribution creation unit 117 (processing performed). Here, the construction period distribution for each passing process pattern is realized by, for example, the information of the construction period distribution 703 for each passing process pattern shown in FIG. In addition, the addition of the parameters that define the distribution of the construction period of the manufacturing process indicated as having passed in the passing process pattern is realized by, for example, calculating equations (8) and (9). .
The first type determining means (process) is realized by, for example, the processing unit A1 (processing performed in FIG. 1) or the processing unit A2 (processing performed in FIG. 15). Here, the product type is realized by, for example, the product type 402 in FIG. 4 and the product type 1700 in FIG.
The type-by-type passing process pattern occurrence rate deriving means (process) is realized by, for example, the type-by-type passing process pattern occurrence rate creating unit 115 (processing performed). Here, the generation probability of the passing process pattern for each type is realized by, for example, information on the passing process pattern occurrence rate 500 in FIG.
The type-specific manufacturing period model deriving means (process) is realized by, for example, the type-specific manufacturing period model creation unit 118 (processing performed). Here, deriving the work period distribution by product type using the work process time distribution by pass process pattern and the occurrence probability of the pass process pattern by product type is realized, for example, by performing calculation of equation (14). The The distribution of work periods by product type is realized, for example, by the production work model 800 by product type shown in FIG.
The second type determination means (process) is realized by, for example, the second type creation unit 122 (processing performed) or the type determination unit 1521 (processing performed) shown in FIG.
The production period derivation means (process) for each type of input order is realized by, for example, the process performed by the manufacturing period calculation unit 124 for each type of input order. Here, derivation of the manufacturing period for each order using the type and the distribution of the period for each type can be realized by, for example, calculating equation (16).
(Claims 2 and 16)
The process passage determination logic creation means (process) is realized by, for example, the process passage determination logic creation unit 112 (processing performed). Here, the process passage determination logic is realized by, for example, the decision tree 300 of FIG.
The first product type creation means (process) is realized by, for example, the first product type creation unit 114 (processing performed).
The second kind creation means (process) is realized by, for example, the second kind creation unit 122 (processing performed).
(Claims 3 and 17)
The product type determination logic creating means (process) is realized by, for example, the product type determination logic creating unit 1512 (processing performed).
The product type determination means (process) is realized by, for example, the product type determination unit 1521 (processing performed in).
(Claims 4 and 18)
The target assortment achievement rate setting means (process) is realized by, for example, the target assortment achievement rate setting unit 123 (processing performed). Here, the target assortment achievement rate is realized by, for example, the target assortment achievement rate 1100 in FIG. In addition, the derivation of the manufacturing period for each order using the type, the distribution of the period for each type of product, and the target assortment achievement rate can be realized by, for example, calculating equation (15).
(Claims 5 and 19)
The manufacturing performance data acquisition means (process) is realized by, for example, the manufacturing performance data acquisition unit 111 (processing to be performed). Here, the manufacturing performance data is realized by the manufacturing performance data 200, for example.
The process-specific work schedule distribution deriving means (process) is realized by, for example, the process-specific work schedule distribution creating unit 116 (processing performed). Here, the parameter which prescribes | regulates the distribution of the construction period according to manufacturing process is implement | achieved by the information of the construction period distribution 600 according to process shown in FIG. 6, for example.
The passage process pattern-specific work period distribution deriving means (process) is realized by, for example, the passage process pattern-specific work period distribution creation unit 117 (processing performed). Here, the construction period distribution for each passing process pattern is realized by, for example, the information of the construction period distribution 703 for each passing process pattern shown in FIG. In addition, the addition of the parameters that define the distribution of the construction period of the manufacturing process indicated as having passed in the passing process pattern is realized by, for example, calculating equations (8) and (9). .
The passing process input order-specific manufacturing work period deriving means (process) is realized by, for example, the passing process input order-specific manufacturing work period calculating unit 2524 (processing performed). Here, using the passage process pattern and the construction period distribution for each passage process pattern, deriving the production period for each order is realized, for example, by calculating equation (18).
(Claims 6 and 20)
The target assortment achievement rate setting means (process) is realized by, for example, the target assortment achievement rate setting unit 123 (processing performed). Here, the target assortment achievement rate is realized by, for example, the target assortment achievement rate 1100 in FIG. In addition, using the passage process pattern, the work period distribution by the passage process pattern, and the target assortment achievement rate, deriving the production work period by order is realized by, for example, calculating Equation (17). Is done.
(Claims 7 and 21)
Deriving the mean and variance of the normal distribution expressed as the distribution of the construction period in the manufacturing process so that the value of the evaluation function that becomes smaller as the product of the likelihood of obtaining the actual construction period of each product increases This is realized, for example, by performing calculation of Expression (7) obtained based on Expression (3).
Each likelihood that the actual work period of each product is obtained is expressed by a normal distribution with the average and variance of the average and variance of the work period distribution of the manufacturing process shown to have passed in the passing process pattern. What is done is obtained by, for example, the expressions (1) and (2).
(Claims 8 and 22)
The passage process pattern-specific work distribution parameter deriving means (process), the log normal distribution deriving means (process), and the manufacturing work period occurrence probability deriving means (process) are, for example, a process performed by the passage process pattern-specific work distribution distribution unit 117 (process). It is realized by. Here, deriving the mean and variance of the work schedule distribution for each passing process pattern by adding the mean and variance of the work schedule distribution of the manufacturing process shown to have passed in the passing process pattern, for example, This is realized by calculating equations (8) and (9). Moreover, the log normal distribution which has the average and dispersion | distribution of construction period distribution according to passage process pattern is implement | achieved by (12) Formula, for example. In addition, the occurrence probability of each work period of a product having a passing process pattern is realized by, for example, PΔ (tΔ | k).
(Claim 9, Claim 23)
Deriving the average and standard deviation of the normal distribution expressed as the distribution of the construction period of the manufacturing process so that the value of the evaluation function that becomes smaller as the product of the likelihood of obtaining the actual construction period of each product increases This is realized, for example, by calculating the equation (26) obtained based on the equation (22).
Each likelihood that the actual work period of each product is obtained is a lognormal distribution with the average and variance of the average and variance of the work period distribution of the manufacturing process shown to have passed in the passing process pattern. The expression is obtained by, for example, Expressions (19) to (21).
(Claim 10, Claim 24)
The passage process pattern-specific work distribution parameter deriving means (process), the log normal distribution deriving means (process), and the manufacturing work period occurrence probability deriving means (process) are, for example, a process performed by the passage process pattern-specific work distribution distribution unit 117 (process). It is realized by. Here, deriving the mean and variance of the work schedule distribution for each passing process pattern by adding the mean and variance of the work schedule distribution of the manufacturing process shown to have passed in the passing process pattern, for example, This is realized by calculating equation (25). In addition, the log normal distribution having the average and variance of the work period distribution for each passing process pattern is realized by, for example, Expression (27). In addition, the occurrence probability of each work period of a product having a passing process pattern is realized by, for example, PΔ (tΔ | k).
(Claims 11 and 25)
Deriving the average of the Poisson distribution expressed as the distribution of the construction period of the manufacturing process so that the value of the evaluation function that becomes smaller as the product of the likelihood of obtaining the actual construction period of each product increases becomes For example, it is realized by performing the calculation of Expression (34) obtained based on Expression (30).
Each likelihood that the actual work period of each product is obtained is expressed by Poisson distribution with the average of each integrated value of the work period distribution of the manufacturing process shown to have passed in the passing process pattern. For example, it is obtained by the equations (28) and (29).
(Claim 12, Claim 26)
The passage process pattern-specific work period distribution parameter deriving means (process), the Poisson distribution deriving means (process), and the manufacturing work period occurrence probability deriving means (process) are performed by, for example, the process-specific distribution creation unit 117 (process to be performed). Realized. Here, deriving the average of the work schedule distribution for each passing process pattern by adding the average of the work schedule distribution of the manufacturing process shown to have passed in the passing process pattern is, for example, Equation (33) This is realized by calculating In addition, the Poisson distribution having the same average as the average construction period distribution for each passing process pattern is realized by, for example, Expression (35). In addition, the occurrence probability of each work period of a product having a passing process pattern is realized by, for example, PΔ (tΔ | k).
(Claims 13 and 27)
The shape parameter and the scale parameter of the gamma distribution expressed as the distribution of the construction period of the manufacturing process so that the value of the evaluation function that becomes smaller as the product of the likelihood of obtaining the actual construction period of each product increases is minimized. Deriving is realized, for example, by calculating the equation (43) obtained based on the equation (38).
Each of the likelihood that the actual work period of each product can be obtained is the integrated value of the shape parameter of the work period distribution of the manufacturing process that is shown to have passed in the passing process pattern. The fact that it is expressed by a gamma distribution with a non-constant value as a scale parameter can be obtained by, for example, Expressions (36) and (37).
(Claims 14 and 28)
The passage process pattern-specific work period distribution parameter deriving means (process), the gamma distribution deriving means (process), and the manufacturing work period occurrence probability deriving means (process) are performed by, for example, a process by the passing process pattern-specific work period distribution creating unit 117 Realized. Here, deriving the shape parameter of the work schedule distribution for each passing process pattern by adding the shape parameters of the work schedule distribution of the manufacturing process shown to have passed in the passing process pattern, for example, This is realized by calculating equation (42). Further, the Gaussian distribution having the same shape parameter as the shape parameter of the work schedule distribution for each passing process pattern is realized by, for example, Expression (44). In addition, the occurrence probability of each work period of a product having a passing process pattern is realized by, for example, PΔ (tΔ | k).

The embodiment of the present invention described above can be realized by a computer executing a program. Further, a means for supplying the program to the computer, for example, a computer-readable recording medium such as a CD-ROM recording such a program, or a transmission medium for transmitting such a program may be applied as an embodiment of the present invention. it can. A program product such as a computer-readable recording medium that records the program can also be applied as an embodiment of the present invention. The programs, computer-readable recording media, transmission media, and program products are included in the scope of the present invention.
In addition, the embodiments of the present invention described above are merely examples of implementation in carrying out the present invention, and the technical scope of the present invention should not be construed as being limited thereto. Is. That is, the present invention can be implemented in various forms without departing from the technical idea or the main features thereof.

100, 1500 Production period prediction device 200 Production result data 300, 1600 Decision tree 402, 1700 Product type by production number 500 Passing process pattern generation rate 600 Process period distribution by process 703 Process period distribution by passing process pattern 800 Production period model by type 900 Order Data 1002 Variety by order 1100 Achievement rate of target assortment 1200 Production period by order

Claims (21)

A manufacturing period prediction device for predicting the period of an order from order data including manufacturing specifications for each order,
Data indicating the manufacturing performance of each product manufactured through a plurality of manufacturing processes, including the manufacturing specifications of the product, the presence / absence of passing results as information indicating the actual value of the presence / absence of passing of each manufacturing process, and the product Manufacturing result data acquisition means for acquiring manufacturing result data including a result work period that is information indicating the result of the work period required to manufacture
By using the past results for each product and the actual work period, a process-specific work period distribution deriving means for deriving parameters defining the distribution of work periods by manufacturing process,
By arranging the value of the passing record presence / absence in a predetermined order using the parameter that defines the distribution of the work period for each manufacturing process derived by the process-based work period distribution deriving means and the passing record presence / absence for each product. Means for deriving work schedule distribution by passing process pattern for deriving work schedule distribution by passing process pattern obtained,
A product included in the manufacturing performance data is a product type obtained by arranging predicted values of the presence or absence of passing through each manufacturing process of the product in the predetermined order using the manufacturing specifications of the product included in the manufacturing performance data. First variety determining means for determining each of the
Using the number of products for each passing process pattern and the number of products for each product type, the passing process pattern occurrence rate deriving means for each product type for deriving the occurrence probability of the passing process pattern for each product type,
The production period model derivation means by type for deriving the work period distribution by type, using the period distribution by passage process pattern and the occurrence probability of the passing process pattern by type,
Using the manufacturing specification included in the order data, a predicted value of whether or not the order included in the order data passes through each manufacturing process is obtained, and the predicted value of whether or not the order passes through each manufacturing process is determined as the predetermined value. Second type determining means for determining the type obtained by arranging in order for each of the orders included in the order data;
Using the product type determined by the second product type determining means and the product time period distribution for each product type derived by the product type manufacturing time model deriving unit, the production time period by product type input order for deriving the production time period for each order Deriving means, and
The passing process pattern-specific work period distribution deriving means is a probability density determined on the basis of the parameters obtained by adding the parameters defining the work period distribution of the manufacturing process shown to have passed in the passing process pattern. As a function, derive the work schedule distribution by the passing process pattern ,
The first product type determination means generates a process pass determination logic for each of the manufacturing processes, which receives the manufacturing specifications of the product and outputs a predicted value of whether or not the product passes the manufacturing process. A process passage determination logic creation means;
By applying the manufacturing specifications of the product to the process passage determination logic, obtaining a predicted value of whether or not each manufacturing process of the product passes, and arranging the acquired predicted values in the predetermined order, the product And a first variety creating means for creating a variety of
The second product type determination means applies the manufacturing specification included in the order data to the process passage determination logic, acquires a predicted value of whether or not the order has passed, and uses the acquired predicted value as the predetermined value. by arranging in order, manufacturing period prediction device, characterized by further have a second cultivar creation means for creating a variety of the orders. A target assortment achievement rate setting means for setting a target assortment achievement rate, which is a target value of the ratio of the amount of products to be delivered by the delivery date with respect to the quantity of the order,
The manufacturing input period according to product type input order includes the product type determined by the second product type determination unit, the manufacturing time distribution by product type calculated by the manufacturing method model output unit by product type, and the target assortment achievement rate. The manufacturing lead time prediction apparatus according to claim 1, wherein a manufacturing lead time for each order is derived using the target assortment achievement rate set by the setting means. The process-specific work period distribution deriving means is expressed as a distribution of the work period of the manufacturing process so that the value of the evaluation function that decreases as the product of the likelihood of obtaining the actual work period of each product increases is minimized. Deriving the mean and variance of the normal distribution
Each of the likelihood that the actual work period of each product is obtained is a normal distribution in which the average value and variance of the average and variance of the work period distribution of the manufacturing process shown to have passed in the passing process pattern are the average and variance. The manufacturing period prediction device according to claim 1 or 2 , characterized in that: The passing process pattern-specific work period distribution deriving means adds the average of the work process distribution and the variance of the manufacturing process that has been shown to have passed in the passing process pattern, respectively, thereby calculating the average of the work schedule distribution by the passing process pattern. And means for deriving a distribution parameter for each passing process pattern for deriving the variance,
Logarithmic normal distribution deriving means for deriving a log normal distribution having a mean and variance derived by the passing process pattern-specific work period distribution parameter deriving means;
Using the log normal distribution obtained by the log normal distribution deriving means as a probability density function, the probability of occurrence of each construction period of the product having the pass process pattern is derived for each of the pass process patterns by deriving for each pass process pattern. The manufacturing period prediction apparatus according to claim 3 , further comprising manufacturing period occurrence probability deriving means for deriving a period distribution. The process-specific work period distribution deriving means is expressed as a distribution of the work period of the manufacturing process so that the value of the evaluation function that decreases as the product of the likelihood of obtaining the actual work period of each product increases is minimized. Deriving the mean and standard deviation of the normal distribution
Each of the likelihood that the actual work period of each product is obtained is a logarithmic normal whose average and variance are the integrated values of the average and variance of the work period distribution of the manufacturing process shown to have passed in the passing process pattern. The production period prediction device according to claim 1 or 2 , characterized in that it is expressed by a distribution. The passing process pattern-specific work period distribution deriving means adds the average of the work process distribution and the variance of the manufacturing process that has been shown to have passed in the passing process pattern, respectively, thereby calculating the average of the work schedule distribution by the passing process pattern. And means for deriving a distribution parameter for each passing process pattern for deriving the variance,
Logarithmic normal distribution deriving means for deriving a log normal distribution having a mean and variance derived by the passing process pattern-specific work period distribution parameter deriving means;
Using the log normal distribution obtained by the log normal distribution deriving means as a probability density function, the probability of occurrence of each construction period of the product having the pass process pattern is derived for each of the pass process patterns by deriving for each pass process pattern. 6. The manufacturing period prediction apparatus according to claim 5 , further comprising manufacturing period occurrence probability deriving means for deriving a period distribution. The process-specific work period distribution deriving means is expressed as a distribution of the work period of the manufacturing process so that the value of the evaluation function that decreases as the product of the likelihood of obtaining the actual work period of each product increases is minimized. Deriving the average of Poisson distribution
Each likelihood that the actual work period of each product is obtained is expressed by a Poisson distribution with the average of the respective integrated values of the work period distributions of the manufacturing process shown to have passed in the passing process pattern. The manufacturing period prediction device according to claim 1 or 2 , wherein The passing process pattern-by-passing process period distribution deriving means derives the average of the passing process pattern-by-passing process pattern by adding the averages of the manufacturing process distributions of the manufacturing processes shown to have passed in the passing process pattern. A means for deriving work period distribution parameters for each passing process pattern,
Poisson distribution deriving means for deriving a Poisson distribution having the same average as the average derived by the passage process pattern-specific work period distribution parameter deriving means,
Using the Poisson distribution obtained by the Poisson distribution deriving means as a probability density function, the occurrence probability of each construction period of the product having the passage process pattern is derived for each of the passage process patterns by deriving the construction period distribution for each passage process pattern. The manufacturing period prediction device according to claim 7 , further comprising manufacturing period occurrence probability deriving means for deriving. The process-specific work period distribution deriving means is expressed as a distribution of the work period of the manufacturing process so that the value of the evaluation function that decreases as the product of the likelihood of obtaining the actual work period of each product increases is minimized. Deriving the shape and scale parameters of the gamma distribution
Each of the likelihood that the actual work period of each product is obtained is defined as the shape parameter that is the integrated value of the shape parameter of the work period distribution of the manufacturing process that is shown to have passed in the passing process pattern. manufacturing period prediction device according to claim 1 or 2, characterized in that it is represented by a gamma distribution to scale parameter a constant value that does not depend on. The passing process pattern-specific work period distribution deriving means adds the shape parameters of the manufacturing process distribution of the manufacturing process that are shown to have passed in the passing process pattern, respectively, to thereby form the work schedule distribution by the passing process pattern A means for deriving a construction period distribution parameter by passing process pattern for deriving a parameter,
Gamma distribution deriving means for deriving a gamma distribution with the same shape parameter and derived shape parameter by the passing step pattern specific implementation schedule distribution parameter deriving means,
Wherein the gamma distribution obtained by the gamma distribution deriving means as a probability density function, the probability of occurrence of each work period of a product having the passage step pattern, the passage step pattern another implementation schedule distribution by deriving for each of the passing step pattern The manufacturing period prediction device according to claim 9 , further comprising manufacturing period occurrence probability deriving means for deriving. A manufacturing period prediction method for predicting the period of an order from order data including manufacturing specifications for each order,
Data indicating the manufacturing performance of each product manufactured through a plurality of manufacturing processes, including the manufacturing specifications of the product, the presence / absence of passing results as information indicating the actual value of the presence / absence of passing of each manufacturing process, and the product Manufacturing result data acquisition step for acquiring manufacturing result data including a result work period that is information indicating the result of the work period required to manufacture the product,
By using the past results for each product and the actual work period, a process-specific work period distribution derivation process for deriving parameters defining the distribution of work periods by manufacturing process, and
By arranging the values of the passing results presence / absence in a predetermined order by using the parameters that define the distribution of the work times by manufacturing steps derived by the process-related work time distribution deriving step and the passing results presence / absence for each product. The work schedule distribution derivation process for each passing process pattern to derive the work schedule distribution for each passing process pattern,
A product included in the manufacturing performance data is a product type obtained by arranging predicted values of the presence or absence of passing through each manufacturing process of the product in the predetermined order using the manufacturing specifications of the product included in the manufacturing performance data. A first variety determining step for determining each of the
Using the number of products for each passing process pattern and the number of products for each kind, deriving the occurrence probability of the passing process pattern for each kind, passing the process pattern occurrence rate for each kind,
The production period model derivation process by product type for deriving the work period distribution by product type, using the work process time distribution by pass process pattern and the occurrence probability of the pass process pattern by product type,
Using the manufacturing specification included in the order data, a predicted value of whether or not the order included in the order data passes through each manufacturing process is obtained, and the predicted value of whether or not the order passes through each manufacturing process is determined as the predetermined value. A second type determining step for determining the type obtained by arranging in order for each of the orders included in the order data;
Using the type determined in the second type determining step and the type-specific manufacturing period distribution derived in the type-specific manufacturing period model deriving step, the manufacturing period by type input order for deriving the manufacturing period by order A derivation step,
Probability density determined based on the parameters obtained by adding the parameters that define the distribution of the construction period of the manufacturing process shown to have passed in the passage process pattern, in the construction process distribution derivation process for each passing process pattern. As a function, derive the work schedule distribution by the passing process pattern ,
In the first product type determination step, a process pass determination logic is generated for each of the manufacturing steps, with the manufacturing specification of the product as an input and a predicted value of whether or not the product passes through the manufacturing step as an output. The process passage determination logic creation process,
By applying the manufacturing specifications of the product to the process passage determination logic, obtaining a predicted value of whether or not each manufacturing process of the product passes, and arranging the acquired predicted values in the predetermined order, the product And a first varieties creation process for creating varieties of
In the second product type determination step, the manufacturing specification included in the order data is applied to the process passage determination logic to obtain a predicted value of whether or not the order has passed, and the acquired predicted value is used as the predetermined value. by arranging in order, manufacturing period prediction method characterized by further have a second cultivar creation step of creating a variety of the orders. A target assortment achievement rate setting step for setting a target assortment achievement rate, which is a target value of the ratio of the amount of products to be delivered by the delivery date with respect to the quantity of the order,
The production time derivation process for each type input order includes the product type determined in the second product type determination step, the work period distribution for each product type derived in the product type production process model derivation step, and the target assortment achievement rate. by using the set target load assortment Percent by setting step, manufacturing period prediction method of claim 1 1, wherein the deriving an order different manufacturing period. The process-specific work period distribution derivation step is expressed as the work period distribution of the manufacturing process so that the value of the evaluation function that becomes smaller as the product of the likelihood of obtaining the actual work period of each product increases is minimized. Deriving the mean and variance of the normal distribution
Each of the likelihood that the actual work period of each product is obtained is a normal distribution in which the average value and variance of the average and variance of the work period distribution of the manufacturing process shown to have passed in the passing process pattern are the average and variance. The manufacturing method forecast method according to claim 11 or 12 , characterized in that: The passing process pattern-by-passing process pattern distribution deriving step is the average of the passing process pattern-by-passing process pattern by adding the average and variance of the manufacturing process distributions shown to have passed in the passing process pattern, respectively. And a process distribution parameter derivation process for each passing process pattern for deriving the variance, and
A log normal distribution derivation step for deriving a log normal distribution having a mean and variance derived by the passing process pattern-specific work period distribution parameter derivation step;
Using the log normal distribution obtained by the log normal distribution derivation step as a probability density function, the occurrence probability of each construction period of the product having the pass step pattern is derived for each of the pass step patterns by deriving for each of the pass step patterns. The manufacturing period prediction method according to claim 13 , further comprising a manufacturing period occurrence probability deriving step for deriving a period distribution. The process-specific work period distribution derivation step is expressed as the work period distribution of the manufacturing process so that the value of the evaluation function that becomes smaller as the product of the likelihood of obtaining the actual work period of each product increases is minimized. Deriving the mean and standard deviation of the normal distribution
Each of the likelihood that the actual work period of each product is obtained is a logarithmic normal whose average and variance are the integrated values of the average and variance of the work period distribution of the manufacturing process shown to have passed in the passing process pattern. The manufacturing method forecast method according to claim 11 or 12 , characterized by being expressed by distribution. The passing process pattern-by-passing process pattern distribution deriving step is the average of the passing process pattern-by-passing process pattern by adding the average and variance of the manufacturing process distributions shown to have passed in the passing process pattern, respectively. And a process distribution parameter derivation process for each passing process pattern for deriving the variance, and
A log normal distribution derivation step for deriving a log normal distribution having a mean and variance derived by the passing process pattern-specific work period distribution parameter derivation step;
Using the log normal distribution obtained by the log normal distribution derivation step as a probability density function, the occurrence probability of each construction period of the product having the pass step pattern is derived for each of the pass step patterns by deriving for each of the pass step patterns. The manufacturing period prediction method according to claim 15 , further comprising a manufacturing period occurrence probability deriving step for deriving a period distribution. The process-specific work period distribution derivation step is expressed as the work period distribution of the manufacturing process so that the value of the evaluation function that becomes smaller as the product of the likelihood of obtaining the actual work period of each product increases is minimized. Deriving the average of Poisson distribution
Each likelihood that the actual work period of each product is obtained is expressed by a Poisson distribution with the average of the respective integrated values of the work period distributions of the manufacturing process shown to have passed in the passing process pattern. The manufacturing construction period prediction method according to claim 11 or 12 , characterized in that: In the passing process pattern-specific work schedule distribution deriving step, the average of the work schedule distribution by the passing process pattern is derived by adding together the average of the work schedule distribution of the manufacturing process shown to have passed in the passing process pattern. A process distribution parameter derivation process for each passing process pattern,
A Poisson distribution derivation step for deriving a Poisson distribution having the same average as the average derived by the passing process pattern-specific work period distribution parameter derivation step;
Using the Poisson distribution obtained by the Poisson distribution derivation process as a probability density function, the occurrence probability of each construction period of the product having the passage process pattern is derived for each of the passage process patterns by deriving the construction period distribution for each passage process pattern. The manufacturing period prediction method according to claim 17 , further comprising: a manufacturing period occurrence probability deriving step for deriving. The process-specific work period distribution derivation step is expressed as the work period distribution of the manufacturing process so that the value of the evaluation function that becomes smaller as the product of the likelihood of obtaining the actual work period of each product increases is minimized. Deriving the shape and scale parameters of the gamma distribution
Each of the likelihood that the actual work period of each product is obtained is defined as the shape parameter that is the integrated value of the shape parameter of the work period distribution of the manufacturing process that is shown to have passed in the passing process pattern. The manufacturing period prediction method according to claim 11 or 12 , characterized by being expressed by a gamma distribution having a constant value not depending on the scale parameter. The process distribution derivation process for each passing process pattern is the shape of the schedule distribution for each passing process pattern by adding together the shape parameters of the process distribution of the manufacturing process shown to have passed in the passing process pattern. The process distribution parameter derivation process for each passing process pattern to derive the parameter,
A gamma distribution deriving step of deriving a gamma distribution with the same shape parameter and the passage step pattern specific implementation schedule distribution parameter derived shape parameter derived by the process,
Wherein the gamma distribution obtained by the gamma distribution deriving process as a probability density function, the probability of occurrence of each work period of a product having the passage step pattern, the passage step pattern another implementation schedule distribution by deriving for each of the passing step pattern The manufacturing period forecasting method according to claim 19 , further comprising: a manufacturing period occurrence probability deriving step for deriving. A computer program that causes a computer to execute each step of the manufacturing period prediction method according to any one of claims 1 to 20 .

Images