WO2020152993A1 - 金属材料の設計支援方法、予測モデルの生成方法、金属材料の製造方法、及び設計支援装置 - Google Patents
金属材料の設計支援方法、予測モデルの生成方法、金属材料の製造方法、及び設計支援装置 Download PDFInfo
- Publication number
- WO2020152993A1 WO2020152993A1 PCT/JP2019/047091 JP2019047091W WO2020152993A1 WO 2020152993 A1 WO2020152993 A1 WO 2020152993A1 JP 2019047091 W JP2019047091 W JP 2019047091W WO 2020152993 A1 WO2020152993 A1 WO 2020152993A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- design
- condition
- conditions
- metal material
- prediction model
- Prior art date
Links
- 238000013461 design Methods 0.000 title claims abstract description 233
- 238000004519 manufacturing process Methods 0.000 title claims abstract description 69
- 238000000034 method Methods 0.000 title claims abstract description 50
- 239000007769 metal material Substances 0.000 title claims abstract description 37
- 239000000203 mixture Substances 0.000 claims abstract description 32
- 239000013598 vector Substances 0.000 claims description 39
- 238000010276 construction Methods 0.000 claims description 29
- 229910052751 metal Inorganic materials 0.000 claims description 12
- 239000002184 metal Substances 0.000 claims description 12
- 229910000831 Steel Inorganic materials 0.000 description 64
- 239000000463 material Substances 0.000 description 64
- 239000010959 steel Substances 0.000 description 64
- 230000006870 function Effects 0.000 description 25
- 238000010586 diagram Methods 0.000 description 20
- 238000011156 evaluation Methods 0.000 description 17
- 238000005457 optimization Methods 0.000 description 9
- 238000013528 artificial neural network Methods 0.000 description 8
- 238000013527 convolutional neural network Methods 0.000 description 7
- 238000004364 calculation method Methods 0.000 description 6
- 238000010438 heat treatment Methods 0.000 description 6
- 238000004458 analytical method Methods 0.000 description 5
- 238000010801 machine learning Methods 0.000 description 5
- 239000000047 product Substances 0.000 description 5
- 238000003860 storage Methods 0.000 description 5
- XEEYBQQBJWHFJM-UHFFFAOYSA-N Iron Chemical compound [Fe] XEEYBQQBJWHFJM-UHFFFAOYSA-N 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 238000005097 cold rolling Methods 0.000 description 3
- 150000002894 organic compounds Chemical class 0.000 description 3
- 239000002245 particle Substances 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 239000000126 substance Substances 0.000 description 3
- 229910000805 Pig iron Inorganic materials 0.000 description 2
- 239000000654 additive Substances 0.000 description 2
- 230000000996 additive effect Effects 0.000 description 2
- 229910052782 aluminium Inorganic materials 0.000 description 2
- 229910052799 carbon Inorganic materials 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 238000001816 cooling Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000005098 hot rolling Methods 0.000 description 2
- 229910052742 iron Inorganic materials 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000007670 refining Methods 0.000 description 2
- OKTJSMMVPCPJKN-UHFFFAOYSA-N Carbon Chemical compound [C] OKTJSMMVPCPJKN-UHFFFAOYSA-N 0.000 description 1
- 235000019738 Limestone Nutrition 0.000 description 1
- 229910052770 Uranium Inorganic materials 0.000 description 1
- XAGFODPZIPBFFR-UHFFFAOYSA-N aluminium Chemical compound [Al] XAGFODPZIPBFFR-UHFFFAOYSA-N 0.000 description 1
- 238000000137 annealing Methods 0.000 description 1
- 229910052787 antimony Inorganic materials 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 229910001563 bainite Inorganic materials 0.000 description 1
- 229910052796 boron Inorganic materials 0.000 description 1
- 229910052791 calcium Inorganic materials 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000005266 casting Methods 0.000 description 1
- 229910052804 chromium Inorganic materials 0.000 description 1
- 239000000571 coke Substances 0.000 description 1
- 239000010960 cold rolled steel Substances 0.000 description 1
- 238000009749 continuous casting Methods 0.000 description 1
- 229910052802 copper Inorganic materials 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000005672 electromagnetic field Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000002068 genetic effect Effects 0.000 description 1
- 239000006028 limestone Substances 0.000 description 1
- 229910052748 manganese Inorganic materials 0.000 description 1
- 229910000734 martensite Inorganic materials 0.000 description 1
- 229910052750 molybdenum Inorganic materials 0.000 description 1
- 229910052759 nickel Inorganic materials 0.000 description 1
- 229910052758 niobium Inorganic materials 0.000 description 1
- 229910052757 nitrogen Inorganic materials 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 229910052698 phosphorus Inorganic materials 0.000 description 1
- 238000005554 pickling Methods 0.000 description 1
- 238000007747 plating Methods 0.000 description 1
- 239000002244 precipitate Substances 0.000 description 1
- 238000007637 random forest analysis Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000005096 rolling process Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000010845 search algorithm Methods 0.000 description 1
- 239000011265 semifinished product Substances 0.000 description 1
- 229910052710 silicon Inorganic materials 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000009628 steelmaking Methods 0.000 description 1
- 229910052717 sulfur Inorganic materials 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 229910052718 tin Inorganic materials 0.000 description 1
- 229910052719 titanium Inorganic materials 0.000 description 1
- 229910052720 vanadium Inorganic materials 0.000 description 1
- 238000004804 winding Methods 0.000 description 1
- 229910000859 α-Fe Inorganic materials 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B22—CASTING; POWDER METALLURGY
- B22D—CASTING OF METALS; CASTING OF OTHER SUBSTANCES BY THE SAME PROCESSES OR DEVICES
- B22D11/00—Continuous casting of metals, i.e. casting in indefinite lengths
- B22D11/16—Controlling or regulating processes or operations
-
- C—CHEMISTRY; METALLURGY
- C21—METALLURGY OF IRON
- C21D—MODIFYING THE PHYSICAL STRUCTURE OF FERROUS METALS; GENERAL DEVICES FOR HEAT TREATMENT OF FERROUS OR NON-FERROUS METALS OR ALLOYS; MAKING METAL MALLEABLE, e.g. BY DECARBURISATION OR TEMPERING
- C21D1/00—General methods or devices for heat treatment, e.g. annealing, hardening, quenching or tempering
- C21D1/26—Methods of annealing
-
- C—CHEMISTRY; METALLURGY
- C21—METALLURGY OF IRON
- C21D—MODIFYING THE PHYSICAL STRUCTURE OF FERROUS METALS; GENERAL DEVICES FOR HEAT TREATMENT OF FERROUS OR NON-FERROUS METALS OR ALLOYS; MAKING METAL MALLEABLE, e.g. BY DECARBURISATION OR TEMPERING
- C21D11/00—Process control or regulation for heat treatments
-
- C—CHEMISTRY; METALLURGY
- C21—METALLURGY OF IRON
- C21D—MODIFYING THE PHYSICAL STRUCTURE OF FERROUS METALS; GENERAL DEVICES FOR HEAT TREATMENT OF FERROUS OR NON-FERROUS METALS OR ALLOYS; MAKING METAL MALLEABLE, e.g. BY DECARBURISATION OR TEMPERING
- C21D8/00—Modifying the physical properties by deformation combined with, or followed by, heat treatment
- C21D8/02—Modifying the physical properties by deformation combined with, or followed by, heat treatment during manufacturing of plates or strips
- C21D8/0221—Modifying the physical properties by deformation combined with, or followed by, heat treatment during manufacturing of plates or strips characterised by the working steps
- C21D8/0226—Hot rolling
-
- C—CHEMISTRY; METALLURGY
- C21—METALLURGY OF IRON
- C21D—MODIFYING THE PHYSICAL STRUCTURE OF FERROUS METALS; GENERAL DEVICES FOR HEAT TREATMENT OF FERROUS OR NON-FERROUS METALS OR ALLOYS; MAKING METAL MALLEABLE, e.g. BY DECARBURISATION OR TEMPERING
- C21D8/00—Modifying the physical properties by deformation combined with, or followed by, heat treatment
- C21D8/02—Modifying the physical properties by deformation combined with, or followed by, heat treatment during manufacturing of plates or strips
- C21D8/0247—Modifying the physical properties by deformation combined with, or followed by, heat treatment during manufacturing of plates or strips characterised by the heat treatment
- C21D8/0263—Modifying the physical properties by deformation combined with, or followed by, heat treatment during manufacturing of plates or strips characterised by the heat treatment following hot rolling
-
- C—CHEMISTRY; METALLURGY
- C22—METALLURGY; FERROUS OR NON-FERROUS ALLOYS; TREATMENT OF ALLOYS OR NON-FERROUS METALS
- C22C—ALLOYS
- C22C38/00—Ferrous alloys, e.g. steel alloys
- C22C38/001—Ferrous alloys, e.g. steel alloys containing N
-
- C—CHEMISTRY; METALLURGY
- C22—METALLURGY; FERROUS OR NON-FERROUS ALLOYS; TREATMENT OF ALLOYS OR NON-FERROUS METALS
- C22C—ALLOYS
- C22C38/00—Ferrous alloys, e.g. steel alloys
- C22C38/002—Ferrous alloys, e.g. steel alloys containing In, Mg, or other elements not provided for in one single group C22C38/001 - C22C38/60
-
- C—CHEMISTRY; METALLURGY
- C22—METALLURGY; FERROUS OR NON-FERROUS ALLOYS; TREATMENT OF ALLOYS OR NON-FERROUS METALS
- C22C—ALLOYS
- C22C38/00—Ferrous alloys, e.g. steel alloys
- C22C38/02—Ferrous alloys, e.g. steel alloys containing silicon
-
- C—CHEMISTRY; METALLURGY
- C22—METALLURGY; FERROUS OR NON-FERROUS ALLOYS; TREATMENT OF ALLOYS OR NON-FERROUS METALS
- C22C—ALLOYS
- C22C38/00—Ferrous alloys, e.g. steel alloys
- C22C38/04—Ferrous alloys, e.g. steel alloys containing manganese
-
- C—CHEMISTRY; METALLURGY
- C22—METALLURGY; FERROUS OR NON-FERROUS ALLOYS; TREATMENT OF ALLOYS OR NON-FERROUS METALS
- C22C—ALLOYS
- C22C38/00—Ferrous alloys, e.g. steel alloys
- C22C38/06—Ferrous alloys, e.g. steel alloys containing aluminium
-
- C—CHEMISTRY; METALLURGY
- C22—METALLURGY; FERROUS OR NON-FERROUS ALLOYS; TREATMENT OF ALLOYS OR NON-FERROUS METALS
- C22C—ALLOYS
- C22C38/00—Ferrous alloys, e.g. steel alloys
- C22C38/12—Ferrous alloys, e.g. steel alloys containing tungsten, tantalum, molybdenum, vanadium, or niobium
-
- C—CHEMISTRY; METALLURGY
- C22—METALLURGY; FERROUS OR NON-FERROUS ALLOYS; TREATMENT OF ALLOYS OR NON-FERROUS METALS
- C22C—ALLOYS
- C22C38/00—Ferrous alloys, e.g. steel alloys
- C22C38/14—Ferrous alloys, e.g. steel alloys containing titanium or zirconium
-
- C—CHEMISTRY; METALLURGY
- C22—METALLURGY; FERROUS OR NON-FERROUS ALLOYS; TREATMENT OF ALLOYS OR NON-FERROUS METALS
- C22C—ALLOYS
- C22C38/00—Ferrous alloys, e.g. steel alloys
- C22C38/60—Ferrous alloys, e.g. steel alloys containing lead, selenium, tellurium, or antimony, or more than 0.04% by weight of sulfur
-
- C—CHEMISTRY; METALLURGY
- C22—METALLURGY; FERROUS OR NON-FERROUS ALLOYS; TREATMENT OF ALLOYS OR NON-FERROUS METALS
- C22F—CHANGING THE PHYSICAL STRUCTURE OF NON-FERROUS METALS AND NON-FERROUS ALLOYS
- C22F1/00—Changing the physical structure of non-ferrous metals or alloys by heat treatment or by hot or cold working
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/04—Manufacturing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
- G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16C—COMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
- G16C60/00—Computational materials science, i.e. ICT specially adapted for investigating the physical or chemical properties of materials or phenomena associated with their design, synthesis, processing, characterisation or utilisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/20—Ensemble learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Definitions
- the present invention relates to a design support method for a metal material having desired characteristics, a prediction model generation method, a metal material manufacturing method, and a design support apparatus.
- the user When designing a new material, the user repeatedly executes the steps of empirically defining the design conditions and the step of reviewing the design based on the trial production result of the material, by trial and error. Therefore, it takes time for the user to grasp the design result, and as a result, the time load required to design a material having sufficient properties increases. In addition, since the design content depends on the user's experience, the past experience accumulated by the user may hinder new ideas.
- Patent Document 1 proposes a method of calculating an optimal design condition by calculating a characteristic value by physical simulation and repeating updating of a forward direction model.
- Patent Document 2 proposes a method of inversely analyzing manufacturing conditions of aluminum having a desired characteristic value by using a neural network. Further, Patent Document 2 describes that the learning result of the neural network is obtained from within the condition range of the learning data set, and the neural network cannot predict the range outside the condition. That is, in Patent Document 2, it is proposed that the manufacturing conditions that deviate from the learning data set should be considered so as not to be searched, and inverse analysis should be performed.
- Non-Patent Document 1 proposes a method of constructing a prediction model for predicting a characteristic value from the chemical structure of an organic compound and specifying the chemical structure of the organic compound having a desired characteristic by inverse analysis.
- An object of the present invention made in view of the above problems is to provide a design support capable of obtaining a predicted value with high accuracy while considering the manufacturing conditions of a metal material and reducing the time load required for design.
- a method, a method of generating a prediction model, a method of manufacturing a metal material, and a design support device are provided.
- a design support method for assisting the design of a metal material having desired characteristics with a computer, It is constructed based on past performance data in which design conditions including the component composition and manufacturing conditions of the metal material and characteristic values of the metal material are associated with each other, and a prediction model for predicting the characteristic value from the design conditions is used, A search step of searching for the design condition for obtaining a desired characteristic, Of the design conditions corresponding to the desired characteristics searched by the searching step, a presentation step of presenting at least a component composition and manufacturing conditions, Including In the searching step, the design condition is searched so as to reduce variations in a plurality of prediction values based on a plurality of different learning data sets.
- a prediction model generation method is A method of generating the prediction model used in the design support method, comprising: An acquisition step of acquiring the past performance data in which the design condition and the characteristic value are associated with each other; A construction step of constructing the prediction model for predicting the characteristic value from the design condition based on the past performance data acquired by the acquisition step; including.
- a method for manufacturing a metal material includes manufacturing the metal material based on the design condition searched by the design support method.
- a design support device for supporting the design of a metal material having desired characteristics, It is constructed based on past performance data in which design conditions including the component composition and manufacturing conditions of the metal material and characteristic values of the metal material are associated with each other, and a prediction model for predicting the characteristic value from the design conditions is used,
- a search unit that searches for the design condition that provides the desired characteristics, Of the design conditions corresponding to the desired characteristics searched by the search unit, a presentation unit that presents at least a component composition and manufacturing conditions, Equipped with The search unit searches for the design condition such that variations in a plurality of prediction values based on a plurality of different learning data sets are reduced.
- the time load required for design is reduced while considering the metal material manufacturing conditions. It is possible.
- 6 is a flowchart showing a first example of the operation of the design support device of FIG. 1.
- 9 is a flowchart showing a second example of the operation of the design support device of FIG. 1.
- the metal material designed in the first embodiment is, for example, steel, but the metal material is not limited to steel and may be any metal.
- FIG. 1 is a functional block diagram showing the configuration of the design support device 1 according to the first embodiment of the present invention.
- the design support device 1 according to the first embodiment is a computer including an acquisition unit 11, a prediction model construction unit 12, a search unit 13, and a presentation unit 14.
- the design support device 1 supports the design of a metal material having desired characteristics.
- the acquisition unit 11 acquires, for example, past performance data relating to the manufacturing of steel materials, which is necessary for constructing a prediction model described later.
- the acquisition unit 11 may include a communication interface for acquiring actual result data.
- the acquisition unit 11 may receive the performance data from a plurality of external devices or the like by using a predetermined communication protocol.
- the actual data includes, for example, data in which design conditions including the component composition and manufacturing conditions of the steel material are associated with the characteristic values of the steel material.
- the manufacturing conditions include, for example, set values and actual values in the manufacturing conditions.
- the data on the component composition of the steel material acquired by the acquisition unit 11 includes, for example, the addition ratio of the element to be dissolved as a component in the steel in the converter or the secondary refining.
- Such elements include, for example, C, Si, Mn, P, S, Al, N, Cr, V, Sb, Mo, Cu, Ni, Ti, Nb, B, Ca, and Sn.
- FIG. 2 is a schematic view showing the manufacturing process of the cold rolled coil of the steel material according to the first embodiment.
- a steel manufacturing process first, raw iron ore is charged into a blast furnace together with limestone and coke to produce molten pig iron.
- the pig iron tapped in the blast furnace has its components such as carbon adjusted in the converter of the steelmaking factory, and the final components adjusted by secondary refining.
- the obtained molten steel is cast by a continuous casting machine or the like to manufacture a semi-finished product called a slab before plate forming.
- a cold rolled coil which is a product is manufactured through a plurality of processing steps such as a heating step in a heating furnace, a hot rolling step, a cooling step, a pickling step, a cold rolling step, an annealing step, and a plating step. ..
- the combination of these plurality of processing steps differs depending on the product to be manufactured.
- the characteristics of steel materials are generally determined by the manufacturing conditions of lower steps such as a hot rolling step, a cooling step, and a cold rolling step, which are steps subsequent to the heating of the slab after casting, except for the conditions relating to the components. Therefore, in the first embodiment, as the manufacturing condition, the condition in the steps after the slab generation is treated as an example.
- the manufacturing conditions include, for example, the following.
- the data of the characteristic values of the steel material acquired by the acquisition unit 11 are, for example, yield point (N/mm 2 ), tensile strength (N/mm 2 ), elongation (%), r value, n value, hole expansion ratio (% ), BH content (N/mm 2 ), and yield ratio. These characteristic values can be obtained, for example, by performing a sampling test for evaluating the characteristics of the steel material from some of the manufactured steel material products.
- the acquisition unit 11 uses the acquired actual values of the characteristic values as actual data, and manages the actual data in association with each other. More specifically, the acquisition unit 11 centrally collects the actual data of the component composition of the steel material, the actual data of the manufacturing conditions, and the actual data of the characteristic value of the steel material for each unit of the manufactured steel material product. They are associated with each other so that they can be handled collectively.
- the acquisition unit 11 may acquire, as input information, constraint conditions including the range of the component composition of the steel material and the range of the manufacturing conditions in the search for the optimum design conditions described below.
- the constraint condition may further include a range of characteristic values of the steel material.
- the constraint condition may further include a condition that guarantees a consistency with the change in the physical quantity of the steel sheet that occurs between different steps in the manufacturing process. For example, the temperature of the steel sheet decreases as the process progresses except when a heating process is performed.
- the constraint condition may include a condition that constrains the manufacturing condition so as not to conflict with such a temperature drop phenomenon.
- the acquisition unit 11 may include, for example, an input interface for a user of a steel material to input a predetermined constraint condition as input information.
- FIG. 3 is a diagram showing an outline of the prediction model according to the first embodiment.
- the prediction model construction unit 12 constructs a prediction model as shown in FIG. 3 that predicts the characteristic value of the steel material from the design conditions based on the past performance data acquired by the acquisition unit 11. More specifically, the prediction model construction unit 12 constructs a prediction model of the characteristic value of the steel material using the component composition of the steel material and the manufacturing conditions as explanatory variables, based on the acquired past performance data.
- Predictive models include models using machine learning techniques including neural networks, local regression models, kernel regression models, and random forests.
- the prediction model construction unit 12 may select a prediction model that can handle a plurality of target variables such as a neural network, or may construct a prediction model for each characteristic value. Good.
- the prediction model constructed by the prediction model construction unit 12 is used as a learned model for searching for optimum design conditions described later.
- the search unit 13 searches for the optimum design condition for obtaining the desired characteristic by using the predictive model constructed by the predictive model constructing unit 12 and predicting the characteristic value of the steel material from the design condition as shown in FIG. .
- the desired characteristic may be, for example, a characteristic that maximizes the characteristic value when the characteristic value is desired to be maximized, or a minimum characteristic value when the characteristic value is desired to be minimized.
- the characteristics may be as follows.
- the desired characteristic may be, for example, any characteristic arbitrarily determined by the user in response to a product request by the user.
- the presentation unit 14 presents the design condition corresponding to the desired characteristic searched by the search unit 13 to the user.
- the user can efficiently design the steel material by using the component composition and the manufacturing condition of the steel material presented by the presentation unit 14 as a target value or a reference value at the time of manufacturing the steel material. If the design conditions include additional conditions other than the component composition and the manufacturing conditions, the presentation unit 14 presents at least the component composition and the manufacturing conditions, and appropriately presents some or all of the additional conditions. ..
- FIG. 4 is a flowchart showing a first example of the operation by the design support device 1 of FIG. FIG. 4 shows a flow in which the design support apparatus 1 generates a prediction model as shown in FIG. 3 based on past performance data.
- step S101 the design support apparatus 1 acquires the past performance data in which the design conditions including the component composition and manufacturing conditions contained in the steel material and the characteristic values of the steel material are associated with each other by the acquisition unit 11.
- step S102 the design support device 1 builds a prediction model for predicting the characteristic value of the steel material from the design condition by the prediction model building unit 12 based on the past performance data acquired in step S101.
- FIG. 5 is a flowchart showing a second example of the operation of the design support device 1 of FIG.
- FIG. 5 shows a flow in which the design support apparatus 1 searches for optimum design conditions using the prediction model constructed by the flow of FIG. 4 and presents it to the user.
- step S201 the search unit 13 of the design support apparatus 1 acquires from the acquisition unit 11 constraint conditions including, for example, a range of component composition of a steel material, a range of manufacturing conditions, and a range of characteristic values of a steel material as input information. ..
- step S202 the search unit 13 of the design support device 1 acquires the above-described prediction model constructed by the prediction model construction unit 12 from the prediction model construction unit 12.
- step S203 the search unit 13 of the design support apparatus 1 performs the optimum design for obtaining desired characteristics for the steel material based on the constraint condition acquired in step S201 and the prediction model acquired in step S202. Search for conditions.
- step S204 the presentation unit 14 of the design support apparatus 1 acquires the optimum design condition corresponding to the desired characteristic searched in step S203 and the corresponding characteristic value from the search unit 13 and presents it to the user.
- the user manufactures a steel material based on the design conditions searched for in step S203 and presented in step S204.
- the search unit 13 acquires, from the acquisition unit 11, the constraint conditions as shown in Table 2 below as input information. More specifically, the search unit 13 acquires the upper limit value and the lower limit value of the component composition of the steel material, which are the design conditions, and the upper limit value and the lower limit value of the manufacturing conditions, as constraint conditions.
- the search unit 13 acquires the upper limit value and the lower limit value of the characteristic value of the steel material as a constraint condition.
- the search unit 13 uses the learned prediction model acquired from the prediction model construction unit 12 in step S202 of FIG. 5 to search for the optimum design condition within the constraint condition in step S203 of FIG.
- a problem is an optimization problem, and the problem can be described as follows.
- x is a design condition expressed as a vector
- k is a type of characteristic
- f k (x) is a predicted value of the characteristic
- ⁇ k is a preset weighting coefficient.
- the function f k (x) of the predicted value of the characteristic in the evaluation function is based on the prediction model constructed by the prediction model construction unit 12.
- F is a set of design conditions x satisfying the constraint conditions acquired in step S201 of FIG. Therefore, the search unit 13 searches for the optimum design condition within the range satisfying the constraint condition.
- L k and U k are the lower limit value and the upper limit value of the characteristic value acquired in step S201 of FIG. 5, respectively.
- the search unit 13 solves such an optimization problem by a method using metaheuristics, a genetic algorithm, mathematical programming, swarm intelligence, and the like.
- the search unit 13 searches for optimum design conditions as a problem that maximizes the evaluation function in Expression 1, but the method of setting the problem is not limited to this.
- the search unit 13 may search for the optimum design condition as a problem of minimizing the absolute value of the evaluation function in Expression 1 by setting the sign of the weighting coefficient ⁇ k to be negative.
- properties that are desired to be maximized include tensile strength and elongation.
- properties that are desired to be minimized include yield ratio.
- the search unit 13 calculates the design condition x obtained by Expression 1 as an optimum solution, but is not limited to this.
- the search unit 13 may set a predetermined condition for the calculation time and calculate the design condition x as the best solution obtained within the corresponding time.
- the search unit 13 may store all the solutions obtained within the corresponding time and output them all at the end.
- the design support device 1 uses the learned prediction model constructed by, for example, the prediction model construction unit 12 based on the past performance data in which the design condition and the characteristic value of the steel material are associated with each other. Performs calculation of material property values.
- the design support device 1 can quickly calculate the characteristic values of the steel material based on a large number of design conditions, and can sufficiently execute the search even within a predetermined time.
- the design support device 1 can search for design conditions corresponding to excellent characteristic values of steel materials.
- the design support apparatus 1 determines that the optimum design condition corresponding to the desired characteristic satisfies the constraint condition as shown in the equation 1, so that the amount of the additive to the steel material and the capacity of the manufacturing facility are temporarily determined from the viewpoint of the manufacturing cost. Even if there is a limit, the design conditions obtained by the inverse analysis can be effectively used. By designing the constraint condition, the design support apparatus 1 can efficiently search within the range of the constraint condition instead of searching for the dark clouds.
- the design support apparatus 1 integrally has the prediction model construction unit 12 that constructs a prediction model, but the configuration of the design support apparatus 1 is not limited to this.
- the design support device 1 may not include the prediction model construction unit 12. That is, the design support device 1 may not have the function related to the construction of the prediction model using the acquisition unit 11 and the prediction model construction unit 12.
- an external device having a configuration unit corresponding to the acquisition unit 11 and the prediction model construction unit 12 acquires the above-described result data to generate a prediction model, and designs the prediction model by the arbitrary method from the external device.
- a mode in which the support device 1 obtains is also possible.
- the search unit 13 has been described as searching for the optimum design condition within the range satisfying the constraint condition, but the present invention is not limited to this.
- the search unit 13 does not acquire the constraint condition from the acquisition unit 11 and does not consider the constraint condition. You may solve the optimization problem of.
- Expression 1 the evaluation function is expressed as the weighted sum of the maximum value or the minimum value of each characteristic, but the content of Expression 1 is not limited to this. Expression 1 may be replaced by Expression 2 below based on the target value ref k of each characteristic.
- the search unit 13 can also search for a design condition that is evaluated higher as it approaches the target value ref k of each characteristic.
- the second embodiment of the present invention will be described below.
- the configuration and basic functions of the design support device 1 according to the second embodiment are the same as the above-mentioned contents regarding the first embodiment described with reference to FIGS. 1 to 5. Therefore, the corresponding contents described in the first embodiment similarly apply to the second embodiment.
- the same components as those in the first embodiment are designated by the same reference numerals and the description thereof will be omitted.
- differences from the first embodiment will be mainly described.
- the variation of the predicted value of the characteristic due to the learning data is considered in the evaluation function of Expression 1.
- the “variation of predicted values” includes, for example, a plurality of predicted value variations based on different learning data sets.
- the design support device 1 solves the optimization problem by changing Equation 1 to the following problem setting.
- V k (x) indicates “variation of predicted values” in which instability in which predicted values change due to different learning data sets of the prediction model is calculated.
- ⁇ k is a preset weighting coefficient.
- FIG. 6 is a schematic diagram for explaining an example of a method for calculating V k (x).
- the search unit 13 calculates V k (x) by an arbitrary method including, for example, the method shown in FIG.
- a plurality of datasets for prediction evaluation are created by randomly extracting a plurality of samples from the dataset for model learning.
- the number of samples to be extracted is a number corresponding to about 70 to 90% of the original model learning data set.
- a plurality of prediction models for prediction evaluation using each data set as learning data are created.
- the prediction model for prediction evaluation is constructed by the prediction model construction unit 12.
- These prediction models for prediction evaluation are created in advance before step S203 of FIG. 5 in which the optimum design condition is searched.
- the prediction value is calculated using the prediction model for each prediction evaluation, and the variance of the plurality of prediction values is V Let k (x).
- K ⁇ N predictive values are obtained by taking the characteristic type k from 1 to K and the predictive model number i from 1 to N.
- the prediction value y ik can be represented by a matrix with K rows and N columns.
- V k (x) is represented by the following Expression 4.
- the second term in parentheses on the right side of Expression 4 is the average value of the predicted values y 1k to y Nk .
- the calculation may be performed by directly using the value of the prediction value y ik , or, for example, a value after normalization is performed for each row in the matrix of K rows and N columns of the prediction value y ik. The calculation may be performed using.
- step S203 of FIG. 5 the variation in the predicted value due to the learning data is minimized.
- V k (x) the objective function of the problem setting in order to consider the certainty of the predicted value
- the search is executed so that the variation in the predicted value becomes small. Therefore, as a result of actually verifying the design conditions obtained by the search by the search unit 13, the possibility that the actual value of the characteristic greatly differs from the predicted value of the characteristic is reduced. That is, it is possible to accurately obtain the predicted value.
- Expression 5 may be adopted instead of Expression 3 based on the target value ref k of each characteristic.
- the search unit 13 can search for a design condition in which the closer to the target value ref k of each characteristic, the higher the evaluation and the reliability of the predicted value are guaranteed.
- the design condition may be searched so that the variation in the predicted value due to the learning data is reduced.
- a predetermined first threshold value may be provided, and the design condition may be searched so that the variation in the predicted value due to the learning data becomes smaller than the predetermined first threshold value.
- the predetermined first threshold includes, for example, a value appropriately determined by the design support device 1 or the user.
- the third embodiment of the present invention will be described below.
- the configuration and the basic function of the design support device 1 according to the third embodiment are the same as the above-described contents regarding the first embodiment and the above-described contents regarding the second embodiment described with reference to FIGS. 1 to 6. is there. Therefore, the corresponding contents described in the first embodiment and the second embodiment apply similarly in the third embodiment.
- the same components as those in the first and second embodiments are designated by the same reference numerals, and the description thereof will be omitted.
- differences from the first embodiment and the second embodiment will be mainly described.
- the difference between the design condition to be searched and the design condition in the past performance data is considered in the evaluation function.
- the search unit 13 tends to search for a range close to the conventionally designed design condition, and may not actively search for a new design condition.
- the following function D(x) is incorporated as the objective function of the problem setting so that the search unit 13 positively and reliably searches for new design conditions.
- ⁇ is a preset weighting coefficient.
- the evaluation function is composed of the weighted sum of the three functions.
- D(x) is the size of the deviation of the actual data used when the prediction model was constructed from the design conditions, and is represented by the following equation 7, for example.
- h si represents the i-th design condition of the s-th actual data.
- ⁇ i is a coefficient related to the i-th design condition.
- Formula 7 indicates that the distance between the design condition to be searched and the design condition of the actual data used when creating the prediction model is integrated for each actual data.
- the search unit 13 does not search for design conditions similar to the actual data, but more actively searches for a new area of design conditions that has not been implemented so far.
- the difference between the design condition to be searched and the design condition in the past actual data is maximized within the range in which the variation in the predicted value due to the learning data is minimized.
- the search unit 13 may search for the design condition including a new area different from the past performance data. As described above, the search unit 13 can not only search for the predicted value of the excellent characteristic, but also more positively search in the direction away from the design conditions so far while considering the accuracy of the prediction. ..
- the search unit 13 can search a range farther from the actual data than the other design condition under the corresponding design condition. For example, when a user wants to find a new design condition by largely changing the composition of the steel material as the design condition from the conventional one, the coefficient ⁇ i related to the production condition is reduced to a coefficient related to the composition of the steel material. ⁇ i may be increased. For example, when the user wants to find a new design condition by largely changing the manufacturing condition as compared with the component composition of the steel material as a design condition, the coefficient ⁇ i related to the component composition of the steel material is reduced and the coefficient related to the manufacturing condition is reduced. ⁇ i may be increased.
- the search unit 13 more actively searches for new design conditions.
- the user can also discover a design condition for obtaining a completely new characteristic. If the user can discover different design conditions for the same characteristic value, the degree of freedom in design can be increased.
- Expression 8 may be adopted instead of Expression 6 based on the target value ref k of each characteristic.
- the search unit 13 is more active as the target value ref k of each characteristic is evaluated higher, and is more aggressive in areas of design conditions that have not been implemented so far in consideration of the certainty of the predicted value. Can be searched for.
- step S203 of FIG. 5 when the optimum design condition is searched within the range of the constraint condition, the difference between the design condition to be searched and the design condition in the past performance data is maximized.
- the search unit 13 can not only search for a predicted value with excellent characteristics, but also more positively in a direction away from the design conditions up to now. In this case, since the term of V k (x) does not exist in Expression 6, the degree of freedom of conditional search is increased.
- Equation 8 the objective function corresponding to Equation 8 is shown as follows.
- the search unit 13 is more active in the area of the design condition that has not been implemented until now, in the state where the evaluation is higher as the target value ref k of each characteristic is closer and the accuracy of the predicted value is taken into consideration. Can be searched for.
- the design condition may be searched such that the difference between the design condition to be searched and the design condition in the past performance data increases.
- a predetermined second threshold value is provided, and the design condition is searched such that the difference between the design condition to be searched and the design condition in the past performance data is larger than the predetermined second threshold value.
- the predetermined second threshold includes, for example, a value appropriately set by the design support device 1 or the user.
- the fourth embodiment of the present invention will be described below.
- the configuration and the basic function of the design support device 1 according to the fourth embodiment are the same as those described above with reference to FIGS. 1 to 6, and the second embodiment and the third embodiment described above. It is the same as the contents of. Therefore, the corresponding contents described in the first embodiment to the third embodiment are similarly applied to the fourth embodiment.
- the same components as those in the first to third embodiments are designated by the same reference numerals, and the description thereof will be omitted.
- differences from the first embodiment to the third embodiment will be mainly described.
- the fourth embodiment uses the image data of the metallographic structure of the steel material as an input to the prediction model shown in FIG.
- the image data of the metal structure requires a size and a resolution with which the design support apparatus 1 can evaluate the structure grain size and the structure fraction of the metal structure including, for example, ferrite, martensite, bainite and the like.
- the design support apparatus 1 may separately use image data having different sizes and resolutions and process a plurality of image data.
- FIG. 7 is a diagram showing an outline of the prediction model according to the fourth embodiment.
- the prediction model is, for example, a model using a neural network.
- the acquisition unit 11 acquires the component composition and manufacturing conditions of the steel material as the input of the prediction model at the time of learning, as shown in FIG. 3, but is not limited to this. As shown in FIG. 7, in the fourth embodiment, in addition to these design conditions, the acquisition unit 11 further acquires image data of the metallographic structure of the steel material for use in inputting a prediction model during learning. May be.
- the prediction model construction unit 12 constructs a prediction model also based on the image data acquired by the acquisition unit 11.
- the search unit 13 searches for design conditions using the prediction model.
- the prediction model construction unit 12 quantifies the image data by a predetermined method. If the prediction model construction unit 12 takes out pixel information of an image as a vector and inputs it as an input, even with image data of a metallographic structure having the same characteristics, if the pixel information is slightly different, input values of different vectors can be obtained. Therefore, parameter learning of the prediction model is additionally generated for each learning image data, and learning efficiency deteriorates.
- the prediction model construction unit 12 uses a convolutional neural network to convert the image data of the metal structure into a feature amount vector having a smaller number of pixels.
- the prediction model construction unit 12 uses the converted feature amount vector as an input value of the prediction model.
- the image data is converted into a low-dimensional feature amount vector. Therefore, the image data of the metal structures of the steel materials having similar characteristic values are the same or similar vectors, and the learning efficiency is improved.
- FIG. 8A is a schematic diagram showing a first example of the correspondence relationship between image data and feature amount vectors.
- FIG. 8B is a schematic diagram showing a second example of the correspondence relationship between image data and feature amount vectors.
- FIG. 8C is a schematic diagram showing a third example of the correspondence relationship between image data and feature amount vectors.
- FIG. 8D is a schematic diagram showing a fourth example of the correspondence relationship between image data and feature amount vectors.
- the prediction model construction unit 12 uses the convolutional neural network to convert the first image data of the metallographic structure into the first feature amount vector ( 0.11, 0.47, 0.94, 0.83).
- the prediction model construction unit 12 uses the convolutional neural network to convert the second image data of the metallographic structure into the second feature amount vector ( 0.10, 0.31, 0.54, 0.89).
- the prediction model construction unit 12 uses the convolutional neural network to convert the third image data of the metallographic structure into the third feature amount vector ( 0.56, 0.91, 0.35, 0.92).
- the prediction model construction unit 12 uses the convolutional neural network to convert the fourth image data of the metallographic structure into the fourth feature amount vector ( 0.41, 0.91, 0.38, 0.20).
- FIG. 8A to 8D show image data of four metal structures, the number of image data of metal structures used as an input to the prediction model may be any number. Although FIG. 8A to FIG. 8D show four elements for each feature amount vector, the number of elements of the feature amount vector may be any number.
- the prediction model construction unit 12 uses the image data of the metal structure in addition to the component composition and the manufacturing conditions of the steel material as the input at the time of learning, and uses the characteristic value as the output. Then, the learning is performed at the same time including the convolutional neural network part.
- the search unit 13 uses a model excluding the convolutional neural network part at the time of searching for design conditions, as indicated by the broken line box in FIG. 7. At this time, the design condition searched by the search unit 13 includes the above feature amount vector instead of the image data.
- FIG. 9 is a schematic diagram showing an example of a method of converting a feature amount vector into image data.
- the feature quantity vector obtained by the search by the search unit 13 may be converted into image data by the following method, for example.
- the search unit 13 calculates the feature amount vector when the image data for learning is input to the learned model as in FIGS. 8A to 8D.
- the search unit 13 stores the data in which the image data and the feature amount vector are associated with each other in an arbitrary storage device.
- the search unit 13 refers to the storage device and selects one of the stored feature amount vectors that approximates to the feature amount vector included in the design condition obtained by the search.
- the search unit 13 obtains (0.40, 0.90, 0, 40, 0.20) as the feature quantity vector included in the design condition obtained by the search.
- the search unit 13 refers to the storage device and approximates the feature amount vector (0.40, 0.90, 0, 40, 0.20) from the stored feature amount vectors to the fourth feature. Select the quantity vector (0.41, 0.91, 0.38, 0.20).
- the search unit 13 outputs the fourth image data of the metal structure corresponding to the selected fourth feature amount vector to the presentation unit 14 as a design condition while referring to the storage device.
- the design conditions searched for in step S203 of FIG. 5 include a feature amount vector based on the image data of the metal structure of the steel material.
- the design support device 1 can acquire the design condition unique to the image data of the metallographic structure, which is different from the component composition and manufacturing conditions of the steel material. Therefore, the prediction accuracy of the prediction model is improved.
- the user does not need to input to the design support apparatus 1 what kind of parameter regarding the metal structure can be obtained from such image data as teacher data. Since the design support apparatus 1 can express the difference in the image data of the metallographic structure as the difference in the feature amount vector, it does not specifically specify what parameter the metallographic structure causes. Also, such a difference can be reflected in the characteristic value of the output.
- Table 4 shows an example of the component composition (unit: mass%) of the steel material that affects the characteristics
- Table 5 shows an example of the manufacturing conditions that affect the characteristics
- Table 6 shows an example of the type of characteristics and characteristic values.
- FIG. 10 shows a scatter diagram related to actual and predicted tensile strength values.
- the horizontal axis of the scatter diagram is the actual value of tensile strength, and the vertical axis is the predicted value of tensile strength.
- the hidden layer of the neural network was one layer, and the number of nodes was 15. The numerical value of each explanatory variable is standardized.
- the prediction accuracy was RMSE (Root Mean Square Error) of 71.94.
- Table 7 shows the constraints of the design conditions used when searching for the design conditions.
- the constraint conditions include the range of component composition of steel materials and the range of manufacturing conditions.
- the constraint conditions may include conditions that guarantee that they are consistent with changes in the physical quantity of the steel sheet that occur between different steps in the manufacturing process.
- the constraint condition may include a condition of heating temperature>finishing rolling temperature>winding temperature.
- Table 8 shows constraint conditions of characteristic values used when searching for design conditions.
- the design support device 1 optimizes the optimization problem as shown in Expression 6 and Expression 7 of the third embodiment.
- the design support device 1 uses swarm intelligence particle swarm optimization as a search algorithm.
- the number of particles was 1000, and the solution was updated 500 times.
- the number of models used for calculating V(x) was 50, and 80% of the learning data was randomly selected and used for learning.
- ⁇ i in D(x) of Equation 7 ⁇ i related to the design conditions of the component composition of the steel material was set to 1 and ⁇ i related to the manufacturing conditions was set to 0.
- a standardized value was used for the calculation of D(x).
- Table 9 shows the design conditions searched by the design support device 1.
- the tensile strength under the design conditions was 1545 MPa, and the maximum tensile strength in the actual data was 1498 MPa, which means that a new design condition included in a new search area with higher tensile strength was discovered.
- Table 10 is a table comparing the result when the optimization problem is solved in the first and second embodiments, and the above result in the third embodiment.
- ⁇ 1.
- the constraints and parameters of the particle swarm optimization are the same as in the case of the third embodiment.
- the first embodiment when trying to search for a design condition corresponding to high tensile strength, there is a tendency for the variation in the predicted value of the characteristics to increase, and there is a large risk that the actual value of the tensile strength will differ from the predicted value.
- design conditions with less variation in the predicted value are searched for as compared with the first embodiment in which the variation in the predicted value is not considered.
- the third embodiment it can be seen that the variation in the predicted value is suppressed to be small, and the design condition in the search region farther from the actual data can be searched for as compared with the second embodiment.
- the present invention can also be realized as a program describing the processing content for realizing each function of the above-described design support apparatus 1 or a storage medium recording the program. It should be understood that these are also included in the scope of the present invention.
Landscapes
- Engineering & Computer Science (AREA)
- Chemical & Material Sciences (AREA)
- Mechanical Engineering (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Materials Engineering (AREA)
- Organic Chemistry (AREA)
- Metallurgy (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Economics (AREA)
- Tourism & Hospitality (AREA)
- Thermal Sciences (AREA)
- Crystallography & Structural Chemistry (AREA)
- General Business, Economics & Management (AREA)
- Marketing (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Entrepreneurship & Innovation (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Computing Systems (AREA)
- Quality & Reliability (AREA)
- Health & Medical Sciences (AREA)
- Development Economics (AREA)
- Medical Informatics (AREA)
- Operations Research (AREA)
- Game Theory and Decision Science (AREA)
- Primary Health Care (AREA)
- General Health & Medical Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computer Hardware Design (AREA)
- Manufacturing & Machinery (AREA)
- Educational Administration (AREA)
Abstract
Description
所望の特性を有する金属材料の設計を計算機により支援する設計支援方法であって、
前記金属材料の成分組成及び製造条件を含む設計条件と前記金属材料の特性値とを関連付けた過去の実績データに基づき構築され、前記設計条件から前記特性値を予測する予測モデルを用いて、前記所望の特性が得られる前記設計条件を探索する探索ステップと、
前記探索ステップにより探索された、前記所望の特性に対応する前記設計条件のうち、少なくとも成分組成及び製造条件を提示する提示ステップと、
を含み、
前記探索ステップにおいて、異なる複数の学習データセットに基づく複数の予測値のばらつきが低減するように前記設計条件を探索する。
上記の設計支援方法に用いられる前記予測モデルの生成方法であって、
前記設計条件と前記特性値とを関連付けた前記過去の実績データを取得する取得ステップと、
該取得ステップにより取得された前記過去の実績データに基づいて、前記設計条件から前記特性値を予測する前記予測モデルを構築する構築ステップと、
を含む。
上記の設計支援方法により探索された前記設計条件に基づいて前記金属材料を製造するステップを含む。
所望の特性を有する金属材料の設計を支援する設計支援装置であって、
前記金属材料の成分組成及び製造条件を含む設計条件と前記金属材料の特性値とを関連付けた過去の実績データに基づき構築され、前記設計条件から前記特性値を予測する予測モデルを用いて、前記所望の特性が得られる前記設計条件を探索する探索部と、
前記探索部により探索された、前記所望の特性に対応する前記設計条件のうち、少なくとも成分組成及び製造条件を提示する提示部と、
を備え、
前記探索部は、異なる複数の学習データセットに基づく複数の予測値のばらつきが低減するように前記設計条件を探索する。
以下、本発明の第1実施形態について説明する。第1実施形態において設計する金属材料は例えば鉄鋼であるが、金属材料は鉄鋼に限定されず、任意の金属であってもよい。
以下、本発明の第2実施形態について説明する。第2実施形態に係る設計支援装置1の構成及び基本的な機能は、図1乃至図5を用いて説明した第1実施形態に関する上記の内容と同一である。したがって、第1実施形態において説明した対応する内容は、第2実施形態においても同様に当てはまる。第1実施形態と同一の構成については同一の符号を付し、説明は省略する。以下では、第1実施形態と異なる点について主に説明する。第2実施形態では第1実施形態と異なり、式1の評価関数において学習データ起因の特性の予測値のばらつきが考慮される。本明細書において、「予測値のばらつき」は、例えば、異なる複数の学習データセットに基づく複数の予測値のばらつきを含む。
以下、本発明の第3実施形態について説明する。第3実施形態に係る設計支援装置1の構成及び基本的な機能は、図1乃至図6を用いて説明した第1実施形態に関する上記の内容、及び第2実施形態に関する上記の内容と同一である。したがって、第1実施形態及び第2実施形態において説明した対応する内容は、第3実施形態においても同様に当てはまる。第1実施形態及び第2実施形態と同一の構成については同一の符号を付し、説明は省略する。以下では、第1実施形態及び第2実施形態と異なる点について主に説明する。第3実施形態では第1実施形態及び第2実施形態と異なり、探索の対象となる設計条件と過去の実績データにおける設計条件との差が評価関数において考慮される。
以下、本発明の第4実施形態について説明する。第4実施形態に係る設計支援装置1の構成及び基本的な機能は、図1乃至図6を用いて説明した第1実施形態に関する上記の内容、並びに第2実施形態及び第3実施形態に関する上記の内容と同一である。したがって、第1実施形態乃至第3実施形態において説明した対応する内容は、第4実施形態においても同様に当てはまる。第1実施形態乃至第3実施形態と同一の構成については同一の符号を付し、説明は省略する。以下では、第1実施形態乃至第3実施形態と異なる点について主に説明する。第4実施形態では第1実施形態乃至第3実施形態と異なり、図3に示す予測モデルに対する入力として鉄鋼材料の金属組織の画像データが利用される。
以下、自動車用冷延鋼板についての鉄鋼材料の設計の例を、主に上述の第3実施形態に基づいて示す。本実施例では鉄鋼材料の特性として引張強度が選択され、所望の特性値として引張強度の最大値を含む設計条件が探索された。
11 取得部
12 予測モデル構築部
13 探索部
14 提示部
Claims (7)
- 所望の特性を有する金属材料の設計を計算機により支援する設計支援方法であって、
前記金属材料の成分組成及び製造条件を含む設計条件と前記金属材料の特性値とを関連付けた過去の実績データに基づき構築され、前記設計条件から前記特性値を予測する予測モデルを用いて、前記所望の特性が得られる前記設計条件を探索する探索ステップと、
前記探索ステップにより探索された、前記所望の特性に対応する前記設計条件のうち、少なくとも成分組成及び製造条件を提示する提示ステップと、
を含み、
前記探索ステップにおいて、異なる複数の学習データセットに基づく複数の予測値のばらつきが低減するように前記設計条件を探索する、
設計支援方法。 - 前記成分組成の範囲、及び前記製造条件の範囲を含む制約条件を入力情報として取得する取得ステップを含み、
前記探索ステップにおいて、前記所望の特性に対応する前記設計条件は前記制約条件を満たす、
請求項1に記載の設計支援方法。 - 前記探索ステップにおいて、探索の対象となる前記設計条件と前記過去の実績データにおける前記設計条件との差が増加するように、前記過去の実績データと異なる新たな領域を含めて前記設計条件を探索する、
請求項1又は2に記載の設計支援方法。 - 前記設計条件は、前記金属材料の金属組織の画像データに基づく特徴量ベクトルを含み、
前記探索ステップにおいて、探索された前記設計条件は、前記特徴量ベクトルを含む、
請求項1乃至3のいずれか1項に記載の設計支援方法。 - 請求項1乃至4のいずれか1項に記載の設計支援方法に用いられる前記予測モデルの生成方法であって、
前記設計条件と前記特性値とを関連付けた前記過去の実績データを取得する取得ステップと、
該取得ステップにより取得された前記過去の実績データに基づいて、前記設計条件から前記特性値を予測する前記予測モデルを構築する構築ステップと、
を含む、
予測モデルの生成方法。 - 請求項1乃至4のいずれか1項に記載の設計支援方法により探索された前記設計条件に基づいて前記金属材料を製造するステップを含む、
金属材料の製造方法。 - 所望の特性を有する金属材料の設計を支援する設計支援装置であって、
前記金属材料の成分組成及び製造条件を含む設計条件と前記金属材料の特性値とを関連付けた過去の実績データに基づき構築され、前記設計条件から前記特性値を予測する予測モデルを用いて、前記所望の特性が得られる前記設計条件を探索する探索部と、
前記探索部により探索された、前記所望の特性に対応する前記設計条件のうち、少なくとも成分組成及び製造条件を提示する提示部と、
を備え、
前記探索部は、異なる複数の学習データセットに基づく複数の予測値のばらつきが低減するように前記設計条件を探索する、
設計支援装置。
Priority Applications (7)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
BR112021014096-4A BR112021014096A2 (pt) | 2019-01-21 | 2019-12-02 | Método de suporte de desenho para material metálico, método de geração de modelo de previsão, método de produção de material metálico, e dispositivo de suporte de desenho |
KR1020217025687A KR20210114993A (ko) | 2019-01-21 | 2019-12-02 | 금속 재료의 설계 지원 방법, 예측 모델의 생성 방법, 금속 재료의 제조 방법 및, 설계 지원 장치 |
JP2020516503A JP7028316B2 (ja) | 2019-01-21 | 2019-12-02 | 金属材料の設計支援方法、予測モデルの生成方法、金属材料の製造方法、及び設計支援装置 |
EP19911629.4A EP3916651A4 (en) | 2019-01-21 | 2019-12-02 | DESIGN ASSISTING METHOD FOR METALLIC MATERIAL, PREDICTION MODEL GENERATION METHOD, METALLIC MATERIAL MANUFACTURING METHOD AND DESIGN ASSISTING DEVICE |
US17/424,164 US20220100932A1 (en) | 2019-01-21 | 2019-12-02 | Design support method for metal material, prediction model generation method, metal material production method, and design support apparatus |
CN201980089525.8A CN113330468A (zh) | 2019-01-21 | 2019-12-02 | 金属材料的设计支援方法、预测模型的生成方法、金属材料的制造方法以及设计支援装置 |
MX2021008722A MX2021008722A (es) | 2019-01-21 | 2019-12-02 | Metodo de asistencia de dise?o para material de metal, metodo de generacion de modelo de prediccion, metodo de produccion de material de metal y aparato de asistencia de dise?o. |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JPPCT/JP2019/001675 | 2019-01-21 | ||
PCT/JP2019/001675 WO2020152750A1 (ja) | 2019-01-21 | 2019-01-21 | 金属材料の設計支援方法、予測モデルの生成方法、金属材料の製造方法、及び設計支援装置 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2020152993A1 true WO2020152993A1 (ja) | 2020-07-30 |
Family
ID=71736136
Family Applications (2)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2019/001675 WO2020152750A1 (ja) | 2019-01-21 | 2019-01-21 | 金属材料の設計支援方法、予測モデルの生成方法、金属材料の製造方法、及び設計支援装置 |
PCT/JP2019/047091 WO2020152993A1 (ja) | 2019-01-21 | 2019-12-02 | 金属材料の設計支援方法、予測モデルの生成方法、金属材料の製造方法、及び設計支援装置 |
Family Applications Before (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2019/001675 WO2020152750A1 (ja) | 2019-01-21 | 2019-01-21 | 金属材料の設計支援方法、予測モデルの生成方法、金属材料の製造方法、及び設計支援装置 |
Country Status (8)
Country | Link |
---|---|
US (1) | US20220100932A1 (ja) |
EP (1) | EP3916651A4 (ja) |
JP (1) | JP7028316B2 (ja) |
KR (1) | KR20210114993A (ja) |
CN (1) | CN113330468A (ja) |
BR (1) | BR112021014096A2 (ja) |
MX (1) | MX2021008722A (ja) |
WO (2) | WO2020152750A1 (ja) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2022054500A1 (ja) | 2020-09-14 | 2022-03-17 | Jfeスチール株式会社 | 材料特性値予測システム及び金属板の製造方法 |
JP2023026082A (ja) * | 2021-08-12 | 2023-02-24 | 株式会社三菱総合研究所 | 情報処理装置、情報処理方法及びプログラム |
WO2023243544A1 (ja) * | 2022-06-15 | 2023-12-21 | Dic株式会社 | 熱硬化性エポキシ樹脂組成物の探索方法、情報処理装置、及びプログラム |
Families Citing this family (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2022132895A (ja) * | 2021-03-01 | 2022-09-13 | 株式会社Uacj | 合金材料の特性を予測する製造支援システム、予測モデルを生成する方法およびコンピュータプログラム |
CN114854978A (zh) * | 2022-04-06 | 2022-08-05 | 武汉钢铁有限公司 | 一种预测带钢跑偏值的方法和装置 |
CN116694919B (zh) * | 2023-06-12 | 2023-11-21 | 无锡市海峰海林精密轴承有限公司 | 一种轴承贝氏体热处理工艺的优化方法及系统 |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2010106314A (ja) * | 2008-10-30 | 2010-05-13 | Jfe Steel Corp | 鋼製品の製造方法 |
JP2010172962A (ja) * | 2009-02-02 | 2010-08-12 | Toshiba Mitsubishi-Electric Industrial System Corp | 圧延製品の特性予測方法 |
JP2011103036A (ja) | 2009-11-10 | 2011-05-26 | Hitachi Ltd | 設計モデル活用型設計探査方法及び装置 |
WO2018062398A1 (ja) | 2016-09-30 | 2018-04-05 | 株式会社Uacj | アルミニウム製品の特性予測装置、アルミニウム製品の特性予測方法、制御プログラム、および記録媒体 |
Family Cites Families (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US9729639B2 (en) * | 2001-08-10 | 2017-08-08 | Rockwell Automation Technologies, Inc. | System and method for dynamic multi-objective optimization of machine selection, integration and utilization |
EP1975860A4 (en) * | 2006-01-13 | 2011-05-04 | Jfe Steel Corp | DEVICE AND METHOD FOR CREATING A PREDICTION FORMULA |
US8392418B2 (en) * | 2009-06-25 | 2013-03-05 | University Of Tennessee Research Foundation | Method and apparatus for predicting object properties and events using similarity-based information retrieval and model |
CN103226739B (zh) * | 2013-04-24 | 2016-11-02 | 辽宁工程技术大学 | 一种基于泛函网络的周期来压预测方法 |
CN106463959B (zh) * | 2014-05-19 | 2019-12-03 | 杰富意钢铁株式会社 | 电力供需指导装置及电力供需指导方法 |
WO2016038705A1 (ja) * | 2014-09-10 | 2016-03-17 | 東芝三菱電機産業システム株式会社 | 圧延シミュレーション装置 |
CN105911037B (zh) * | 2016-04-19 | 2018-05-15 | 湖南科技大学 | 锰矿区土水界面污染流中锰及伴生重金属分布预测方法 |
CN106355330A (zh) * | 2016-08-31 | 2017-01-25 | 郑州航空工业管理学院 | 基于径向基神经网络预测模型的多响应参数优化方法 |
JP6610518B2 (ja) | 2016-11-30 | 2019-11-27 | 株式会社ダイフク | 検査装置 |
-
2019
- 2019-01-21 WO PCT/JP2019/001675 patent/WO2020152750A1/ja active Application Filing
- 2019-12-02 US US17/424,164 patent/US20220100932A1/en active Pending
- 2019-12-02 EP EP19911629.4A patent/EP3916651A4/en active Pending
- 2019-12-02 WO PCT/JP2019/047091 patent/WO2020152993A1/ja unknown
- 2019-12-02 JP JP2020516503A patent/JP7028316B2/ja active Active
- 2019-12-02 KR KR1020217025687A patent/KR20210114993A/ko active IP Right Grant
- 2019-12-02 BR BR112021014096-4A patent/BR112021014096A2/pt unknown
- 2019-12-02 MX MX2021008722A patent/MX2021008722A/es unknown
- 2019-12-02 CN CN201980089525.8A patent/CN113330468A/zh active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2010106314A (ja) * | 2008-10-30 | 2010-05-13 | Jfe Steel Corp | 鋼製品の製造方法 |
JP2010172962A (ja) * | 2009-02-02 | 2010-08-12 | Toshiba Mitsubishi-Electric Industrial System Corp | 圧延製品の特性予測方法 |
JP2011103036A (ja) | 2009-11-10 | 2011-05-26 | Hitachi Ltd | 設計モデル活用型設計探査方法及び装置 |
WO2018062398A1 (ja) | 2016-09-30 | 2018-04-05 | 株式会社Uacj | アルミニウム製品の特性予測装置、アルミニウム製品の特性予測方法、制御プログラム、および記録媒体 |
Non-Patent Citations (4)
Title |
---|
ADACHI, YOSHITAKA ET AL.: "Materials Integration System 'MIPHA' powered by Machine Learning", SYSTEMS, CONTROL AND INFORMATION, vol. 61, no. 5, 15 May 2017 (2017-05-15), pages 20 - 25, XP055725666 * |
HASEGAWA, HIROSHI: "Systematic approach: From requirements/requests to optimum design", PROCEEDINGS OF THE 28TH JAPAN SIMULATION CONFERENCE, 11 June 2009 (2009-06-11), pages 113 - 116 * |
See also references of EP3916651A4 |
TAKEDA SEIJI ET AL.: "Material discovery by AI", ANNUAL CONFERENCE OF THE JAPANESE SOCIETY FOR ARTIFICIAL INTELLIGENCE, 2018 |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2022054500A1 (ja) | 2020-09-14 | 2022-03-17 | Jfeスチール株式会社 | 材料特性値予測システム及び金属板の製造方法 |
JP2023026082A (ja) * | 2021-08-12 | 2023-02-24 | 株式会社三菱総合研究所 | 情報処理装置、情報処理方法及びプログラム |
JP7270005B2 (ja) | 2021-08-12 | 2023-05-09 | 株式会社三菱総合研究所 | 情報処理装置、情報処理方法及びプログラム |
WO2023243544A1 (ja) * | 2022-06-15 | 2023-12-21 | Dic株式会社 | 熱硬化性エポキシ樹脂組成物の探索方法、情報処理装置、及びプログラム |
Also Published As
Publication number | Publication date |
---|---|
US20220100932A1 (en) | 2022-03-31 |
CN113330468A (zh) | 2021-08-31 |
EP3916651A1 (en) | 2021-12-01 |
WO2020152750A1 (ja) | 2020-07-30 |
KR20210114993A (ko) | 2021-09-24 |
BR112021014096A2 (pt) | 2021-09-21 |
EP3916651A4 (en) | 2022-04-06 |
JP7028316B2 (ja) | 2022-03-02 |
JPWO2020152993A1 (ja) | 2021-02-18 |
MX2021008722A (es) | 2021-08-24 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2020152993A1 (ja) | 金属材料の設計支援方法、予測モデルの生成方法、金属材料の製造方法、及び設計支援装置 | |
KR101011546B1 (ko) | 예측식 작성장치 및 예측식 작성방법 | |
Sanz-García et al. | GA-PARSIMONY: A GA-SVR approach with feature selection and parameter optimization to obtain parsimonious solutions for predicting temperature settings in a continuous annealing furnace | |
Wang et al. | Applying input variables selection technique on input weighted support vector machine modeling for BOF endpoint prediction | |
Cui et al. | An improved Lagrangian relaxation approach to scheduling steelmaking-continuous casting process | |
Peng et al. | Coupling physics in machine learning to predict properties of high-temperatures alloys | |
JP5516390B2 (ja) | 品質予測装置、品質予測方法、プログラムおよびコンピュータ読み取り可能な記録媒体 | |
CN114611844B (zh) | 一种转炉出钢过程合金加入量的确定方法和系统 | |
CN106802977A (zh) | 一种用于烧结矿性能指标预测及综合质量评价方法 | |
Chakraborti | Critical assessment 3: The unique contributions of multi-objective evolutionary and genetic algorithms in materials research | |
Xing et al. | Hybrid intelligent parameter estimation based on grey case-based reasoning for laminar cooling process | |
JP2008112288A (ja) | 予測式作成装置、結果予測装置、品質設計装置、予測式作成方法及び製品の製造方法 | |
WO2021004198A1 (zh) | 一种板材性能的预测方法及装置 | |
CN113128124B (zh) | 基于改进神经网络的多牌号C-Mn钢力学性能预测方法 | |
JP2013080458A (ja) | 品質予測装置、操業条件決定方法、品質予測方法、コンピュータプログラムおよびコンピュータ読み取り可能な記憶媒体 | |
JP2017120640A (ja) | 電磁鋼板の磁気特性予測装置及び磁気特性制御装置 | |
Chen et al. | Application of grey relational analysis and extreme learning machine method for predicting silicon content of molten iron in blast furnace | |
Wang et al. | Strip hardness prediction in continuous annealing using multiobjective sparse nonlinear ensemble learning with evolutionary feature selection | |
WO2020148918A1 (ja) | 金属材料の設計支援方法及び設計支援装置 | |
JP5845945B2 (ja) | 操業予測装置、操業予測方法、及びコンピュータプログラム | |
Dong et al. | Local multi-model integrated soft sensor based on just-in-time learning for mechanical properties of hot strip mill process | |
JP5682131B2 (ja) | 鋼材の材質予測装置 | |
Wu et al. | Deformation resistance prediction of tandem cold rolling based on grey wolf optimization and support vector regression | |
CN115456264B (zh) | 一种中小型转炉的终点碳含量和终点温度预测方法 | |
Zhang et al. | Deep learning-based prediction framework of temperature control time for wide-thick slab hot rolling production |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
ENP | Entry into the national phase |
Ref document number: 2020516503 Country of ref document: JP Kind code of ref document: A |
|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 19911629 Country of ref document: EP Kind code of ref document: A1 |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
REG | Reference to national code |
Ref country code: BR Ref legal event code: B01A Ref document number: 112021014096 Country of ref document: BR |
|
ENP | Entry into the national phase |
Ref document number: 20217025687 Country of ref document: KR Kind code of ref document: A |
|
ENP | Entry into the national phase |
Ref document number: 2019911629 Country of ref document: EP Effective date: 20210823 |
|
ENP | Entry into the national phase |
Ref document number: 112021014096 Country of ref document: BR Kind code of ref document: A2 Effective date: 20210716 |