JP5402621B2 - Manufacturing load prediction apparatus, method, computer program, and computer-readable storage medium - Google Patents

Description

  In a manufacturing plant that manufactures a plurality of products having different product attributes by processing in a plurality of manufacturing processes, products having similar product attributes are grouped and defined as a plurality of varieties. This technique is suitable for predicting the production load and determining the quality of the production plan.

  In manufacturing processes in many industries including the steel industry, it is required to manufacture products appropriately according to the contents of orders from customers. For example, steel products in the steel manufacturing industry have a wide variety of product order information such as standards that prescribe the materials and sizes according to customer applications. For this reason, the manufacturing specifications of steel products, for example, the chemical composition, the forming method, the heat treatment method, etc. are extremely diverse. Further, it is required to deliver a product having the product order information requested by the customer in the requested amount on the requested date.

  On the other hand, in the manufacturing process, from the viewpoint of improving productivity by mass production, it is required to produce a plurality of orders with the same manufacturing specification in batch units. The manufacturing process of steel products includes a plurality of manufacturing facilities such as steel making, rolling, refining, and shipping, and the optimum lot conditions (conditions for combining a plurality of orders into lots) of each manufacturing facility are different. Therefore, “pursuit of productivity improvement by lot sizing” for a certain manufacturing facility reduces the productivity of other manufacturing facilities, or lot sizing in the upstream process leads to concentration of manufacturing load in the downstream process, resulting in an increase in work in progress. Or increase the manufacturing period. Therefore, it is required to create a lot considering the trade-off between manufacturing facilities. In addition, pre-fabrication of a product for producing a proper-sized lot causes extra product inventory and a corresponding increase in work period. In the present specification, the production load refers to a processing amount (production amount) in each production facility.

In contrast, the steel industry often performs the following production management. First, the sales department receives an order inquiry from a customer and, based on the manufacturing specifications of the order and the requested delivery date, negotiates whether the order is accepted and if necessary, and then instructs the steelworks to produce the order information. . Steelworks will consider more detailed manufacturing specifications based on order information instructed by the sales department. In the steelworks, the production lots and production timings of each production facility are planned based on the production specifications while considering the delivery date, production period, productivity in the production process, and the like, and production instructions are issued to each production facility.
In this way, although rough delivery date negotiations are being conducted at the order receiving stage, the complexity of the manufacturing process as described above, combined with the size and complexity of the production management that handles large quantities of various types and small lot orders, are combined. Currently, it is difficult for the sales department to appropriately negotiate the delivery date in consideration of the grouping of lots at each manufacturing facility and the manufacturing load at the order receiving stage.

  In addition, after order information is instructed from the sales department to the steelworks, that is, after the delivery date target is determined, the determination of the timing for starting production is an important planning factor. For example, in a steelmaking facility, it is desired to produce orders of the same components as much as possible from the viewpoint of productivity and yield, but on the other hand, it is necessary to level the production load in the processes after steelmaking. Therefore, it is necessary to devise an optimal plan with the entire manufacturing process as an integrated process so that a large lot in the steelmaking facility and a leveling of the manufacturing load in the downstream process are compatible. However, in reality, because of the variety of products, the production flow is diversified, and the production load of refining equipment (one of the production equipment), such as the maintenance of quality defects such as defects found in inspection equipment, is the product. It is difficult to predict the manufacturing load accurately at the planning stage and to take countermeasures because it is not determined until after the inspection. For this reason, as a result, the production load after production increases, and the work period may increase due to an increase in work in progress.

In such a situation, at the current production site, the manufacturing process manager classifies the products into rough varieties based on the characteristics of the manufacturing process, and the actual average value of the manufacturing load for each type is used to predict the manufacturing load. The method of using is often used.
Further, in Patent Document 1, in an assembly processing line for high-mix low-volume production, a group of products with similar requirements for the product component configuration or production line is handled as a group of the same product, and the load on the production line is appropriately leveled. A production instruction quantity leveling device has been proposed.
Further, in Patent Document 2, in a manufacturing process for mass production of a large variety of small lot products, a logic for grouping products having similar processing steps is created by a decision tree based on manufacturing performance data, and is grouped. There has been proposed a method of using the average value of the production load for each product type for the production load prediction.

JP-A-10-138102 JP 2008-27150 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 above-described method for predicting the manufacturing load at the current production site, the classification is roughly made into groups of varieties based on the knowledge about the manufacturing process. For this reason, there is a problem that the accuracy of predicting the manufacturing load is not always sufficient, for example, the manufacturing load varies depending on the product even in the same product type.
In addition, the production instruction quantity leveling device disclosed in Patent Document 1 creates a group of varieties in that a plurality of varieties having similar requirements for the product component configuration or production line are handled as one cultivar group. The conditions are clear. However, it is difficult to predict the production load such as the maintenance of stochastic quality defects such as defects found in inspection facilities, such as work processes that occur probabilistically, and the conditions for creating product groups from product attributes There is a problem that is not clearly described.

  Moreover, in the manufacturing load prediction apparatus disclosed in Patent Document 2, since the decision tree model is used for the logic for grouping products having similar processing patterns in all manufacturing processes, the number of processing facilities is large. In the manufacturing process, there is a problem that the number of processing patterns (combination of manufacturing processes) in a plurality of steps increases, that is, the objective variable of the decision tree increases, resulting in a large model structure and a difficult to manage model. Furthermore, when creating a decision tree model, it is difficult to perform detailed learning specialized for each manufacturing process because it is impossible to select explanatory variables for each manufacturing process.

  The present invention has been made in view of such problems, and is a product production plan in a manufacturing factory that manufactures a plurality of types of products through processing in a plurality of manufacturing steps, each manufacturing facility In order to quickly and accurately predict the production load for each product type, which is essential when creating production plans that meet manufacturing requirements such as capacity constraints, a decision tree for creating product types is manufactured. By creating separately, it is possible to select explanatory variables for each manufacturing process, to suppress the large-scale decision tree, and to perform detailed model learning specialized for each manufacturing process For the purpose.

The manufacturing load prediction apparatus of the present invention uses the manufacturing load in each manufacturing process of a manufacturing factory that manufactures a plurality of products having different product attributes by processing in a plurality of manufacturing processes as manufacturing result data for products manufactured in the past. A production load prediction apparatus for predicting based on product order record information including at least the size and weight of a product manufactured in the past as a product attribute, and a record of presence / absence of processing of the product in each manufacturing step And a plurality of varieties for bringing together products having the same or a predetermined range of product attributes based on the manufacturing performance data A decision tree created for each manufacturing process and classifying the product into one of the multiple varieties for each manufacturing process is created as the type classification logic for each process. Based on the production type classification logic creation means and the manufacturing performance data, the production type classification logic is used for each production process, for each type, for the type of product for the total number of products belonging to the type. Manufacturing load prediction model creating means for creating a model for calculating an occurrence rate indicating the expected value of the ratio of the number of passing through the manufacturing process of the product to which the product belongs, as a manufacturing load prediction model, the product type classification logic by process, and the manufacturing load Production load prediction means for predicting the production load of each of the production steps for a production plan related to a new product order using a prediction model is provided.
In addition, the manufacturing load prediction method of the present invention provides the manufacturing load for each product manufactured in the past as the manufacturing load in each manufacturing process of a manufacturing factory that manufactures a plurality of products having different product attributes by processing in a plurality of manufacturing processes. A method for predicting a manufacturing load based on data, which includes product order record information including at least the size and weight of a product manufactured in the past as product attributes, and whether or not the product is processed in each manufacturing process. A plurality of varieties for bringing together products having the same or a predetermined range of product attributes based on the manufacturing result data and an input step of inputting manufacturing result information including at least the actual results as manufacturing result data A decision tree is created for each manufacturing process and the product is classified into one of the plurality of varieties for each manufacturing process. For each manufacturing process, for each product type, for the total number of products belonging to the product type, based on the production result data and the production result data, A production load prediction model creating step of creating a model for calculating an occurrence rate indicating an expected value of the ratio of the number of passing through the production process of the product belonging to the product type as a production load prediction model; A production load prediction step of predicting the production load of each of the production steps for a production plan related to ordering a new product using a production load prediction model is provided.
Furthermore, the computer program according to the present invention converts the manufacturing load in each manufacturing process of a manufacturing factory that manufactures a plurality of products having different product attributes by processing in a plurality of manufacturing processes into manufacturing performance data for products manufactured in the past. A computer program for causing a computer to function as a manufacturing load prediction device for predicting based on product order performance information including at least the size and weight of a product manufactured in the past as product attributes, and each of the products Products that have the same or a predetermined range of product attributes based on the manufacturing result data and an input unit that inputs manufacturing result information including at least the results of presence or absence of processing in the manufacturing process as the manufacturing result data A plurality of varieties are created for each manufacturing process, and the product is assigned to each manufacturing process. The manufacturing process using the process-specific product category logic creation unit for creating a decision tree to be classified into any of a plurality of product types as process-specific product class logic, and the process-specific product class logic based on the manufacturing performance data A model for calculating an occurrence rate indicating an expected value of a ratio of the number of products passing through the manufacturing process to the total number of products belonging to the product for each product is created as a manufacturing load prediction model. A manufacturing load prediction unit that predicts a manufacturing load of each manufacturing process with respect to a production plan related to a new product order using the manufacturing load prediction model creation unit, the product type classification logic according to the process, and the manufacturing load prediction model. When,
A computer is caused to function as a manufacturing load prediction apparatus including:
A storage medium storing the computer program is also provided.

  According to the present invention, since decision trees for creating varieties are created for each manufacturing process, it is possible to suppress an increase in the scale of the decision tree and to select explanatory variables for the decision tree for each manufacturing process. It is possible to perform detailed learning specialized for each manufacturing process. Thereby, it is possible to quickly and highly accurately predict the production load for each type required when creating a production plan that satisfies the production requirements such as capacity constraints of each production facility.

It is the figure which showed an example of the outline of the thick plate manufacturing process in the steel industry. It is the figure which showed an example of schematic structure of the manufacturing load prediction apparatus in 1st Embodiment. It is the flowchart which showed an example of the outline of the manufacturing load prediction procedure of a manufacturing load prediction apparatus. It is the figure which showed an example of product order information. It is the figure which showed an example of manufacture performance information. It is the figure which showed an example of the decision tree of a cutting process. It is the figure which showed an example of the decision tree of a care process. It is the figure which showed an example of the decision tree of the correction process. It is the figure which showed an example of the decision tree of an inspection process. It is the figure which showed an example of the kind according to process. It is the figure which showed an example of leaf ID classified by process calculated | required from manufacture performance. It is the figure which showed an example of the kind name for every kind (leaf ID) of a cutting process, the number of processed sheets, a total number of sheets, and a generation rate (manufacturing load prediction model). It is the figure which showed an example of the kind name for every kind (leaf ID) of a maintenance process, the number of processed sheets, a total number of sheets, and an incidence rate (manufacturing load prediction model). It is the figure which showed an example of the kind name for every kind (leaf ID) of a correction process, the number of processed sheets, a total number of sheets, and an incidence rate (manufacturing load prediction model). It is the figure which showed an example of the kind name for every kind (leaf ID) of an inspection process, the number of processed sheets, a total number of sheets, and an incidence rate. It is the figure which showed an example of new product order information. It is the figure which showed an example of leaf ID classified by process with respect to a new order. It is the figure which showed an example of leaf ID of a new product, an objective variable, and an incidence rate. It is the figure which showed an example of the production plan according to manufacture date and each purchase kind (leaf ID). It is the figure which showed an example of the manufacturing load prediction value of the care process according to the manufacture date and the purchase type (leaf ID). It is the figure which showed an example of the comparison result (prediction precision of the manufacturing load of an inspection process) with the manufacturing load of a test | inspection facility, and a track record value. It is the figure which showed an example of the kind classification logic produced by the prior art. It is the figure which showed an example of schematic structure of the manufacturing load prediction apparatus in 2nd Embodiment. It is the flowchart which showed an example of the outline of the manufacturing load prediction procedure of a manufacturing load prediction apparatus. It is the figure which showed an example of the kind for planning given to each product in new product order information. It is the figure which showed an example of the incidence rate for every kind for planning. It is the figure which showed an example of the production planned quantity (number of sheets) of the product according to manufacture date and the plan article type. It is the figure which showed an example of the manufacturing load estimated value according to manufacture date and manufacturing process. It is the figure which showed an example of the production plan amount (number of sheets) after a change. It is the figure which showed an example of the manufacturing load estimated value after a change. It is the figure which showed the manufacturing load prediction precision of the test | inspection process in 2nd Embodiment.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
First, with reference to FIG. 1, an example of a schematic configuration of a manufacturing factory for a thick steel plate (thick plate), which is an example to which the present invention is applied and is a representative product in the steel industry, will be described. It is needless to say that the application destination of the present invention is not limited to a thick steel plate manufacturing factory. In FIG. 1, arrows indicate the product flow.
In the converter process 1001, the steel output component, which is a chemical component of the steel intermediate product (molten steel) in a high-temperature molten state, is adjusted in units of about 300 [tons], for example, and is discharged into a molten steel pan. The steel output unit in the converter 1001 is called charge.
In the continuous casting process 1002, the molten steel produced in the converter 1001 is continuously cast for a plurality of charges, and then cut to a specified length, for example, a plate-shaped intermediate called a slab of about 20 [ton] units. Manufacture products. A series of production units in the continuous casting process 1002 is called casting. Although it depends on the manufacturing specifications, it is generally manufactured as 8 to 12 charges as one cast.

In the rolling step 1003, the slab is heated and then molded to a predetermined thickness and width. In the refining (cutting) step 1004, the slab after rolling is cut to the size of the order specification. Further, in the refining (correction) step 1005, the slab after rolling is corrected in order to ensure the quality of the shape and the like. In the refining (maintenance) step 1006, the slab after rolling is maintained to ensure quality. The products that have been all processed are placed in the warehouse 1007. A product cut to the size of the product order information is called a plate. Here, in the correction process 1005 and the care process 1006, whether or not it is necessary is determined after manufacturing in the rolling process 1003. Depending on the shape of the product, the correction process 1005 is performed again after the process in the correction process 1005. Therefore, it is difficult to accurately set the manufacturing load for each product at the time of production planning, because reprocessing such as processing in the manufacturing process and processing in the care process 1006 may be necessary.
Therefore, in the first and second embodiments described below, an appropriate manufacturing load that appropriately reflects the actual manufacturing condition is set for each order, and the manufacturing load of the planned production plan is predicted before the start of manufacturing. The quality of the production plan can be evaluated in advance.

<First Embodiment>
FIG. 2 shows an example of a schematic configuration of the production load prediction apparatus 100 that predicts the production load for each product. FIG. 3 shows an example of each step of the production load prediction method performed using the production load prediction apparatus 100. The hardware of the manufacturing load prediction apparatus 100 can be realized by using a computer system (for example, a personal computer) including a CPU, ROM, RAM, HDD, communication interface, user interface, and the like.

(Input unit 1: Input step S1)
The input unit 1 receives as input the product order record information including at least the product size and weight as product attributes and the production record information including at least the record of presence or absence of processing in the product manufacturing process. FIG. 4 shows an example of product order record information 400. In FIG. 4, the product 1 has a standard “a”, a flatness “B”, a cutting specification “present”, a correction specification “present”, an inspection specification “none”, and a manufacturing specification 1 “none”. Indicates that production specification 2 is “Yes”, customer is “B3”, product weight is 4 [ton], product thickness is 12 [mm], product width is 1000 [mm], and product length is 7000 [mm]. . FIG. 5 shows an example of the manufacturing performance information 500. In FIG. 5, the product 1 represents that only the cutting and correction are processed in the refining process (only the cutting process 1004 and the correction process 1005 have passed).
In the input unit 1, for example, information input means (keyboard, electronic file drive, network I / O, etc.) performs input operation by a user, communication with an external device, etc. The actual information 500 is acquired, and the CPU stores the product order actual information 400 and the manufacturing actual information 500 in the HDD or the like according to the program.

(Explanatory variable setting unit 2: explanatory variable setting step S2)
The explanatory variable setting unit 2 sets (predetermined) one or a plurality of product attributes included in the product order record information 400 as explanatory variables. “Explanation variables” and “object variables” will be described later.
The explanatory variable setting unit 2 is realized, for example, when the CPU reads the product order record information 400 stored in the HDD or the like according to the program and extracts information on predetermined product attributes therefrom.

(Objective variable setting unit 3: Objective variable setting step S3)
The objective variable setting unit 3 sets the presence / absence of processing results of each manufacturing process as an objective variable based on the manufacturing results information 500. Specifically, in the present embodiment, the actual value of the presence / absence of processing in each of the processing steps of the cutting step 1004, the care step 1006, the correction step 1005, and the inspection step, which are refining steps, is “1” when processed. If not, the expression “0” is used as the objective variable of each manufacturing process. The inspection process is executed as a process immediately before entering the warehouse 1007. However, the inspection process may be executed at a predetermined timing after entering the warehouse 1007. In the example shown in FIG. 5, the objective variable of the cutting process 1004 of the product 1 is “1”, the objective variable of the care process 1006 is “0”, the objective variable of the correction process 1005 is “1”, and the objective variable of the inspection process is “0”.
The objective variable setting unit 3 is realized, for example, when the CPU reads the manufacturing performance information 500 stored in the HDD or the like according to a program and derives the objective variable from the contents.

(Design parameter setting unit 4: Design parameter input step S4)
The design parameter setting unit 4 sets design parameters required when creating the process-specific product type classification logic. The design parameters will be described later.
In the design parameter setting unit 4, for example, an information input unit included in the manufacturing load prediction device 100 acquires a design parameter by performing an input operation by a user, communication with an external device, and the like. Is stored in an HDD or the like.

Here, “0” and “1” representing the ease of the manufacturing process of the product are assigned to each product as a passage presence / absence variable according to the product attribute of the product. ) ”Will be described. It should be noted that a product whose pass / fail variable of a certain manufacturing process is “1” means that it easily passes through the manufacturing process, and a product whose value is “0” means that it is difficult to pass through the manufacturing 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 closely related to the prediction accuracy and ease of handling of the decision tree.

  Here, the design parameter setting unit 4 sets the parameters related to the structure of the decision tree, such as the number of leaf nodes of the decision tree to be created and the depth of the tree structure, as the design parameters. Specifically, the design parameter setting unit 4 gives an upper limit value of the number of data held by one leaf node. 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.

(Process-specific product category logic creation unit 5: Process-specific product category logic creation step S5)
The process-specific product category logic creation unit 5 creates a decision tree according to the decision tree construction method based on the manufacturing performance information 500, the explanatory variable, the objective variable, and the design parameter. The created decision tree becomes the process-specific kind classification logic.
In the process-specific product type classification logic creation unit 5, for example, the CPU reads the explanatory variables, the objective variables, and the design parameters stored in the HDD or the like in accordance with a program, and uses them as a decision tree construction method. This is realized by performing a process according to the above and creating a decision tree.
(Process-specific product category logic storage unit 6, process-specific product category logic storage step S6)
The process-specific product category logic storage unit 6 stores the process-specific product category logic generated by the process-specific product category logic generation unit 5 in the form of a database, a file, or the like.
The process-specific product category logic storage unit 6 is realized, for example, when the CPU stores the process-specific product category logic generated by the process-specific product category logic generation unit 5 in an HDD or the like according to a program.

6 to 9 show an example of a decision tree constructed by the decision tree construction method. Specifically, FIG. 6 shows an example of the decision tree 600 of the cutting step 1004, FIG. 7 shows an example of the decision tree 700 of the maintenance step 1006, FIG. 8 shows an example of the decision tree 800 of the correction step 1005, FIG. 9 shows an example of a decision tree 900 for the inspection process.
For example, in Fig. 8, the product with a correction specification of “None”, a product length of over 4500 [mm], a product width of 800 [mm] or less, a product weight of over 4 [ton], and a product thickness of 8 [mm] or less Indicates that the product is classified as leaf ID 5 and the objective variable is “0”, that is, the product is difficult to be corrected. The leaf ID means a unique number assigned to each leaf node and is defined as a product type.

(Process-specific product classifying unit 7: Process-specific product classifying step S7)
The process-specific product classifying unit 7 assigns a pass / fail variable (object variable) and a product (leaf ID) to the product order record information 400 according to the process-specific product classification logic (decision trees 600, 700, 800, 900). FIG. 10 shows an example of the result (passage presence / absence variable for each process (objective variable)) given to each product. FIG. 11 shows an example of the result of assigning the product type (leaf ID) (process-specific product type obtained from the manufacturing results). For example, the product 1 included in the product performance information 400 shown in FIG. 4 has the standard “a”, the flatness “B”, the cutting specification “present”, the correction specification “present”, and the inspection specification “none”. "Production specification 1 is" None ", Production specification 2 is" Yes ", Consumer is" B3 ", Product weight is 4 [ton], Product thickness is 12 [mm], Product width is 1,000 [mm], Product Since the length is 7,000 [mm], according to the decision trees 600, 700, 800, and 900 shown in FIGS. 6 to 10, the passage presence / absence variable of the cutting step 1004 is “1”, the type is “1”, and the maintenance step The passage presence / absence variable of 1006 is “0”, the product type is “2”, the passage presence / absence variable of the correction step 1005 is “1”, the product type is “1”, the inspection process pass / fail variable is “0”, and the product type is “4”. It becomes.
In the process-specific product classifying unit 7, for example, the CPU reads the product order record information 400 and the decision trees 600, 700, 800, 900 from the HDD or the like according to the program, and determines the product attributes of each product in the decision trees 600, 700. , 800, 900, and by determining a passing variable and a product type in each manufacturing process of each product.

(Production load prediction model creation unit 8: Production load prediction model creation step S8)
The production load prediction model creation unit 8 calculates, for each product type, an occurrence rate that is an expected value of “the number of processed processes corresponding to the product type” per product belonging to the product type. Specifically, the production load prediction model creation unit 8 totals the number of processed processes and the total number of production processes corresponding to the product type for each product type, and the occurrence rate load_rate [i] for each product type from the following equation (1). [j] is calculated. However, i is an index related to the manufacturing process, i = 0 is cutting, i = 1 is care, i = 2 is correction, i = 3 is inspection, j (j = 0,1,2, ...・) Means variety.

Here, sum_plate [i] [j] is the total number of products belonging to the manufacturing process i (i = 0,1,2,3) and product type j (j = 0,1,2,...). . sum_load [i] [j] is processed in the manufacturing process among products belonging to manufacturing process i (i = 0,1,2,3) and product type j (j = 0,1,2, ...) This is the actual value of the number of processed products.
12 to 15, the results of totaling or calculating the number of processed sheets sum_load [i] [j], the total number of sheets sum_plate [i] [j], and the incidence rate load_rate [i] [j] for each type are shown for each manufacturing process. Shown in
For example, in FIG. 13, the product (leaf ID) “3” in the maintenance process 1006 has a passage presence variable “1”, and the total number of products belonging to the product (leaf ID) “3” is “500”. This means that the number of sheets processed in the maintenance process 1006 is “450 sheets” and the occurrence rate is “0.90” (= 450/500). Here, as shown as an example in FIGS. 12 to 15, models in which the occurrence rate is obtained from the product type are referred to as manufacturing load prediction models 1200, 1300, 1400, and 1500. For example, in the production load prediction model 1300 in the care process shown in FIG. 13, when the type of a certain product is “3”, the output (occurrence rate) 0.90 obtained with the type “3” as an input is the occurrence of the product. Become a rate.
The manufacturing load prediction model creation unit 8 is realized, for example, by the CPU reading the product by process (see FIG. 11) and the manufacturing performance information 500 from the HDD or the like according to the program and performing the calculation of the formula (1). The

(Production load prediction model storage unit 9: Production load prediction model storage step S9)
The production load prediction model storage unit 9 stores the production load prediction model 1200, 1300, 1400, 1500 created by the production load prediction model creation unit 8 in the form of a database, a file, or the like.
The production load prediction model storage unit 9 is realized, for example, by the CPU storing the production load prediction models 1200, 1300, 1400, and 1500 created by the production load prediction model creation unit 8 in an HDD or the like according to a program. .

(Production load prediction unit 10: Production load prediction step S10)
For a new product order, the production load predicting unit 10 includes an order attribute included in the order information (new product order information) of the new product and a process-specific product type classification logic (decision trees 600, 700, 800, 900) is given to each new product. Then, the product load prediction unit 10 assigns, for each new product, an occurrence rate for each manufacturing process obtained from the product type and the manufacturing load prediction models 1200, 1300, 1400, and 1500.
FIG. 16 is a diagram showing an example of new product order information 1600. FIG. 17 is a diagram showing an example of a result of assigning a product type to each new product (product type for each new order).
In FIG. 16, for example, the product 101 has a standard “d”, a flatness “B”, a cutting specification “none”, a correction specification “none”, an inspection specification “present”, and a manufacturing specification 1 “none”. "Production specification 2 is" None ", Customer is" A2 ", Product weight is 5 [ton], Product thickness is 20 [mm], Product width is 900 [mm], Product length is 3,000 [mm] . Therefore, according to the decision trees 600, 700, 800, and 900 shown in FIGS. 6 to 9, the type of the cutting step 1004 is “2”, the type of the maintenance step 1006 is “5”, and the type of the correction step 1005 is “2”. "The type of inspection process is" 1 ".

Moreover, FIG. 18 is the result of giving the incidence rate for every manufacturing process for every new product. This result is obtained from the type (see FIG. 17) assigned to each new product and the production load prediction models 1200, 1300, 1400, 1500 (see FIGS. 12 to 15).
In FIG. 18, for example, in the product 101, the type of the cutting process 1004 is “2”, the type of the maintenance process 1006 is “5”, the type of the correction process 1005 is “2”, and the type of the inspection process is “1”. is there. Therefore, according to the manufacturing load prediction models 1200, 1300, 1400, and 1500 shown in FIGS. 12 to 15, the passage presence / absence variable (object variable) of the cutting step 1004 is “0”, and the occurrence rate is “0.02”, which is the maintenance step. The passage presence / absence variable (objective variable) of 1006 is “1” and the occurrence rate is “0.90”, the passage presence / absence variable (objective variable) of the correction process 1005 is “1”, and the occurrence rate is “0.90”. The presence / absence variable (objective variable) is “1” and the occurrence rate is “1.00”.

Furthermore, the manufacturing load prediction unit 10 can also predict the manufacturing load for each manufacturing date with respect to the production plan in which the manufacturing date of some or all new products included in the new product order information 1600 is determined. . Hereinafter, as a specific example, description will be given focusing on the care process 1006.
FIG. 19 is a diagram showing an example of a production plan 1900 in which new products are aggregated for each production date and for each type of product (procurement leaf ID) 1006. In FIG. 19, for example, the production date 1 indicates that 200 products of the product type “1” are included. FIG. 20 shows a result of calculating the manufacturing load prediction value 2000 of the care process 1006 for each production date and each care leaf ID using the production load prediction model 1300 shown in FIG. For example, in the production plan 1900 shown in FIG. 19, the type “3” and the number of products on the manufacturing date 1 are 800. In the production load prediction model 1300 shown in FIG. 13, the occurrence rate of the type “3” is 0.9. Therefore, as shown in FIG. 20, the predicted production load for the product type “3” and production date 1 is 800 × 0.9 = 720. For other varieties, the predicted production load value can be calculated in the same manner, and the estimated production load value for each production date can be calculated by summing up for each production date.

  The manufacturing load prediction unit 10 is realized by the following processing by hardware, for example. That is, first, the information input means included in the manufacturing load prediction apparatus 100 acquires the new product order information 1600 and the production plan 1900 by performing an input operation by the user, communication with an external apparatus, etc., and the CPU according to the program. The process-specific product type classification logic (decision trees 600, 700, 800, 900) and the production load prediction models 1200, 1300, 1400, 1500 are read from the HDD or the like. Then, according to the program, the CPU assigns a product type to each new product using the new product order information 1600 and the process-specific product type classification logic (decision trees 600, 700, 800, 900), the product type, and the production load. An occurrence rate for each manufacturing process obtained from the prediction models 1200, 1300, 1400, and 1500 is assigned to each new product. Further, the CPU uses the production plan 1900 in which the production date of the new product is determined according to the program and the production load prediction models 1200, 1300, 1400, 1500, and the production load prediction value for each production process and each production date. 2000 is calculated.

  Next, the case where the occurrence rate for each new product derived as described above is used for production planning work will be described. In planning a production plan, the productivity and delivery date in the continuous casting process 1002 are taken into consideration, but it is also important to consider the manufacturing load in the subsequent process after the continuous casting facility. Therefore, based on the occurrence rate of each product included in the production plan once formulated, the production load of each production facility is predicted, and the validity of the production plan is evaluated by comparing with the capacity and operation schedule of each production facility. If there is a problem, actions such as changing the timing of steel production can be taken before production.

  In addition, when using the manufacturing load prediction models 1200, 1300, 1400, and 1500 to evaluate the validity of the production plan, the accuracy is important. FIG. 21 shows an example of a comparison result between the manufacturing load of the inspection equipment predicted using the manufacturing load prediction models 1200, 1300, 1400, and 1500 used in the present embodiment and the actual value. In this evaluation, product order record information and manufacturing record information for 10 months are used, and product category logic and production load prediction model are created using the data for the first half 7 months, and the data for the latter half 3 months are used. The accuracy was evaluated. The horizontal axis of FIG. 21 represents the period when the unit of 5 days is set to the first season, and the vertical axis represents the actual number of processed sheets generated at the inspection facility and the predicted value. From the results shown in FIG. 21, the standard error, which is the standard deviation of the error, is about 188 [sheets / 5 days], which is about 8.5 [%] standard error with respect to the average processed number of 2200 sheets. Therefore, it can be seen that the prediction can be made with high accuracy.

  Further, FIG. 22 shows a product type classification logic created by the method disclosed in Patent Document 2. In the method disclosed in Patent Document 2, a passing process pattern of a combination of “1” and “0” corresponding to the presence / absence of passing of each manufacturing process for all processes (cutting, care, correction, inspection) is an objective variable. It is said. For example, the objective variable “1000” means that it is processed only by cutting. Comparing the method of this embodiment (FIGS. 6 to 9) and the conventional method disclosed in Patent Document 2 (FIG. 22), the kind classification logic (decision trees 600, 700, The number of branches of 800, 900) is 1 to 6 in each manufacturing process, whereas the number of branches of the type division block of the conventional method is 11 or more. Therefore, it can be seen that the method of the present embodiment can create a small and simple kind classification logic.

  In addition, in the conventional method, when it is desired to improve the prediction accuracy of a specific manufacturing process, it is necessary to recreate the entire product category logic, which may adversely affect other manufacturing processes that do not need to improve the prediction accuracy. there were. On the other hand, in the method of this embodiment, since the kind classification logic is created for each manufacturing process, it is possible to recreate the kind classification logic only for the focused manufacturing process without affecting other manufacturing processes. It is. Furthermore, since the kind classification logic is created for each manufacturing process, it is possible to set explanatory variables and design parameters for each manufacturing process, and fine learning specialized for each manufacturing process is possible. For example, when it is desired to improve only the prediction accuracy of the maintenance process 1006, an action such as increasing the explanatory variable or reducing the upper limit value of the number of data held by the leaf node may be performed by specializing in the maintenance process 1006. It becomes possible. In addition, it is possible to evaluate the quality of the production plan in advance, and it is possible to make a production plan that accounts for productivity, delivery date, cost, and the like.

  As described above, in the present embodiment, when predicting a manufacturing load that is the number of products manufactured in each manufacturing process in a manufacturing factory that manufactures a plurality of products having different product attributes by processing in a plurality of manufacturing processes, Perform the process. First, the manufacturing result information 500 including at least the results of the presence or absence of processing in each manufacturing process of the product and the manufacturing result order information 400 including at least the size and weight of the product manufactured in the past as product attributes are input. Then, based on the manufacturing result order information 400 and the manufacturing result information 500, the product type classification logic for each process (decision trees 600 and 700 for classifying products having the same product attributes or within a predetermined range as the same product type. , 800, 900) for each manufacturing process. In addition, based on the manufacturing result order information 400 and the manufacturing result information 500, using the process-specific product type classification logic (decision trees 600, 700, 800, 900), the product belonging to the product type for each manufacturing process and product type. A model for calculating an occurrence rate indicating an expected value of the ratio of the number of passages of the manufacturing process of the product belonging to the product to the total number is calculated as a manufacturing load prediction model 1200, 1300, 1400, 1500. Then, each production process is performed for the production plan 1900 related to the order of a new product by using the type-by-process classification logic (decision trees 600, 700, 800, 900) and the production load prediction models 1200, 1300, 1400, 1500. Predict the production load. Accordingly, it is possible to predict an appropriate manufacturing load reflecting the actual manufacturing status.

<Second Embodiment>
Next, a second embodiment of the present invention will be described. In the present embodiment, the content of the manufacturing load prediction unit (manufacturing load prediction step) is different from that of the first embodiment described above. That is, this embodiment is the same as the first embodiment from the input step S1 to the production load prediction model storage step S9. 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. 23 shows an example of a schematic configuration of a manufacturing load prediction apparatus 2300 that predicts a manufacturing load for each type of product for planning. FIG. 24 shows an example of each step of the manufacturing load prediction method performed using the manufacturing load prediction apparatus. Here, the product for planning is a product that groups products with similar manufacturing load in each manufacturing process, and the manufacturing load is controlled by controlling the number of products manufactured by production date for each product type for planning. It is possible to easily create a production plan in consideration.
(Planning Variety Creation Unit 101: Planning Variety Creation Step S101)
In accordance with the process-specific product category logic (decision trees 600, 700, 800, 900), the planning product creation unit 101 determines whether a new product order information 1600 has a pass / fail variable for each manufacturing process (objective). Variable), a combination pattern for all processes of the pass / fail variable assigned for each manufacturing process is created as a planning type, and a planning type is assigned to each new product order information 1600 for each product.

FIG. 25 is a diagram showing an example of a result of assigning a planning product type for each new product to the new product order information 1600.
In FIG. 25, the passage presence / absence variable of each manufacturing process of the product 101 is “0” for the cutting process 1004, “1” for the care process 1006, “1” for the correction process 1005, and “1” for the inspection process ( (See FIG. 18). Therefore, the planning variety is “0111”.
The planning type creation unit 101 reads, for example, the process-specific type classification logic (decision trees 600, 700, 800, 900) and the new product order information 1600 from the HDD or the like, and for each new product, This is realized by generating a variety for planning which is a combination of passage presence / absence variables given for each product manufacturing process.

(Planned product type production load prediction model creation unit 102: Planned product type production load prediction model creation step S102)
The planned product type production load prediction model creation unit 102 calculates, for each planned product type and production process, an average value of the occurrence rate in the production process of products belonging to a plurality of product types included in the planned product type. The calculated average value is used as the planning type and the production rate of the manufacturing process. This occurrence rate is the ratio of the number of new products belonging to the planning product to the number of passing through the manufacturing process with respect to the total number of new products belonging to the planning product for each manufacturing process and planning product. This is an example of the expected value of the occurrence rate, and hereinafter, this occurrence rate is referred to as the planned product type occurrence rate. That is, the planned product type occurrence rate plan_load_rate [i] [k] for the planning variety k (k = 0, 1, 2,...) And the manufacturing process i (i = 0, 1, 2, 3) is The following equation (2) is obtained.

FIG. 26 shows an example of the planned product type production load prediction model 2600. The planned product type production load prediction model 2600 is a model in which the planned product type occurrence rate of each production process is obtained from the planned product type. For example, the composition number of the product maintenance process 1006 belonging to the planning variety “1000” is 3 for the variety “1”, 5 for the variety “2”, 10 for the variety “4”, and “ Since “7” was 6 sheets, the planned product type occurrence rate in the maintenance process 106 for the planning variety “1000” is 0.07 (= (3 × 0.08 + 5 × 0.05 + 10 × 0.05 + 6 × 0.13) / (3 + 5 + 10 + 6)).
The planned product type production load prediction model creation unit 102 is realized by, for example, the following processing by hardware. That is, first, the information input means included in the manufacturing load prediction device 100 acquires new product order information 1600 by performing an input operation by a user, communication with an external device, and the like, and the CPU performs a product by process according to the program. The classification logic (decision trees 600, 700, 800, 900) and the production load prediction models 1200, 1300, 1400, 1500 are read out. Then, the CPU obtains the product type for the new product according to the program and calculates the equation (2).

(Planned product type production load prediction unit 103: Planned product type production load prediction step S103)
The planned product type production load prediction unit 103 predicts the production load for each production date for the production plan 1900 in which the production date of some or all products included in the new product order information 1600 is determined.
FIG. 27 is a diagram illustrating an example of a result (production planned amount 2700) of the number of products collected for each production plan by planned product type and production date. In FIG. 27, for example, production date 1 (first day) represents that 450 types of planning variety “0000” are produced.
Planned product type generation rate plan_load_rate [i] [k] of production process i (i = 0,1,2,3), planning variety k (k = 0,1,2,...) The production date using the production plan quantity plan_amount [k] [t] of the planning variety k (k = 0,1,2, ...) and the production date t (t = 1,2, ...) The estimated production load load [i] [t] for t (t = 1,2,...) and manufacturing process i (i = 0,1,2,3) is expressed by the following equation (3). Find it.

FIG. 28 shows the plan of the production plan quantity 2700 (= plan_amount [k] [t]) for each planned product type and production date shown in FIG. 27 and the planned product type production load prediction model 2600 shown in FIG. The production load predicted amount 2800 (= load [i] [t]) obtained by the expression (3) using the planned product type occurrence rate (= plan_load_rate [i] [k]) is represented. In FIG. 28, for example, it means that the manufacturing load predicted value of the cutting process on the manufacturing date 1 (first day) is 724 sheets.
Next, the case where it is assumed that the daily processing capacity of each manufacturing process is 800 sheets will be considered. As can be seen from FIG. 28, both the correction process 1005 on the manufacturing date 2 (second day) and the care process 1006 on the manufacturing date 3 (third day) both have a processing capacity exceeding 800 sheets. . From the production plan amount 2700 shown in FIG. 27, the second day includes a large number of planning varieties “0010” having a high occurrence rate of the planned product type in the correction process 1005, and the third day of the maintenance process 1006. It can be seen that there are many types of planning products “0100” having a high incidence of planned product type. Therefore, for example, the planning variety “0100” on the third day is moved about 600 sheets on the second day, and the planning variety “0010” on the second day is moved about 600 [sheets] on the third day. FIG. 29 shows the production plan amount 2900 when moved to FIG. 29, and FIG. 30 shows the predicted production load value 3000 at that time. As can be seen from the predicted production load value 3000 shown in FIG. 30, the correction process 1005 on the manufacturing date 2 (second day) and the care process 1006 on the manufacturing date 3 (third day) are below the processing capacity. I understand that. In this way, it is possible to easily control the manufacturing load by collecting the products into planning varieties and creating or correcting the production plan for each production planning varieties.
The planned product type production load prediction unit 103 reads out the production plan from, for example, an HDD and creates a production plan quantity 2700 for each planned product type and production date, and at the same time, the planned product type production load prediction model creation unit 102 This is realized by reading the planned product type production load prediction model 2600 created in step (3) and calculating the equation (3).

  In addition, when using the manufacturing load prediction model 2600 in order to evaluate the validity of a production plan, the precision becomes important. FIG. 31 shows an example of a comparison result between the production load of the inspection facility predicted using the production load prediction model 2600 used in the present embodiment and the actual value. In this evaluation, product order record information and manufacturing record information for 10 months are used, and product category logic and production load prediction models are created using the data for the first seven months, and data for the latter three months are used. The accuracy was evaluated. The horizontal axis in FIG. 31 represents a period when the unit of 5 days is set to the first season, and the vertical axis represents the actual number of processed sheets generated in the inspection facility and the predicted value. From the results shown in FIG. 31, the standard error is about 189 [5 days / day], and even when compared with the accuracy evaluation results of the first embodiment shown in FIG. It can be seen that there is almost no deterioration of the prediction accuracy due to this. In other words, in the present embodiment, the products are grouped into the planning varieties, and the production plan is created and corrected in units of the production planning varieties, so that the manufacturing load can be easily controlled.

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.

DESCRIPTION OF SYMBOLS 1 Input part 2 Explanation variable setting part 3 Objective variable setting part 4 Design parameter input part 5 Process type classification logic creation part 6 Process type classification logic storage part 7 Process type classification part 8 Production load prediction model creation part 9 Production load Prediction model storage unit 10 Manufacturing load prediction unit 101 Planning product type creation unit 102 Planning product type production load prediction model creation unit 103 Production load prediction unit

Claims (8)

A manufacturing load prediction device that predicts the manufacturing load in each manufacturing process of a manufacturing plant that manufactures multiple products with different product attributes by processing in multiple manufacturing processes, based on manufacturing performance data for products manufactured in the past. There,
The production performance data includes product order performance information including at least the size and weight of the product manufactured in the past as product attributes, and production performance information including at least the performance of presence or absence of processing in the manufacturing process of the product. As input means,
Based on the manufacturing performance data, a plurality of varieties for collecting products having the same product attribute or within a predetermined range are created for each manufacturing process, and the products are classified into the plurality of varieties for each manufacturing process. Process-specific product category logic creation means for creating a decision tree to be classified into any of the above as product-specific product category logic;
Based on the manufacturing performance data, the number of passing of the manufacturing process of the product belonging to the product type with respect to the total number of products belonging to the product type for each manufacturing process and for each product type using the process type classification logic. A production load prediction model creating means for creating a model for calculating an occurrence rate indicating an expected value of a ratio as a production load prediction model;
Production load prediction means for predicting the production load of each of the production steps for a production plan related to the order of a new product, using the type-by-process classification logic and the production load prediction model,
A manufacturing load prediction apparatus comprising:   The manufacturing load predicting means applies the product attribute of the new product to the type-specific logic for each process, obtains a production plan amount for each manufacturing process and for each type, and sets the production load to the production plan amount. The manufacturing load according to claim 1, wherein the manufacturing load of each manufacturing process is predicted by multiplying the production plan amount calculated from a prediction model by the occurrence rate at which the manufacturing process and the product match. Load prediction device. The manufacturing load predicting means applies the product attribute of the new product to each of the process type classification logics for each manufacturing process, and obtains the type of the new product for each of the manufacturing processes. Each of the new products is determined as a planning pattern which is a combination pattern of passage presence / absence variables, which is a variable indicating the ease of passing through the manufacturing process of the new product, which is determined according to the obtained product type. Varieties creation means for planning to create about,
With respect to the total number of new products belonging to the planning product type for each manufacturing process and each planning product type, using the occurrence rate for each manufacturing process and each product type included in the manufacturing load prediction model A model for calculating the planned product type occurrence rate, which is an expected value of the rate of occurrence of the new product belonging to the planning product, is shown as the planned product type production load prediction model. Production load prediction model creation means for each type of planned product to be created,
Using the result of applying the product attribute of the new product to the product type classification logic for each process, the production plan amount for each production process and each product type for planning is obtained, and the production plan amount A plan that predicts the production load of each production process by multiplying the production plan quantity calculated from the type production load prediction model and the planned product type occurrence rate in which the production process and the product for planning coincide. Product type production load prediction means,
The manufacturing load prediction apparatus according to claim 1, further comprising: A manufacturing load prediction method that predicts the manufacturing load in each manufacturing process of a manufacturing plant that manufactures multiple products with different product attributes by processing in multiple manufacturing processes, based on manufacturing performance data for products manufactured in the past. There,
The production performance data includes product order performance information including at least the size and weight of the product manufactured in the past as product attributes, and production performance information including at least the performance of presence or absence of processing in the manufacturing process of the product. An input step to enter as
Based on the manufacturing performance data, a plurality of varieties for collecting products having the same product attribute or within a predetermined range are created for each manufacturing process, and the products are classified into the plurality of varieties for each manufacturing process. A process-specific product category logic creation step for creating a decision tree to be classified into any of the above as product-specific product class logic,
Based on the manufacturing performance data, the number of passing of the manufacturing process of the product belonging to the product type with respect to the total number of products belonging to the product type for each manufacturing process and for each product type using the process type classification logic. A production load prediction model creating step for creating a model for calculating an occurrence rate indicating an expected value of the ratio as a production load prediction model;
A production load prediction step for predicting a production load of each of the production steps for a production plan related to an order for a new product using the type-by-process classification logic and the production load prediction model;
A manufacturing load prediction method comprising:   The manufacturing load prediction step applies a product attribute of the new product to the process-specific product type classification logic to determine a production plan amount for each manufacturing process and each product type, The manufacturing load according to claim 4, wherein the production load of each manufacturing process is predicted by multiplying the production plan amount calculated from a prediction model by the occurrence rate at which the manufacturing process and the product type match. Load prediction method. In the manufacturing load prediction step, the product attribute of the new product is applied to each of the product type classification logics for each manufacturing process, and the product type of the new product is obtained for each of the manufacturing processes. Each of the new products is determined as a planning pattern which is a combination pattern of passage presence / absence variables, which is a variable indicating the ease of passing through the manufacturing process of the new product, which is determined according to the obtained product type. Varieties creation step for planning to create about,
With respect to the total number of new products belonging to the planning product type for each manufacturing process and each planning product type, using the occurrence rate for each manufacturing process and each product type included in the manufacturing load prediction model A model for calculating the planned product type occurrence rate, which is an expected value of the rate of occurrence of the new product belonging to the planning product, is shown as the planned product type production load prediction model. A step of creating a production load prediction model for each type of planned product to be created;
Using the result of applying the product attribute of the new product to the product type classification logic for each process, the production plan amount for each production process and each product type for planning is obtained, and the production plan amount A plan that predicts the production load of each production process by multiplying the production plan quantity calculated from the type production load prediction model and the planned product type occurrence rate in which the production process and the product for planning coincide. Product type production load prediction step,
The manufacturing load prediction method according to claim 4, further comprising: As a manufacturing load prediction device that predicts the manufacturing load in each manufacturing process of a manufacturing plant that manufactures a plurality of products with different product attributes by processing in a plurality of manufacturing processes based on manufacturing result data of products manufactured in the past A computer program for causing a computer to function,
The production performance data includes product order performance information including at least the size and weight of the product manufactured in the past as product attributes, and production performance information including at least the performance of presence or absence of processing in the manufacturing process of the product. An input unit to input as,
Based on the manufacturing performance data, a plurality of varieties for collecting products having the same product attribute or within a predetermined range are created for each manufacturing process, and the products are classified into the plurality of varieties for each manufacturing process. A process-specific product category logic creation unit that creates a decision tree to be classified into any of the above as product-specific product category logic,
Based on the manufacturing performance data, the number of passing of the manufacturing process of the product belonging to the product type with respect to the total number of products belonging to the product type for each manufacturing process and for each product type using the process type classification logic. A production load prediction model creation unit for creating a model for calculating an occurrence rate indicating an expected value of a ratio as a production load prediction model;
A production load prediction unit that predicts a production load of each of the production steps for a production plan related to a new product order using the type-by-process classification logic and the production load prediction model;
A computer program that causes a computer to function as a manufacturing load prediction apparatus.   A computer-readable storage medium having recorded thereon the computer program according to claim 7.

Images