Manufacturing support system for predicting characteristics of alloy materials, method for generating prediction model, and computer program

Abstract

To provide a manufacturing support system capable of improving accuracy of a prediction model which predicts characteristics even out of a data range of manufacturing results, and selecting an appropriate manufacturing condition using the prediction model.SOLUTION: According to a program, a processor: obtains a plurality of manufacturing parameters and a measured value of at least one characteristic of alloy material; calculates a pre-prediction value of characteristics showing an approximate value of a target prediction value, which is a target value of the characteristic, based on a first manufacturing parameter, included in the plurality of manufacturing parameters, using a prediction expression which describes a relation between the first manufacturing parameter and the pre-prediction value; calculates a difference between the pre-prediction value and the measured value of the characteristic; trains a model using a second manufacturing parameter and a learning data set including the difference; and generates a learned model which is used for predicting the at least one characteristic.SELECTED DRAWING: Figure 5

Description

Technical Field

[0001] The present disclosure relates to a manufacturing support system for predicting characteristics of alloy materials, a method for generating a prediction model, and a computer program.

Background Art

[0002] A wide variety of industrial materials are used in various technical fields. Industrial materials are classified and tested according to various standards (e.g., JIS, ASTM), but the characteristics of individual products vary due to various factors (composition, manufacturing process, manufacturing conditions (e.g., heat treatment temperature, speed, or time), etc.). Therefore, it is not easy to select or develop a material that best suits the required characteristics. Also, industrially, not only characteristics but also cost, supply stability, product life, etc. can be points for material selection.

[0003] So far, the selection, adjustment, or change of manufacturing conditions has mainly been carried out based on human experience. That is, they have been human-dependent operations. However, with the recent progress of ICT technology, the development of technology to support the selection, adjustment, or change of manufacturing conditions using a computer has been promoted, and the aspect of human-dependent operations is being eliminated.

[0004] Patent Document 1 discloses a material prediction device that calculates the similarity between past manufacturing conditions stored in a performance database and manufacturing conditions to be predicted, generates a prediction model using the similarity, and predicts the material of steel using the prediction model. In this material prediction device, an evaluation function weighted by similarity is used to evaluate the prediction error of the prediction model. The manufacturing conditions of the steel are controlled based on the predicted material of the steel.

[0005] Patent Document 2 discloses a design support device that calculates manufacturing conditions other than those selected by a designer that satisfy the required quality characteristic values based on data in a quality database that stores the manufacturing conditions manufactured in the past and the quality characteristic values obtained from those manufacturing conditions. This design support device calculates an influence coefficient indicating the degree of influence of the manufacturing conditions selected by the designer and the conditions other than those conditions on the required quality characteristic values from the neighboring data of the manufacturing conditions stored in the quality database.

Prior Art Documents

Patent Documents

[0006]

Patent Document 1

Patent Document 2

Summary of the Invention

Problems to be Solved by the Invention

[0007] According to the method for predicting the material of the steel material described in Patent Document 1 and the method for calculating the influence coefficient described in Patent Document 2, in both cases, the obtained predicted values and the calculation results of the influence coefficient are only within the range of past manufacturing results. Therefore, it becomes difficult to obtain a solution outside that range.

[0008] The present invention has been made in view of the above problems, and its object is to improve the accuracy of a prediction model for predicting the characteristics of an alloy material not only within the range of manufacturing conditions in past manufacturing results but also outside that range, and to use that prediction model to select manufacturing conditions that satisfy the design range (or standard range) of target characteristics. It is to provide a manufacturing support system, a method for generating a prediction model, and a computer program.

Means for Solving the Problems

[0009] The manufacturing support system of the present disclosure is a system that predicts at least one property of an alloy material manufactured through a plurality of manufacturing processes in a non-limiting and exemplary embodiment, comprising a processor, a memory storing a program for controlling the operation of the processor, and a storage device storing data including a plurality of manufacturing parameters representing manufacturing conditions of various manufacturing processes and measured values of the at least one property of the alloy material manufactured under the manufacturing conditions of the manufacturing processes. The processor accesses the storage device according to the program, acquires the plurality of manufacturing parameters and the measured values of the at least one property, obtains a preliminary prediction formula describing the relationship between a first manufacturing parameter included in the plurality of manufacturing parameters and a preliminary prediction value of a property representing an approximate value of a target prediction value which is the target value of the at least one property, calculates the preliminary prediction value based on the first manufacturing parameter using the preliminary prediction formula, calculates the difference between the calculated preliminary prediction value and the measured value of the at least one property corresponding to the preliminary prediction value, trains a model using a learning data set including a second manufacturing parameter different from the first manufacturing parameter and the calculated difference included in the plurality of manufacturing parameters, and generates a trained model used to predict the at least one property.

[0010] The method of the present disclosure is a method for generating a prediction model used to predict at least one property of an alloy material manufactured through a plurality of manufacturing processes in a non-limiting and exemplary embodiment, which respectively accesses data including a plurality of manufacturing parameters representing manufacturing conditions of various manufacturing processes and measured values of the at least one property of the alloy material manufactured under the manufacturing conditions of the manufacturing process to obtain the plurality of manufacturing parameters and the measured values of the at least one property, obtains a pre-prediction value calculation formula describing the relationship between a first manufacturing parameter included in the plurality of manufacturing parameters and a pre-prediction value of a property representing an approximate value of a target prediction value which is the target value of the at least one property, calculates the pre-prediction value based on the first manufacturing parameter using the pre-prediction value calculation formula, calculates a difference between the calculated pre-prediction value and the measured value of the at least one property corresponding to the pre-prediction value, and trains a model using a learning data set including a second manufacturing parameter different from the first manufacturing parameter included in the plurality of manufacturing parameters and the difference to generate the prediction model.

[0011] The computer program of the present disclosure is a computer program that causes a computer to generate a prediction model for predicting at least one property of an alloy material manufactured through a plurality of manufacturing processes in a non-limiting and exemplary embodiment. The computer is caused to access data including a plurality of manufacturing parameters each representing manufacturing conditions of various manufacturing processes and measured values of the at least one property of the alloy material manufactured under the manufacturing conditions of the manufacturing processes, and to obtain the plurality of manufacturing parameters and the measured values of the at least one property; obtain a pre-prediction value calculation formula that describes the relationship between a first manufacturing parameter included in the plurality of manufacturing parameters and a pre-prediction value of a property representing an approximate value of a target prediction value that is a target value of the at least one property; calculate the pre-prediction value based on the first manufacturing parameter using the pre-prediction value calculation formula; calculate a difference between the calculated pre-prediction value and the measured value of the at least one property corresponding to the pre-prediction value; and train a model using a second manufacturing parameter different from the first manufacturing parameter included in the plurality of manufacturing parameters and a learning data set including the difference, and generate the prediction model.

Advantages of the Invention

[0012] Exemplary embodiments of the present disclosure improve the accuracy of a prediction model for predicting properties of an alloy material not only within but also outside the range of manufacturing conditions in past manufacturing performance, and enable selection of manufacturing conditions that satisfy a design range of target properties using the prediction model, and provide a manufacturing support system, a method for generating a prediction model, and a computer program.

Brief Description of the Drawings

[0013]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12A

Figure 12B

MODE FOR CARRYING OUT THE INVENTION

[0014] Material manufacturers classify materials according to, for example, JIS standards and accumulate test data for each. For example, aluminum alloys (hereinafter referred to as "aluminum alloys") are classified into A1000 series, A2000 series, A3000 series, A4000 series, A5000 series, A6000 series, and A7000 series based on, for example, the material symbols of JIS standards. Also, for each aluminum alloy, a tensile test conforming to, for example, JIS standards is conducted, and test data such as stress-strain curves is accumulated.

[0015] Each alloy system is known to exhibit mechanical properties (also referred to as "mechanical properties") such as tensile strength, yield strength, and elongation. However, the mechanical properties of individual products can vary depending on the manufacturing process and manufacturing conditions (e.g., heat treatment temperature, speed, or time), even if the composition of the aluminum alloy is the same. It is not easy to efficiently select a product with optimal required characteristics (including not only mechanical properties but also cost, supply stability, product life, etc.) from among a large number of such products. Such problems exist not only for aluminum alloys but also for various other materials such as other alloy materials or polymer materials.

[0016] Alloy materials such as aluminum alloys are manufactured through a plurality of manufacturing processes including various processes. For example, heat-treated aluminum alloy plates (aluminum plates) are manufactured through various manufacturing processes such as casting, homogenization, hot rolling, cold rolling, and annealing. In order to obtain the optimal mechanical properties of the aluminum alloy required for aluminum plates manufactured through such processes, for example, selection and adjustment of manufacturing conditions in the annealing process are required because disturbance conditions such as temperature vary due to seasonal fluctuations. The mechanical properties of alloy materials are predicted based on metallurgical or empirical prediction formulas, and selection and adjustment of manufacturing conditions have been carried out based on those prediction formulas.

[0017] According to the study by the present inventor, the adjustment of various parameters used in the prediction formula being personal and the fact that such adjustment is laborious and time-consuming are major issues. The relationship between manufacturing conditions and mechanical properties can be learned by utilizing machine learning typified by deep learning or statistical models. However, in actual manufacturing sites, from the perspective of preventing the provision of non-conforming defective products to customers, the manufacturing conditions are controlled so as to satisfy the required range of mechanical properties, that is, to stabilize the mechanical properties. As a result, the data accumulated in the database becomes biased. Also, in such control, the manufacturing conditions must be changed after the mechanical properties fluctuate. Therefore, when searching for manufacturing conditions that satisfy the standard range of mechanical properties, unexpected search results may be obtained, and the predicted mechanical properties may deviate significantly from the standard values. Thus, there is a problem that the relationship between manufacturing conditions and mechanical properties cannot be learned correctly. This is due to the interpolation / extrapolation problem that can occur when using a so-called machine learning model.

[0018] In view of such problems, the inventor of the present application has constructed a model for predicting the properties of an alloy material from manufacturing conditions by combining machine learning and a theoretical formula, and using that model, has devised a novel method capable of selecting manufacturing conditions that satisfy the standard range of properties even in the range that cannot be learned by machine learning, that is, in extrapolation.

[0019] Hereinafter, with reference to the accompanying drawings, a manufacturing support system for predicting the properties of an alloy material, a prediction method, and a method for generating a prediction model according to the present disclosure will be described in detail. However, a more detailed description than necessary may be omitted. For example, a detailed description of well-known matters and a redundant description of substantially the same configuration or process may be omitted. This is to avoid making the following description unnecessarily redundant and to facilitate the understanding of those skilled in the art. Also, the same reference numerals may be assigned to substantially the same configuration or process.

[0020] The following embodiments are illustrative, and the manufacturing support system, prediction method, and method for generating a prediction model for predicting the properties of an alloy material according to the present disclosure are not limited to the following embodiments. For example, the numerical values, shapes, materials, steps, the order of those steps, etc. shown in the following embodiments are merely examples, and various modifications are possible as long as there is no technical contradiction. Also, as long as there is no technical contradiction, it is possible to combine one aspect with another aspect.

[0021] FIG. 1 is a block diagram illustrating a schematic configuration of a manufacturing support system 1000 for predicting the properties of an alloy material according to the present embodiment. The manufacturing support system (hereinafter simply referred to as the "system") 1000 includes a database 100 and a data processing device 200. In the present embodiment, the database 100 stores a data group including a plurality of manufacturing parameters each representing the manufacturing conditions of various manufacturing processes and the measured values of at least one property of the alloy material manufactured under the manufacturing conditions of the manufacturing process.

[0022] In the present embodiment, the alloy material is an aluminum alloy. The system 1000 can predict the properties of the aluminum alloy or select the manufacturing conditions that satisfy the standard range of the required properties. However, the system 1000 can be used as a system for supporting the manufacture of various alloy materials other than aluminum alloys. Hereinafter, the manufacturing conditions that satisfy the standard range of the required properties are described as "optimal manufacturing conditions".

[0023] An example of the properties of the alloy material is mechanical properties. The properties of the alloy material may include electrical properties, thermal properties, magnetic properties, optical properties, etc. In the present embodiment, the properties of the aluminum alloy are, for example, mechanical properties such as yield stress YS, ultimate tensile strength, tensile strength TS, elongation EL, elastic modulus (Young's modulus), Poisson's ratio, or YS / TS ratio (yield ratio). Such mechanical properties are obtained, for example, based on the stress-strain curve obtained by a tensile test conforming to the standard. The stress-strain curve shows different shapes depending on the composition of the material, the manufacturing process, the manufacturing conditions, the test conditions, etc.

[0024] The material manufacturer classifies the materials according to, for example, JIS standards and accumulates test data for each of them. Aluminum alloys are classified into A1000 series, A2000 series, A3000 series, A4000 series, A5000 series, A6000 series, and A7000 series based on, for example, the material symbols of JIS standards. Also, for each aluminum alloy, a tensile test conforming to, for example, JIS standards is conducted, and test data such as stress-strain curves is accumulated.

[0025] The material manufacturer can accumulate a vast amount of time-series process data obtained during the manufacturing stage in the database 100 for, for example, several years, 10 years, 20 years or more. The time-series process data can be accumulated in the database 100 in association with design and development information, manufacturing conditions such as heat treatment temperature, speed or time, and climate data during manufacturing, and test data 700. Such a data group is called big data.

[0026] In this embodiment, the plurality of manufacturing processes may include at least one of a raw material blending process, a melting process, a casting process, a homogenization process, a hot rolling process, a hot extrusion process, a hot forging process, a cold rolling process, a foil rolling process, a straightening process, a solution treatment process, a tempering process, and an aging process.

[0027] In this embodiment, the multiple manufacturing conditions used in multiple manufacturing processes include, for example, the type of aluminum alloy, the amount of Si, the amount of Mg, the amount of Cu, the amount of Fe, the hot rolling coiling temperature (°C), the cold rolling coiling temperature (°C), the heat treatment temperature (°C), the heat treatment time (seconds), the coiling temperature after heat treatment (°C), the sheet width (mm) of the sheet material during heat treatment, the sheet thickness (mm) during heat treatment, the weight (kg) of the coil after heat treatment, the number of days of natural aging after heat treatment (days), and the climate data during natural aging. In this specification, these are described as "manufacturing parameters" representing the manufacturing conditions of each manufacturing process. Among the multiple manufacturing parameters, the heat treatment time and the coiling temperature respectively mean the heat treatment time and the coiling temperature in the heat treatment furnace. As equipment for performing heat treatment, for example, a continuous annealing line (CAL) can be used. The heat treatment temperature may be the set temperature of the heat treatment furnace or the temperature obtained by measuring the sheet material during heat treatment.

[0028] In this embodiment, among the multiple manufacturing parameters described above, the heat treatment temperature is described as the "first manufacturing parameter", and those other than the heat treatment temperature are described as the "second manufacturing parameter". Details of the first manufacturing parameter and the second manufacturing parameter will be described later.

[0029] The types of aluminum alloys are roughly classified into heat-treatable alloys and non-heat-treatable alloys. The types of aluminum alloys can be classified based on, for example, the material symbol, alloy symbol, or quality classification symbol of the JIS standard. For example, the types of aluminum alloys can be classified based on the quality classification symbol into those that are cold-worked and then naturally aged (T3) after solution treatment, or those that are naturally aged (T4) after solution treatment.

[0030] The climate data includes air temperature, humidity, sunshine hours, precipitation, etc. For example, these climate data can be obtained based on the meteorological information of the area where the manufacturing factory is located, announced by the Meteorological Agency, and can be the average air temperature of a day, the average humidity of a day, the average air temperature of a week, the average humidity of a week, the average precipitation of a week, the average sunshine hours of a week, etc.

[0031] The database 100 can store, for example, a data group that associates the types of aluminum alloys, the above-described plurality of manufacturing parameters, with mechanical properties such as the yield stress YS, the proof stress, the tensile strength TS, the elongation EL, the elastic modulus, the Poisson's ratio, or the YS / TS ratio. The database 100 is a storage device such as a semiconductor memory, a magnetic storage device, or an optical storage device.

[0032] The data processing device 200 can access the vast amount of data stored in the database 100 to obtain a plurality of manufacturing parameters and the measured values of at least one mechanical property associated therewith.

[0033] The data processing device 200 includes a main body 201 of the data processing device and a display device 220. For example, software (or firmware) used to generate a prediction model for predicting the mechanical properties of an aluminum alloy by utilizing the data stored in the database 100, and software for selecting optimal manufacturing conditions by utilizing the learned prediction model are installed in the main body 201 of the data processing device. Such software is recorded on a computer-readable recording medium such as an optical disk, sold as package software, or provided via the Internet.

[0034] The display device 220 is, for example, a liquid crystal display or an organic EL display. The display device 220 can display, for example, the predicted values of the mechanical properties of the aluminum alloy and / or the optimal manufacturing conditions based on the output data output from the main body 201.

[0035] A typical example of the data processing device 200 is a personal computer. Alternatively, the data processing device 200 can be a dedicated device that functions as a manufacturing support system.

[0036] Figure 2 is a block diagram showing an example of the hardware configuration of the data processing apparatus 200. The data processing apparatus 200 includes an input device 210, a display device 220, a communication I / F 230, a storage device 240, a processor 250, a ROM (Read Only Memory) 260, and a RAM (Random Access Memory) 270. These components are connected to be communicable with each other via a bus 280.

[0037] The input device 210 is a device for converting an instruction from a user into data and inputting it to the computer. The input device 210 is, for example, a keyboard, a mouse, or a touch panel.

[0038] The communication I / F 230 is an interface for performing data communication between the data processing apparatus 200 and the database 100. The form and protocol of the data are not limited as long as the data can be transferred. For example, the communication I / F 230 can perform wired communication compliant with USB, IEEE1394 (registered trademark), or Ethernet (registered trademark), etc. The communication I / F 230 can perform wireless communication compliant with the Bluetooth (registered trademark) standard and / or the Wi-Fi standard. Any of these standards includes a wireless communication standard using a frequency in the 2.4 GHz band or the 5.0 GHz band.

[0039] The storage device 240 is, for example, a magnetic storage device, an optical storage device, a semiconductor storage device, or a combination thereof. Examples of the optical storage device are an optical disk drive or a magneto-optical disk (MD) drive. Examples of the magnetic storage device are a hard disk drive (HDD), a floppy disk (FD) drive, or a magnetic tape recorder. An example of the semiconductor storage device is a solid state drive (SSD).

[0040] The processor 250 is a semiconductor integrated circuit, also referred to as a central processing unit (CPU) or a microprocessor. The processor 250 sequentially executes a computer program stored in the ROM 260 that describes a set of instructions for training a prediction model or utilizing a learned model, thereby realizing a desired process.

[0041] In addition to or instead of the processor 250, the data processing device 200 may include an FPGA (Field Programmable Gate Array) equipped with a CPU, a GPU (Graphics Processing Unit), an ASIC (Application Specific Integrated Circuit), an ASSP (Application Specific Standard Product), or a combination of two or more circuits selected from these circuits.

[0042] The ROM 260 is, for example, a writable memory (e.g., PROM), a rewritable memory (e.g., flash memory), or a read-only memory. The ROM 260 stores a program for controlling the operation of the processor. The ROM 260 does not necessarily have to be a single recording medium and may be a collection of multiple recording media. A part of the multiple aggregates may be a removable memory.

[0043] The RAM 270 provides a working area for temporarily expanding the control program stored in the ROM 260 at boot time. The RAM 270 does not necessarily have to be a single recording medium and may be a collection of multiple recording media.

[0044] Hereinafter, several representative configuration examples of the system 1000 of the present disclosure will be described.

[0045] In the first configuration example, the system 1000 includes the database 100 and the data processing device 200 shown in FIG. 1. The database 100 is a different hardware from the data processing device 200. Alternatively, by loading a storage medium such as an optical disk storing a large amount of data into the main body 201 of the data processing device 200, it is possible to access the storage medium instead of the database 100 and read out a large amount of data.

[0046] In the second configuration example, the system 1000 is a single data processing device 200. In that case, a large amount of test data 700 is previously stored in a storage device 240 such as an HDD. In this configuration example, examples of the data processing device 200 can be a laptop PC, a tablet terminal, a smartphone, or the like.

[0047] FIG. 3 is a hardware block diagram showing a configuration example of a cloud server 300 having a database 340 storing a large amount of data.

[0048] In one aspect of the third configuration example, the system 1000 includes one or more data processing devices 200 and the database 340 of the cloud server 300 as shown in FIG. 3. The cloud server 300 has a processor 310, a memory 320, a communication I / F 330, and a database 340. A large amount of data can be stored in the database 340 on the cloud server 300. For example, the plurality of data processing devices 200 can be connected via a local area network (LAN) 400 built within the company. The local area network 400 is connected to the Internet 500 via an Internet service provider (ISP). Each data processing device 200 can access the database 340 of the cloud server 300 via the Internet 500.

[0049] In another aspect, the system 1000 may include one or more data processing devices 200 and a cloud server 300. In that case, instead of or in cooperation with the processor 250 included in the data processing device 200, the processor 310 included in the cloud server 300 can sequentially execute a computer program that describes a set of instructions for training a prediction model or utilizing a learned model. Alternatively, for example, a plurality of data processing devices 200 connected to the same LAN 400 may cooperate to execute a computer program that describes such a set of instructions. By distributing the processing among a plurality of processors in this way, it becomes possible to reduce the computational load on individual processors.

[0050] FIG. 4 illustrates a general manufacturing flow including a manufacturing process of an aluminum alloy represented by the quality identification symbol T4. In the illustrated example, the plurality of manufacturing processes include a casting process, a homogenization process, a hot rolling process, a cold rolling process, a tempering process, and a natural aging process. In the present embodiment, among these processes, the manufacturing processes from the casting process to the cold rolling process are referred to as the upper processes, and the manufacturing processes including the tempering process and the natural aging process are referred to as the lower processes to distinguish between the two. The tempering process in the lower process can be carried out, for example, using a continuous annealing line.

[0051] The aluminum alloy in the present embodiment is, for example, an A6000 series alloy manufactured according to the T4 manufacturing flow illustrated in FIG. 4. Measured values of mechanical properties such as the yield stress YS, the tensile strength TS, and the elongation EL for each of the plurality of manufacturing parameters described above are stored in the database 100 as manufacturing results.

[0052] FIG. 5 is a functional block diagram showing the functions of the system 1000 processed by the processor 250 in functional block units. FIG. 6 is a flowchart illustrating a processing procedure for generating a prediction model for predicting the mechanical properties of an aluminum alloy.

[0053] The processor 250 has an input unit 251, a pre-predicted value calculation unit 252, a difference calculation unit 253, a prediction model unit 254, a target predicted value calculation unit 255, and an output unit 256. For the sake of convenience in explanation, the word "unit" is appended to each functional block name. For example, the block for calculating the pre-predicted value is denoted as the pre-predicted value calculation unit 252. Typically, the processing (or tasks) of the functional blocks corresponding to each unit are described in a computer program in units of software modules. However, when using an FPGA or the like, all or part of these functional blocks can be implemented as a hardware accelerator.

[0054] The input unit 251 accesses the database 100 and acquires, as input variables, a plurality of manufacturing parameters and the measured values of at least one mechanical property (step S110). The measured values of the mechanical properties are obtained from a tensile test of an aluminum alloy manufactured under the manufacturing conditions of the manufacturing process. In the illustrated example, the input unit 251 acquires the above-described second manufacturing parameter obtained from past manufacturing results as an input variable.

[0055] The input unit 251 acquires, as an input variable, a pre-predicted value calculation formula that can be selected by a user such as an operator or a developer (step S120). The pre-predicted value calculation formula includes a mathematical formula for predicting the mechanical properties in a data-unacquired region that could not be obtained from past manufacturing results, using a metallurgical theoretical formula or empirical formula in materials engineering, and is a formalized know-how of the manufacturing site. Each time manufacturing is carried out, the metallurgical theoretical formula or empirical formula can be accumulated. When generating a prediction model, the user can select one of the accumulated plurality of metallurgical theoretical formulas or empirical formulas as the pre-predicted value calculation formula and input it into the system 1000. Alternatively, a predetermined function without user selectivity may be used as the pre-predicted value calculation formula.

[0056] The input unit 251 further acquires, as input variables, the measured values of the tensile strength TS, which is a mechanical property. The input unit 251 may further acquire the measured values of at least one of the yield stress YS, elongation EL, and proof stress. Hereinafter, the tensile strength, yield stress, proof stress, and elongation are denoted as TS, YS, YS_2, and EL, respectively.

[0057] The input unit 251 accesses the database 100 and acquires a plurality of data sets obtained from past multiple manufacturing processes. Each data set includes a plurality of manufacturing parameters acquired for each coil (or lot) representing a series of processes from the casting process to the natural aging process. Each data set further includes the measured value of TS measured for each coil. The input unit 251 acquires, for example, data sets for 1000 coils.

[0058] The pre-prediction value calculation unit 252 calculates a pre-prediction value based on the first manufacturing parameter using a pre-prediction value calculation formula (step S130). The pre-prediction value calculation formula describes the relationship between the first manufacturing parameter included in the plurality of manufacturing parameters acquired by the input unit 251 and the pre-prediction value of the mechanical property. The pre-prediction value represents an approximate value of the target prediction value, which is the target value of the mechanical property. In the present embodiment, the first manufacturing parameter is the heat treatment time HT. The pre-prediction value P_PTS of TS represents an approximate value of the target prediction value T_PTS, which is the target value of TS. The pre-prediction value calculation formula describes the relationship between the heat treatment time HT and the pre-prediction value P_PTS of TS.

[0059] In the present embodiment, the pre-prediction value calculation formula is given by a quadratic equation of a variable based on the heat treatment time HT, which is the first manufacturing parameter. However, the pre-prediction value calculation formula is not limited to a quadratic equation and may be represented by a linear equation or a mathematical equation of the third degree or higher. Hereinafter, a model example of the pre-prediction value calculation formula that can be used in the case of an aluminum alloy in the heat treatment system is shown. The pre-prediction value calculation formula in this example is represented by Equation 1. [Equation 1] P_PTS = -206.85HT 2 + 544.63HT - 122.61

[0060] The difference calculation unit 253 calculates the difference between the pre-predicted value calculated by the pre-prediction value calculation unit 252 and the measured value of at least one characteristic corresponding to the pre-predicted value (step S140). In the present embodiment, the difference calculation unit 253 calculates the difference ΔTS between the pre-predicted value P_PTS and the measured value of TS corresponding to the pre-predicted value P_PTS. The difference ΔTS represents the difference between the measured value of TS in the manufacturing results and the approximate value of the predicted value of TS, which is the prediction result obtained by using the pre-prediction value calculation formula based on the manufacturing conditions in the manufacturing results.

[0061] The prediction model unit 254 trains a prediction model using a learning data set (or training data) that includes a second manufacturing parameter different from the first manufacturing parameter and the difference ΔTS calculated by the difference calculation unit 253, which are included in a plurality of manufacturing parameters (step S150). In the present embodiment, the second manufacturing parameter includes climate data. An example of the climate data is the past average temperature in the manufacturing results.

[0062] In the present embodiment, the learning data set does not include the first manufacturing parameter that mainly contributes to the input variables input to the pre-prediction value calculation unit 252 among the plurality of manufacturing parameters input to the input unit 251, and includes a second manufacturing parameter different from the first manufacturing parameter. The first manufacturing parameter in the present embodiment is the heat treatment time HT. That is, the learning data set does not include the heat treatment time HT.

[0063] In the present embodiment, the prediction model unit 254 is a supervised prediction model and is constructed by a neural network (NN). An example of the neural network is a multi-layer perceptron (MLP). The MLP is also referred to as a forward propagation type neural network. However, the supervised prediction model is not limited to a neural network and may be, for example, a support vector machine or a random forest.

[0064] Figure 7 is a diagram showing a configuration example of a neural network. The illustrated neural network is an MLP composed of N layers from an input layer as the first layer to an output layer as the Nth layer (final layer). Among the N layers, the second layer to the N-1th layer are intermediate layers (also referred to as "hidden layers"). The number of units (also referred to as "nodes") constituting the input layer is the same n as the number of dimensions of the feature amounts that are input data. In the present embodiment, the input layer is composed of six units corresponding to six manufacturing parameters included in the learning data set. The output layer is composed of four units. In the present embodiment, the number of intermediate layers is three, and the total number of units is 300.

[0065] In the MLP, information propagates unidirectionally from the input side to the output side. One unit receives a plurality of inputs and calculates one output. When the plurality of inputs are [x 1 , x 2 , x 3 , ···, x i (i is an integer of 2 or more)], the total input to the unit is given by a formula of Equation 2, which is obtained by multiplying each input x by a different weight w, adding them, and adding a bias b thereto. Here, [w 1 , w 2 , w 3 , ···, w i are the weights for the respective inputs. The output z of the unit is given by the output of a function f of Equation 3 called an activation function with respect to the total input u. The activation function is generally a monotonically increasing non-linear function. An example of the activation function is the logistic sigmoid function, which is given by Equation 4. e in Equation 4 is the Napier's constant. [Equation 2] u = x 1 w 1 + x 2 w 2 + x 3 w 3 + ··· w i w i + b [Equation 3] z = f(u) [Equation 4] f(u) = 1 / (1 + e ‐u )

[0066] All units included in each layer are coupled between layers. As a result, the output of the units in the left layer becomes the input of the units in the right layer, and signals propagate unidirectionally from the right layer to the left layer through this coupling. By sequentially determining the output of each layer while optimizing the parameters of the weights w and the bias b, the final output of the output layer is obtained.

[0067] The learning dataset further includes the difference ΔTS output from the difference operation unit 253. The difference ΔTS is used as teacher data. The parameters of the weights w and the bias b are optimized based on a loss function (mean squared error) so that the output of the output layer in the neural network approaches the difference ΔTS. In this embodiment, the number of epochs is about 1000 times. The prediction model can be represented by the formula of Equation 5. The above-described second manufacturing parameter, which is an input variable of the neural network, is an explanatory variable and is within the data range in the manufacturing results. The output of the neural network is the target variable. It is preferable to optimize the prediction model using an optimization algorithm such as Adam. [Equation 5] Output (correction value CV) = NN (second manufacturing parameter)

[0068] As a result of training the prediction model, a learned model is generated. The prediction model unit 254 inputs input variables including the average temperature, which is the second manufacturing parameter, into the learned model, and obtains a correction value CV for correcting the difference ΔPTS between the target prediction value T_PTS and the pre-prediction value P_PTS (step S160). The input variables input to the learned model may include other manufacturing parameters (for example, winding temperature, natural aging days) other than the average temperature within the data range in the manufacturing results.

[0069] The correction value CV is the difference between the approximate value (pre-prediction value P_PTS) of the predicted value of TS predicted based on the past manufacturing conditions in the manufacturing results and the target prediction value T_PTS. The correction value CV indicates the estimation error of the target prediction value T_PTS.

[0070] Similar to TS, the prediction model unit 254 obtains a correction value CV for correcting the difference ΔPYS between the target prediction value T_PYS, which is the target value of YS, and the preliminary prediction value P_PTS, a correction value CV for correcting the difference ΔPEL between the target prediction value T_PEL, which is the target value of EL, and the preliminary prediction value P_PTS, and a correction value CV for correcting the difference ΔPYS_2 between the target prediction value T_PYS_2, which is the target value of YS_2, and the preliminary prediction value P_PTS. These four correction values CV are respectively output from the four outputs of the output layer in the neural network.

[0071] The correction value CV for YS indicates the estimation error of the target prediction value T_PYS. The correction value CV for EL indicates the estimation error of the target prediction value T_PEL. The correction value CV for YS_2 indicates the estimation error of the target prediction value T_PYS_2.

[0072] The target prediction value calculation unit 255 calculates the target prediction value from the preliminary prediction value and the correction value based on a mathematical formula that models the relationship between the preliminary prediction value, the correction value, and the target prediction value (step S170). The mathematical formula that models this relationship is represented by a linear equation of the preliminary prediction value P_PTS, and is represented by, for example, the mathematical formulas from Equation 6 to Equation 9. Equation 6 models the relationship between the preliminary prediction value P_PTS, the correction value CV for TS, and the target prediction value T_PTS. Equation 7 models the relationship between the preliminary prediction value P_PTS, the correction value CV for YS, and the target prediction value T_PYS. Equation 8 models the relationship between the preliminary prediction value P_PEL, the correction value CV for EL, and the target prediction value T_PEL. Equation 9 models the relationship between the preliminary prediction value P_PYS_2, the correction value CV for YS_2, and the target prediction value T_PYS_2. [Equation 6] T_PTS = P_PTS + CV [Equation 7] T_PYS = 0.40 * P_PTS + CV + 28 [Equation 8] T_PEL = -0.065 * P_PTS + CV + 44 [Equation 9] T_PYS_2 = 0.42 * P_PTS + CV + 112

[0073] The target predicted value calculation unit 255 calculates the target predicted value T_PTS based on the formula of Equation 6. The target predicted value calculation unit 255 calculates the target predicted value T_PYS based on the formula of Equation 7. The target predicted value calculation unit 255 calculates the target predicted value T_PEL based on the formula of Equation 8. The target predicted value calculation unit 255 calculates the target predicted value T_PYS_2 based on the formula of Equation 9.

[0074] The output unit 256 can output the predicted value of the mechanical properties to a dedicated driver or controller (not shown) of the display device 220 and display it on the display device 220. For example, the calculated target predicted values T_PTS, T_PYS, T_PEL, and T_PYS_2 are displayed on the display device 220.

[0075] In this way, for example, an operator or developer can utilize the manufacturing support system 1000 to predict mechanical properties such as TS, YS, EL, and YS_2 from past manufacturing conditions in the manufacturing results.

[0076] Next, a method for selecting optimal manufacturing conditions using the learned model will be described. By using the learned model described above, it becomes possible to select optimal manufacturing conditions in a range that cannot be learned by machine learning, that is, in the extrapolation range.

[0077] The selection of optimal manufacturing conditions can be performed according to various processing procedures (that is, algorithms) by using the learned model according to the present embodiment. Hereinafter, the first to third implementation examples of the algorithm will be described. A computer program including an instruction group describing these algorithms can be provided via, for example, the Internet. In the following description, the entity that executes each process is assumed to be the data processing device 200 including the processor 250.

[0078] [First implementation example] FIG. 8 is a flowchart showing the processing procedure according to the first implementation example.

[0079] The data processing device 200 acquires at least one input design parameter that defines the design range of the target mechanical properties (step S210). For example, an operator can input, via the input device 210, an input design parameter that defines the design range of the target TS or the target predicted value T_PTS into the data processing device 200. The data processing device 200 can input the input design parameter into the learned model and output a group of manufacturing parameters representing one or more manufacturing conditions that satisfy the design range (step S220).

[0080] According to this implementation example, it becomes possible to obtain a group of manufacturing parameters that satisfy the design range of the target mechanical properties by using a learned model. For example, if an input design parameter that defines the design range of the target TS is input into the data processing device 200, two manufacturing conditions, namely the heat treatment time and the winding temperature that satisfy the design range, can be selected. The selected heat treatment time and winding temperature can be displayed, as a selection result, on, for example, the display device 220 of the system 1000.

[0081] [Second Implementation Example] FIG. 9 is a flowchart showing a processing procedure according to the second implementation example.

[0082] First, the data processing device 200 trains a prediction model by using one or more manufacturing parameters obtained from past manufacturing processes, that is, one or more manufacturing parameters indicating past manufacturing conditions in manufacturing results (S310). This process is the same as the training of the model in step S150 described above.

[0083] Next, the data processing device 200 acquires at least one input design parameter that defines the design range of the target mechanical properties (S320). For example, an operator can input, via the input device 210, an input design parameter that defines the design range of the target TS or the target predicted value T_PTS into the data processing device 200.

[0084] Next, the data processing device 200 inputs at least one of a group of manufacturing parameters representing one or more manufacturing conditions used from the first manufacturing process to the intermediate manufacturing processes that have been performed in the current manufacturing process, and the input design parameters into the learned model (step S330).

[0085] The data processing device 200 acquires a group of downstream manufacturing parameters representing one or more manufacturing conditions that satisfy the design range and are used in the manufacturing process of the downstream process following the intermediate manufacturing process (step S340).

[0086] In a certain manufacturing process that is currently in progress, for example, assume that the upstream process from the casting process, which is the first manufacturing process, to the cold rolling process, which is the intermediate manufacturing process, illustrated in FIG. 4, has been completed. In that case, the data processing device 200 inputs, into the learned model, for example, climate data (average temperature, average humidity, etc.) representing past manufacturing conditions in the manufacturing results from the casting process to the cold rolling process, and the input design parameters. For example, the group of downstream manufacturing parameters may include manufacturing parameters representing manufacturing conditions in the annealing process. The data processing device 200 can acquire, as the output of the learned model, two manufacturing parameters, namely, the heat treatment time that satisfies the design range of the TS and the winding temperature after heat treatment, to be used in the annealing process.

[0087] According to this implementation example, by using the manufacturing conditions of the manufacturing processes that have been performed in the upstream process as constraint conditions and utilizing the learned model, it becomes possible to acquire a group of manufacturing parameters to be used in the downstream process that satisfies the design range of the target mechanical properties.

[0088] [Third Implementation Example] FIG. 10 is a flowchart showing a processing procedure according to the third implementation example.

[0089] First, the data processing device 200 trains a prediction model using one or more manufacturing parameters indicating past manufacturing conditions in the manufacturing results (S410). This process is the same as the training of the model in step S150 described above.

[0090] Next, the data processing device 200 acquires at least one input design parameter that defines the design range of the target mechanical characteristics (S420). For example, an operator can input, via the input device 210, an input design parameter that defines the design range of the target TS or the target predicted value T_PTS to the data processing device 200.

[0091] Next, in the currently ongoing manufacturing process, the data processing device 200 inputs at least one of a group of manufacturing parameters representing one or more manufacturing conditions used up to the intermediate manufacturing process and a first group of downstream manufacturing parameters including predicted values of one or more manufacturing conditions to be used in the downstream manufacturing process following the intermediate manufacturing process, and the input design parameter into the learned model (step S430). Similar to the second implementation example, assume that in a certain ongoing manufacturing process, for example, the upstream process from the casting process to the cold rolling process has been completed. In that case, for example, the data processing device 200 inputs past climate data (such as average temperature, average humidity, etc.) representing the manufacturing conditions in the manufacturing results of the upstream process, predicted values of future climate data (such as average temperature, average humidity, etc.) to be used in the downstream manufacturing process, and the input design parameter into the learned model.

[0092] The data processing device 200 acquires a second group of downstream manufacturing parameters that represent one or more manufacturing conditions different from the one or more manufacturing conditions represented by the first group of downstream manufacturing parameters and that satisfy the design range (step S440). For example, the data processing device 200 can acquire, as the output of the learned model, manufacturing parameters of the number of days of natural aging that are different from the climate data, represent the manufacturing conditions used in the natural aging process, and satisfy the design range of TS.

[0093] According to this implementation example, by using the past manufacturing conditions of the manufacturing process already implemented in the upper process and the lower process as constraint conditions and utilizing the learned model, it becomes possible to obtain a group of future manufacturing parameters for the unimplemented manufacturing process in the lower process that satisfies the design range of the target mechanical properties.

[0094] FIG. 11 is a graph illustrating the relationship between the heat treatment time and the target predicted value T_PTS. The vertical axis represents the target predicted value T_PTS, and the horizontal axis represents the heat treatment time. In the figure, the simulation result by the conventional method using only machine learning is shown by a broken line, and the simulation result by the hybrid method combining machine learning and a logical formula according to this embodiment is shown by a solid line. Note that, unlike this embodiment, in the conventional method, the heat treatment time is included in the training data during machine learning.

[0095] When using the conventional method, the change amount of the target predicted value T_PTS with respect to the change amount of the heat treatment time, that is, the slope of the graph is small. This is because the heat treatment time is included in the training data as a manufacturing parameter, and the heat treatment time is adjusted so that the characteristics of TS fall within the target design range. As a result, the influence that the heat treatment time can exert on other factors (manufacturing conditions) is offset. Therefore, when searching for the optimal manufacturing conditions using the conventional method, there is a possibility that extremely large or small manufacturing conditions (out-of-specification conditions) may be proposed by the system.

[0096] When using the hybrid method according to this embodiment, compared with the conventional method, the change amount of the target predicted value T_PTS with respect to the change amount of the heat treatment time is large. This is because, as a manufacturing parameter, the heat treatment time is included in the input variable of the logical formula instead of the training data, and the influence of the heat treatment time on other factors is reflected in the logical formula. As a result, the influence that the heat treatment time and other factors exert on each other can be separated, and it becomes possible to select appropriate manufacturing conditions even outside the range of machine learning, that is, in the extrapolation range.

[0097] In particular, by using predicted values of climate data such as average temperature as input variables and inputting the predicted values into a learned model, appropriate future manufacturing conditions can be selected. Therefore, the possibility that the system proposes out-of-specification conditions can be reduced. As a result, it becomes possible to select optimal manufacturing conditions that satisfy the target design range by making a small change in the heat treatment time.

[0098] The inventor of the present application compared the prediction accuracy of material properties and the selection of optimal manufacturing conditions between the hybrid method according to the present embodiment and the conventional method by performing cross-validation. Here, the prediction accuracy of material properties is related to interpolation in the learned model, and the selection of optimal manufacturing conditions is related to extrapolation. The prediction accuracy of material properties was found to be comparable between the two, and it was found that the hybrid method is superior to the conventional method in terms of the selection of optimal manufacturing conditions.

[0099] Furthermore, the inventor of the present application examined the prediction accuracy of material properties by comparing the target predicted value and the measured value. FIGS. 12A and 12B are graphs showing the evaluation results of the prediction accuracy of material properties by the hybrid method according to the present embodiment. FIG. 12A shows the result of comparing the target predicted value T_PTS with the measured value of TS, and FIG. 12B shows the result of comparing the target predicted value T_PYS with the measured value of YS. The vertical axis in the figure indicates the measured value, and the horizontal axis indicates the target predicted value. The coefficient of determination R of TS and YS 2 was 0.61 and 0.66, respectively. As a result, it can be seen that the target predicted values T_PTS and T_PYS can be predicted accurately.

Industrial Applicability

[0100] The technology of the present disclosure can be widely used in a manufacturing support tool used for selecting manufacturing conditions that satisfy the standard range of material properties in addition to predicting the properties of alloy materials from manufacturing conditions.

Explanation of Signs

[0101] 100, 340: Database 200: Data Processing Device 201: Main Body 210: Input Device 220: Display Device 230: Communication I / F 240: Memory Device 250: Processor 251: Input Unit 252: Pre - prediction Value Calculation Unit 253: Difference Calculation Unit 254: Prediction Model Unit 255: Target Prediction Value Calculation Unit 256: Output Unit 260: ROM 270: RAM 280: Bus 300: Cloud Server 310: Processor 320: Memory 330: Communication I / F 400: Local Area Network 500: Internet 1000: Manufacturing Support System (System)

Claims

A manufacturing support system for predicting at least one property of an alloy material manufactured through a plurality of manufacturing processes including heat treatment, comprising: a processor; a memory storing a program for controlling the operation of the processor; a storage device each storing data including a plurality of manufacturing parameters representing manufacturing conditions of various manufacturing processes and measured values of the at least one property of the alloy material manufactured under the manufacturing conditions of the manufacturing processes; comprising: The processor, according to the program, accesses the storage device, acquires the plurality of manufacturing parameters and the measured values of the at least one property, acquires a pre-prediction value calculation formula describing the relationship between a first manufacturing parameter related to heat treatment included in the plurality of manufacturing parameters and a pre-prediction value of a property representing an approximate value of a target prediction value which is the target value of the at least one property, calculates the pre-prediction value based on the first manufacturing parameter using the pre-prediction value calculation formula, calculates a difference between the calculated pre-prediction value and the measured value of the at least one property corresponding to the pre-prediction value, trains a model using a second manufacturing parameter different from the first manufacturing parameter included in the plurality of manufacturing parameters and a learning data set including the calculated difference, and generates a learned model used for predicting the at least one property; The second manufacturing parameter includes climate data, a manufacturing support system.

2. The processor inputs the second manufacturing parameter into the learned model to obtain a correction value for correcting the difference between the target prediction value and the pre-prediction value, calculates the target prediction value from the pre-prediction value and the correction value based on a mathematical formula modeling the relationship among the pre-prediction value, the correction value, and the target prediction value. The manufacturing support system according to claim 1.

3. The pre-prediction value calculation formula is given by a quadratic formula of a variable based on the first manufacturing parameter. The manufacturing support system according to claim 1 or 2.

4. The first manufacturing parameter is heat treatment time. The manufacturing support system according to any one of claims 1 to 3.

5. The second manufacturing parameter further includes a manufacturing parameter representing any one of the chemical composition of the alloy material, heat treatment temperature, coiling temperature, and natural aging time. The manufacturing support system according to claim 1.

6. The manufacturing support system according to claim 5, wherein the second manufacturing parameter further includes a manufacturing parameter representing the weight of the alloy material.

7. The first manufacturing parameter is a heat treatment time, the second manufacturing parameter includes climate data, The manufacturing support system according to claim 1, wherein the processor trains the model using the learning data set including the second manufacturing parameter.

8. The manufacturing support system according to claim 6 or 7, wherein the climate data is an average temperature.

9. The manufacturing support system according to any one of claims 1 to 8, wherein the alloy material is an aluminum alloy.

10. The manufacturing support system according to claim 9, wherein the plurality of manufacturing processes includes at least one of a raw material blending process, a melting process, a casting process, a homogenization process, a hot rolling process, a hot extrusion process, a hot forging process, a cold rolling process, a foil rolling process, a straightening process, a solution treatment process, a tempering process, and an aging process.

11. The processor further acquires at least one input design parameter defining a design range of target characteristics, inputs the input design parameter into the learned model, and outputs a group of manufacturing parameters representing one or more manufacturing conditions that satisfy the design range. The manufacturing support system according to any one of claims 1 to 10.

12. After training the model using one or more manufacturing parameters obtained from past manufacturing processes, the processor further acquires at least one input design parameter defining a design range of target characteristics, In the current manufacturing process, at least one of a group of manufacturing parameters representing one or more manufacturing conditions used from the first manufacturing process to the intermediate manufacturing processes already carried out and the input design parameter are input into the learned model, and one or more manufacturing conditions that satisfy the design range and are used in the subsequent downstream manufacturing processes after the intermediate manufacturing process are output. The manufacturing support system according to any one of claims 1 to 10.

13. After training the model using one or more manufacturing parameters obtained from past manufacturing processes, the processor further acquires at least one input design parameter defining a design range of target characteristics, In the current manufacturing process, at least one of a group of manufacturing parameters representing one or more manufacturing conditions used from the first manufacturing process to the intermediate manufacturing process that has been carried out, and a predicted value of one or more manufacturing conditions to be used in a downstream manufacturing process following the intermediate manufacturing process, a first downstream manufacturing parameter group including the predicted value, and the input design parameters are input to the learned model, and a second downstream manufacturing parameter group representing one or more manufacturing conditions different from the one or more manufacturing conditions represented by the first downstream manufacturing parameter group and satisfying the design range is output. The manufacturing support system according to any one of claims 1 to 10.

14. The manufacturing support system according to claim 12 or 13, wherein the intermediate manufacturing process is a cold rolling process.

15. The downstream process includes a tempering process, The manufacturing support system according to any one of claims 12 to 14, wherein the downstream manufacturing parameter group includes a manufacturing parameter representing a manufacturing condition in the tempering process.

16. The manufacturing support system according to any one of claims 1 to 15, wherein at least one characteristic of the alloy material includes a mechanical characteristic of the alloy material.

17. A method for generating a prediction model used to predict at least one characteristic of an alloy material manufactured through a plurality of manufacturing processes including heat treatment, Each accesses data including a plurality of manufacturing parameters representing manufacturing conditions of various manufacturing processes and measured values of the at least one characteristic of the alloy material manufactured under the manufacturing conditions of the manufacturing process, and obtains the plurality of manufacturing parameters and the measured values of the at least one characteristic. Obtain a pre-prediction value calculation formula that describes the relationship between a first manufacturing parameter related to heat treatment included in the plurality of manufacturing parameters and a pre-prediction value of a characteristic representing an approximate value of a target prediction value that is a target value of the at least one characteristic. Using the pre-prediction value calculation formula, calculate the pre-prediction value based on the first manufacturing parameter. Calculate the difference between the calculated pre-prediction value and the measured value of the at least one characteristic corresponding to the pre-prediction value. Including training a model using a second manufacturing parameter different from the first manufacturing parameter included in the plurality of manufacturing parameters and a learning data set including the difference to generate the prediction model. A method, wherein the second manufacturing parameter includes climate data.

18. A computer program for causing a computer to generate a prediction model used for predicting at least one property of an alloy material manufactured through a plurality of manufacturing processes including heat treatment, wherein the computer accesses data including a plurality of manufacturing parameters each representing manufacturing conditions of various manufacturing processes and measured values of the at least one property of the alloy material manufactured under the manufacturing conditions of the manufacturing process, to obtain the plurality of manufacturing parameters and the measured values of the at least one property; obtains a pre-prediction value calculation formula describing the relationship between a first manufacturing parameter related to heat treatment included in the plurality of manufacturing parameters and a pre-prediction value of a property representing an approximate value of a target prediction value which is the target value of the at least one property; calculates the pre-prediction value based on the first manufacturing parameter using the pre-prediction value calculation formula; calculates a difference between the calculated pre-prediction value and the measured value of the at least one property corresponding to the pre-prediction value; and causes the computer to execute training of a model using a second manufacturing parameter different from the first manufacturing parameter included in the plurality of manufacturing parameters and a learning data set including the difference, to generate the prediction model, wherein the second manufacturing parameter includes climate data.

Images