JP7407893B1 - Composition proposal device - Google Patents

Abstract

The present invention provides a composition proposal device that proposes parameters for a polymer composite material having desired properties. [Solution] A composition proposal device according to the present invention is a device that proposes the type and mass ratio of a type of polymer and a type of component other than the polymer, which are included as constituent components in a polymer composite material. A setting section that sets target values for the physical properties of a virtual polymer composite material, and a trained model that has been trained so that parameters including descriptors of the polymer composite material correspond to physical properties of the polymer composite material. , an optimization unit that changes the parameters so that the physical properties of the virtual polymer composite material, which are predicted by inputting parameters including descriptors of the virtual polymer composite material, approach target values. , the optimization unit changes the parameters each time the physical properties of the virtual polymer composite material are predicted, thereby determining the type of polymer and the type of polymer contained in the virtual polymer composite material as constituent components. Optimize the types and mass ratios of other components. [Selection diagram] Figure 4

Description

The present invention relates to a composition proposal device.

Polymer composite materials formed from compositions containing multiple polymeric materials are widely used as materials for automobile parts, home appliances, food and medical containers, building materials and civil engineering materials, and the like. As the polymer composite material, for example, a propylene resin composition containing a propylene polymer is used.

Examples of the propylene resin composition include a propylene homopolymer, a random copolymer of propylene and a monomer other than propylene, or a propylene polymer containing a heterophasic propylene polymer material and an ethylene-α-olefin copolymer. A propylene resin composition containing a polymer has been disclosed (see, for example, Patent Document 1).

Japanese Patent Application Publication No. 2018-178107

Here, when manufacturing a polymer composite material, the type of polymer used for manufacturing the polymer composite material and It is necessary to determine the composition of the polymer composite material. In order to determine the type of polymer and the composition of the polymer composite material, we actually conduct experiments to manufacture polymer composite materials using compositions containing various polymers and measure their physical properties. Since it is necessary to search for the composition of a polymer composite material with specific physical properties, it requires a great deal of effort and preparation of raw materials and equipment. Therefore, from the viewpoint of improving the energy efficiency that occurs when selecting polymer composite materials, there is a need for a method that can efficiently suggest the physical properties of polymer composite materials from parameters such as composition and structure of polymer composite materials. There is.

Furthermore, one aspect of the present invention aims to provide a composition proposal device that proposes parameters for a polymer composite material having desired properties.

One aspect of the present invention is
A composition proposal device that proposes the types and mass ratios of a type of polymer and components other than the type of polymer contained as constituent components in a polymer composite material, comprising:
a setting section for setting target values of physical properties of a virtual polymer composite material;
The descriptor of the virtual polymer composite material is added to a trained model trained using a training data set in which parameters including the descriptor of the polymer composite material are associated with physical properties of the polymer composite material. an optimization unit that changes parameters including a descriptor of the virtual polymer composite material so that physical properties of the virtual polymer composite material predicted by inputting parameters including a descriptor of the virtual polymer composite material approach the target value; and,
has
The optimization unit changes the parameters including the descriptor of the virtual polymer composite material each time the physical properties of the virtual polymer composite material are predicted. This is a composition proposal device that optimizes the types and mass ratios of the above-mentioned one type of polymer and components other than the above-mentioned one type of polymer, which are included as constituent components.

One aspect of the present invention can provide a composition proposal device that proposes parameters for a polymer composite material having desired properties.

FIG. 2 is an explanatory diagram showing that a combination of the composition, structure of constituent components, etc. of a polymer composite material is proposed based on the physical properties of the polymer composite material. FIG. 1 is a block diagram showing a schematic configuration of a learning device. FIG. 2 is a block diagram showing the configuration of a prediction device. FIG. 1 is a block diagram showing the configuration of a composition proposal device according to an embodiment of the present invention. FIG. 2 is a block diagram showing another configuration of the composition proposal device according to the embodiment of the present invention. FIG. 2 is a block diagram showing the hardware configurations of a learning device, a prediction device, and a composition proposal device. It is a flowchart explaining a learning method. It is a flowchart explaining a prediction method. 1 is a flowchart illustrating a composition proposal method according to an embodiment of the present invention. 7 is another flowchart illustrating a composition proposal method according to an embodiment of the present invention. It is a figure showing the relationship between the measured value of a bending elastic modulus, and the predicted value of a bending elastic modulus. It is a figure showing the relationship between the measured value of room temperature IZOD, and the predicted value of room temperature IZOD. It is a figure showing the relationship between the measured value of low-temperature IZOD, and the predicted value of low-temperature IZOD. FIG. 3 is a diagram showing a relationship between an actual measured value of MFR and a predicted value of MFR. It is a figure showing the relationship between the predicted value of bending elastic modulus, and the predicted value of normal temperature IZOD.

Embodiments of the present invention will be described in detail below. In order to facilitate understanding of the explanation, the same components in each drawing are denoted by the same reference numerals, and redundant explanation will be omitted. Furthermore, in the present embodiment, "~" indicating a numerical range means that the lower and upper limits include the numerical values written before and after it, unless otherwise specified.

A composition proposal device according to this embodiment will be explained. Figure 1 proposes combinations of the composition of polymer composite materials and the structures of constituent components such as polymer materials (hereinafter simply referred to as polymers) contained in polymer composite materials based on the physical properties of polymer composite materials. FIG.

As shown in FIG. 1, the composition proposal device according to the present embodiment calculates the physical properties of a polymer composite material obtained by inputting information regarding parameters including descriptors such as the composition and structure of the composition into a trained model. We optimize and propose parameters including descriptors such as the composition and structure of the composition so that the physical properties of the polymer composite material approach the desired target values.

In this embodiment, the polymer composite material is a composition containing a polymer as a component. Further, the polymer composite material may be a molded product formed into a predetermined shape depending on its use. In this embodiment, a case will be described in which the polymer composite material is a propylene-based resin composition containing a propylene-based polymer.

In explaining the composition proposing device according to this embodiment, a polymer composite material will be explained. Note that the polymer composite material is not limited to a propylene-based resin composition, but may be any composition containing a polymer as a constituent component.

<Propylene resin composition>
The propylene resin composition contains at least one propylene polymer (A).

[Propylene polymer (A)]
The propylene polymer (A) is a polymer having monomer units derived from propylene. Examples of the propylene polymer (A) include propylene homopolymers, random copolymers of propylene and monomers other than propylene, and heterophasic propylene polymer materials. The propylene-based resin composition may contain only one type of propylene-based polymer (A), or may contain two or more types of propylene-based polymers (A). The propylene polymer (A) should contain at least one member selected from the group consisting of a propylene homopolymer and a heterophasic propylene polymer material from the viewpoint of the rigidity and impact resistance of the molded article of the propylene resin composition. is preferred.

(Propylene homopolymer)
When the propylene polymer (A) contains a propylene homopolymer, the intrinsic viscosity (η) of the propylene homopolymer depends on the fluidity of the propylene resin composition when melted and the molded product of the propylene composition. From the viewpoint of toughness, it is preferably 0.10 dL/g to 2.00 dL/g.

In addition, in this embodiment, the intrinsic viscosity number (unit: dL/g) is a value measured at a temperature of 135° C. using tetralin as a solvent by the following method.

Using an Ubbelohde viscometer, the reduced viscosity is measured at three concentrations: 0.1 g/dL, 0.2 g/dL, and 0.5 g/dL. The limiting viscosity number is determined by an extrapolation method in which the reduced viscosity is plotted against the concentration and the concentration is extrapolated to zero. A method for calculating the limiting viscosity by extrapolation is described, for example, in "Polymer Solution, Polymer Experimental Science 11" (Kyoritsu Shuppan Co., Ltd., 1982), page 491.

The molecular weight distribution (Mw/Mn) of the propylene homopolymer is preferably 3.0 or more. The molecular weight distribution of the propylene polymer (A) is preferably 3.0 to 30.0.

In this embodiment, the molecular weight distribution is calculated using the weight average molecular weight (Mw) and number average molecular weight (Mn) measured by gel permeation chromatography (GPC) under the following conditions. It refers to the ratio (Mw/Mn) of weight average molecular weight (Mw) to Mn).
((conditions))
Equipment: Tosoh Corporation HLC-8121 GPC/HT
Separation column: 3 GMHHR-H(S)HT manufactured by Tosoh Corporation Measurement temperature: 140°C
Carrier: Orthodichlorobenzene Flow rate: 1.0 mL/min Sample concentration: Approximately 1 mg/mL
Sample injection volume: 400μL
Detector: Differential refraction calibration curve creation method: Use standard polystyrene

A propylene homopolymer can be produced, for example, by polymerizing propylene using a polymerization catalyst.

Examples of the polymerization catalyst include Ziegler type catalysts; Ziegler-Natta type catalysts; catalysts consisting of a compound of transition metal of Group 4 of the periodic table having a cyclopentadienyl ring and alkylaluminoxane; periodic catalysts having a cyclopentadienyl ring A catalyst consisting of a transition metal compound in Group 4 of the table, a compound that reacts with the transition metal compound to form an ionic complex, and an organoaluminum compound; Examples include catalysts that are modified by supporting a compound of a transition metal of Group 4 of the periodic table having a pentadienyl ring, a compound that forms an ionic complex, an organoaluminum compound, etc.

Examples of polymerization methods include bulk polymerization, solution polymerization, and gas phase polymerization.

Examples of the polymerization method include a batch method, a continuous method, and a combination thereof. The polymerization method may be a multistage method in which a plurality of polymerization reactors are connected in series.

Various conditions (polymerization temperature, polymerization pressure, monomer concentration, catalyst input amount, polymerization time, etc.) in the polymerization step may be appropriately determined depending on the molecular structure of the target polymer.

After the polymerization process, the polymer is dried at a temperature below the melting temperature of the polymer, if necessary, in order to remove residual solvent contained in the polymer and ultra-low molecular weight oligomers produced as by-products during manufacturing. You may.

(Random copolymer of propylene and a monomer other than propylene)
A random copolymer of propylene and a monomer other than propylene contains a monomer unit derived from propylene and a monomer unit derived from a monomer other than propylene. The random copolymer preferably contains 0.01% by mass to 20% by mass of monomer units derived from monomers other than propylene, based on the mass of the random copolymer.

Examples of monomers other than propylene include ethylene and α-olefins having 4 to 12 carbon atoms. Among these, at least one selected from ethylene and α-olefin having 4 to 10 carbon atoms is preferred, and at least one selected from the group consisting of ethylene, 1-butene, 1-hexene, and 1-octene is more preferred.

Note that "α-olefin" means an aliphatic unsaturated hydrocarbon having a carbon-carbon unsaturated double bond at the α position.

Examples of the random copolymer include propylene-ethylene random copolymer, propylene-1-butene random copolymer, propylene-1-hexene random copolymer, propylene-1-octene random copolymer, and propylene-ethylene random copolymer. -1-butene random copolymer, propylene-ethylene-1-hexene random copolymer, and propylene-ethylene-1-octene random copolymer.

When the propylene polymer (A) includes a random copolymer of propylene and a monomer other than propylene, the intrinsic viscosity (η) of the random copolymer is determined by the flow rate when the propylene resin composition is melted. From the viewpoint of performance, for example, it is preferably 1.00 dL/g to 10.00 dL/g.

The random copolymer preferably has a molecular weight distribution (Mw/Mn) of 3.0 to 30.0.

The random copolymer can be produced, for example, by polymerizing propylene and a monomer other than propylene according to a polymerization catalyst, a polymerization method, and a polymerization system that can be used in the production of a propylene homopolymer.

(Heterophagic propylene polymer material)
The heterophasic propylene polymer material consists of a polymer (I) containing 80% by mass or more of monomer units derived from propylene (however, the total mass of the polymer (I) is 100% by mass), and ethylene. and a polymer (II) containing a monomer unit derived from at least one α-olefin selected from the group consisting of α-olefins having 4 to 12 carbon atoms and a monomer unit derived from propylene. . Heterophasic propylene polymeric material means a mixture having a structure in which polymer (II) is dispersed in a matrix of polymer (I).

The heterophasic propylene polymer material can be produced, for example, by performing a first polymerization step to form polymer (I) and a second polymerization step to form polymer (II). Examples of the polymerization catalyst, polymerization method, and polymerization system employed in these polymerization steps are the same as those described above.

Polymer (I) may be, for example, a propylene homopolymer, or may contain monomer units derived from monomers other than propylene. When the polymer (I) contains a monomer unit derived from a monomer other than propylene, this content is, for example, 0.01% by mass or more and 20% by mass based on the total mass of the polymer (I). It may be less than % by mass.

Examples of monomers other than propylene include ethylene and α-olefins having 4 or more carbon atoms. Among these, at least one selected from the group consisting of ethylene and α-olefins having 4 to 10 carbon atoms is preferred, and at least one selected from the group consisting of ethylene, 1-butene, 1-hexene, and 1-octene is more preferred. .

Examples of polymers containing monomer units derived from monomers other than propylene include propylene-ethylene copolymer, propylene-1-butene copolymer, propylene-1-hexene copolymer, propylene-1 -octene copolymer, propylene-ethylene-1-butene copolymer, propylene-ethylene-1-hexene copolymer and propylene-ethylene-1-octene copolymer.

Polymer (I) is a propylene homopolymer, a propylene-ethylene copolymer, a propylene-1-butene copolymer, a propylene-ethylene-1- Butene copolymers are preferred, and propylene homopolymers are more preferred.

The content of polymer (I) is preferably 50% by mass to 99% by mass based on the total mass of the heterophasic propylene polymer material.

Polymer (II) contains 40% by mass or more of monomer units derived from at least one α-olefin selected from the group consisting of ethylene and α-olefins having 4 to 12 carbon atoms, and It is preferable to contain monomer units derived from the above.

In the polymer (II), the content of monomer units derived from at least one α-olefin selected from the group consisting of ethylene and α-olefins having 4 to 12 carbon atoms is 40 to 70% by mass. It may be 45% to 60% by mass.

In the polymer (II), at least one α-olefin selected from the group consisting of ethylene and α-olefins having 4 to 12 carbon atoms is selected from the group consisting of ethylene and α-olefins having 4 to 10 carbon atoms. At least one selected from the group consisting of ethylene, 1-butene, 1-hexene, 1-octene and 1-decene is more preferable, and at least one selected from the group consisting of ethylene and 1-butene is preferable. One type is more preferable.

Examples of the polymer (II) include propylene-ethylene copolymer, propylene-ethylene-1-butene copolymer, propylene-ethylene-1-hexene copolymer, propylene-ethylene-1-octene copolymer, Examples include propylene-ethylene-1-decene copolymer, propylene-1-butene copolymer, propylene-1-hexene copolymer, propylene-1-octene copolymer and propylene-1-decene copolymer. Among these, propylene-ethylene copolymer, propylene-1-butene copolymer and propylene-ethylene-1-butene copolymer are preferred, and propylene-ethylene copolymer is more preferred.

The content of polymer (II) is preferably 1% by mass to 50% by mass based on the total mass of the heterophasic propylene polymer material.

The heterophasic propylene polymeric material may include a xylene insoluble (CXIS) component and a xylene soluble (CXS) component.

The content of the CXIS component in the heterophasic propylene polymer material is preferably 50% by weight to 99% by weight based on the total weight of the heterophasic propylene polymer material.

The content of the CXS component in the heterophasic propylene polymer material is preferably 1% by mass to 50% by mass based on the total mass of the heterophasic propylene polymer material.

In this embodiment, the CXIS component in the heterophasic propylene polymeric material is mainly composed of polymer (I), and the CXS component in the heterophasic propylene polymeric material is mainly composed of polymer (II). it is conceivable that.

Examples of the heterophasic propylene polymer material include (propylene)-(propylene-ethylene) polymer material, (propylene)-(propylene-ethylene-1-butene) polymer material, (propylene)-(propylene-ethylene-1- hexene) polymeric material, (propylene)-(propylene-ethylene-1-octene) polymeric material, (propylene)-(propylene-1-butene) polymeric material, (propylene)-(propylene-1-hexene) polymeric material, ( (propylene)-(propylene-1-octene) polymeric material, (propylene)-(propylene-1-decene) polymeric material, (propylene-ethylene)-(propylene-ethylene) polymeric material, (propylene-ethylene)-(propylene- (ethylene-1-butene) polymeric material, (propylene-ethylene)-(propylene-ethylene-1-hexene) polymeric material, (propylene-ethylene)-(propylene-ethylene-1-octene) polymeric material, (propylene-ethylene) -(Propylene-ethylene-1-decene) polymeric material, (propylene-ethylene)-(propylene-1-butene) polymeric material, (propylene-ethylene)-(propylene-1-hexene) polymeric material, (propylene-ethylene) -(Propylene-1-octene) polymeric material, (propylene-ethylene)-(propylene-1-decene) polymeric material, (propylene-1-butene)-(propylene-ethylene) polymeric material, (propylene-1-butene) -(Propylene-ethylene-1-butene) polymeric material, (propylene-1-butene)-(propylene-ethylene-1-hexene) polymeric material, (propylene-1-butene)-(propylene-ethylene-1-octene) Polymerized material, (propylene-1-butene)-(propylene-ethylene-1-decene) polymerized material, (propylene-1-butene)-(propylene-1-butene) polymerized material, (propylene-1-butene)-( Propylene-1-hexene) polymeric material, (propylene-1-butene)-(propylene-1-octene) polymeric material, (propylene-1-butene)-(propylene-1-decene) polymeric material, (propylene-1-butene)-(propylene-1-decene) polymeric material, hexene)-(propylene-1-hexene) polymeric material, (propylene-1-hexene)-(propylene-1-octene) polymeric material, (propylene-1-hexene)-(propylene-1-decene) polymeric material, ( Examples include propylene-1-octene)-(propylene-1-octene) polymeric materials and (propylene-1-octene)-(propylene-1-decene) polymeric materials.

Note that "(propylene)-(propylene-ethylene) polymer material" refers to "heterophasic propylene in which polymer (I) is a propylene homopolymer and polymer (II) is a propylene-ethylene copolymer. "polymerized material". The same applies to other similar expressions.

The intrinsic viscosity (ηI) of the polymer (I) is preferably from 0.10 dL/g to 2.00 dL/g, for example.

The intrinsic viscosity (ηII) of the polymer (II) is preferably 1.00 dL/g to 10.00 dL/g.

The ratio of the intrinsic viscosity number (ηII) of the polymer (II) to the intrinsic viscosity number (ηI) of the polymer (I) ([η]II/[η]I) is preferably from 1 to 20.

Examples of the method for measuring the intrinsic viscosity (ηI) of the polymer (I) include a method of forming the polymer (I) and then measuring the intrinsic viscosity of the polymer.

The intrinsic viscosity number (ηII) of the polymer (II) is, for example, the intrinsic viscosity number (ηTotal) of the heterophasic propylene polymer material, the intrinsic viscosity number (ηI) of the polymer (I), and the intrinsic viscosity number (ηI) of the polymer (II) and the polymer. It can be calculated by the following formula (I) using the content of the combined substance (I).
ηII=(ηTotal-ηI×XI)/XII...(I)

In addition, in formula (I), ηTotal, ηI, XI, and XII each have the following meanings.
ηTotal: Intrinsic viscosity number (dL/g) of heterophasic propylene polymer material
ηI: Intrinsic viscosity number (dL/g) of polymer (I)
XI: Ratio of the mass of polymer (I) to the total mass of heterophasic propylene polymer material (mass of polymer (I)/mass of heterophasic propylene polymer material)
XII: Ratio of mass of polymer (II) to total mass of heterophasic propylene polymer material (mass of polymer (II)/mass of heterophasic propylene polymer material)

Here, XI and XII can be determined from the mass balance during polymerization.

Note that XII may be calculated by measuring the heat of fusion of the polymer (I) and the heat of fusion of the heterophasic propylene polymer material, and using the following formula (II).
XII=1-(ΔHf)T/(ΔHf)P...(II)

In formula (II), ΔHfT and ΔHfP have the following meanings.
ΔHfT: Heat of fusion of heterophasic propylene polymer material (J/g)
ΔHfP: Heat of fusion of polymer (I) (J/g)

The intrinsic viscosity (ηCXIS) of the CXIS component is preferably 0.10 dL/g to 2.00 dL/g.

The intrinsic viscosity (ηCXS) of the CXS component is preferably 1.00 dL/g to 10.00 dL/g.

The ratio of the limiting viscosity number (ηCXS) of the CXS component to the limiting viscosity number (ηCXIS) of the CXIS component (ηCXS/ηCXIS) is preferably from 1 to 20.

The molecular weight distribution (Mw(I)/Mn(I)) of the polymer (I) is preferably 3.0 or more.

The molecular weight distribution (Mw(CXIS)/Mn(CXIS)) of the CXIS component is preferably 3.0 or more.

The melt flow rate (MFR) of the propylene polymer (A) at a temperature of 230°C and a load of 2.16 kgf is 5 g/10 minutes to 300 g/10 minutes from the viewpoint of moldability of the propylene resin composition. It is preferable.

Note that in this embodiment, MFR refers to a value measured in accordance with JIS K7210.

[Polyolefin elastomer (B)]
The propylene resin composition may also contain a polyolefin elastomer (B). The polyolefin elastomer (B) is based on the content of monomer units derived from ethylene contained in the polyolefin elastomer (B) and the α-olefin having 4 or more carbon atoms, with the total mass of the polyolefin elastomer (B) being 100% by mass. The total content of the derived monomer units may be 100% by mass.

Examples of α-olefins having 4 or more carbon atoms include α-olefins having 4 to 12 carbon atoms. Examples of α-olefins having 4 to 12 carbon atoms include 1-butene, 1-pentene, 1-hexene, 4-methyl-1-pentene, 1-octene, and 1-decene. Among them, 1-butene, 1-hexene, and 1-octene are preferred. The above α-olefin may be an α-olefin having a cyclic structure such as vinylcyclopropane and vinylcyclobutane.

Examples of the polyolefin elastomer (B) include ethylene-1-butene copolymer, ethylene-1-hexene copolymer, ethylene-1-octene copolymer, ethylene-1-decene copolymer, ethylene-(3 -methyl-1-butene) copolymer, and a copolymer of ethylene and an α-olefin having a cyclic structure (ethylene-α-olefin copolymer).

The ethylene-α-olefin copolymer is a copolymer containing a monomer unit derived from ethylene and a monomer unit derived from an α-olefin having 4 or more carbon atoms. This means that it does not contain any monomeric units.

In the polyolefin elastomer (B), the content of monomer units derived from α-olefin having 4 or more carbon atoms is 1% by mass to 49% by mass based on the total mass of the polyolefin elastomer (B). is preferred.

The MFR of the polyolefin elastomer (B) at a temperature of 190° C. and a load of 2.16 kgf is preferably 0.1 g/10 minutes to 80 g/10 minutes.

The density of the polyolefin elastomer (B) is preferably 0.850 g/cm ³ to 0.890 g/cm ³ from the viewpoint of impact resistance of the molded article of the propylene resin composition.

The polyolefin elastomer (B) can be produced by polymerizing ethylene and an α-olefin having 4 or more carbon atoms using a polymerization catalyst.

Examples of the polymerization catalyst include homogeneous catalysts represented by metallocene catalysts, and Ziegler-Natta type catalysts.

As a homogeneous catalyst, for example, a catalyst consisting of a compound of a transition metal of group 4 of the periodic table having a cyclopentadienyl ring and an alkylaluminoxane; a compound of a transition metal of group 4 of the periodic table having a cyclopentadienyl ring , a catalyst consisting of a compound that reacts with the transition metal compound to form an ionic complex and an organoaluminum compound; Examples include catalysts modified by supporting a group 4 transition metal compound, a compound forming an ionic complex, an organoaluminum compound, etc.).

Examples of Ziegler-Natta type catalysts include catalysts in which a titanium-containing solid transition metal component and an organic metal component are combined.

A commercially available product may be used as the polyolefin elastomer (B). Commercially available products include, for example, Engage (registered trademark) manufactured by Dow Chemical Japan Co., Ltd., Tafmer (registered trademark) manufactured by Mitsui Chemicals, Inc., Neozex (registered trademark) manufactured by Prime Polymer Co., Ltd., Urtozex (registered trademark), Sumitomo Examples include Excelen FX (registered trademark), Sumikasen (registered trademark), and Espren SPO (registered trademark) manufactured by Kagaku Co., Ltd.

[Filler (C)]
The propylene resin composition may also contain a filler (C). The filler (C) contained in the propylene resin composition has, for example, a function of increasing the dimensional stability of a molded article of the propylene resin composition. Examples of the filler (C) include inorganic fillers and organic fillers. The propylene resin composition may contain only one type of filler (C), or may contain two or more types of filler (C).

Inorganic fillers include glass, silicate minerals, alumina, silica, silicon dioxide, titanium oxide, iron oxide, aluminum oxide, magnesium oxide, antimony oxide, barium ferrite, strontium ferrite, beryllium oxide, magnesium hydroxide, Mention may be made of aluminum hydroxide, basic magnesium carbonate, calcium carbonate, magnesium carbonate, carbonate minerals, calcium sulfate, magnesium sulfate, basic magnesium sulfate, calcium sulfite, carbon black and cadmium sulfide.

Organic fillers include polyester, aromatic polyamide, cellulose and vinylon.

The filler (C) is preferably an inorganic filler from the viewpoint of the rigidity, impact resistance, and dimensional stability of the molded article, and more preferably talc, which is a plate-like silicate mineral.

The shape of the filler may be plate-like, needle-like, or fibrous.

The average particle diameter D ₅₀ of the filler (C) may be, for example, 0.5 μm to 20.0 μm from the viewpoint of the rigidity, impact resistance, and dimensional stability of the molded article.

The average particle diameter _D50 is determined based on volume-based particle diameter distribution measurement data measured by laser diffraction or centrifugal sedimentation according to the method specified in JIS R1629. In the diameter distribution measurement data, it means the particle diameter when the cumulative number of particles from the smaller particle diameter side reaches 50% (50% equivalent particle diameter). The particle diameter defined in this way is generally referred to as a "50% equivalent particle diameter" and is expressed as "D50."

The content of the filler (C) is 10 parts by mass based on a total of 100 parts by mass of the propylene polymer (A) and the polyolefin elastomer (B), from the viewpoint of dimensional stability of the molded article of the propylene resin composition. parts to 60 parts by weight.

The propylene resin composition may also contain components other than those mentioned above. Examples of such components include neutralizing agents, antioxidants, ultraviolet absorbers, nucleating agents, lubricants, antistatic agents, anti-blocking agents, processing aids, organic peroxides, colorants (inorganic pigments, , organic pigments, pigment dispersants, etc.), foaming agents, foaming nucleating agents, plasticizers, flame retardants, crosslinking agents, crosslinking aids, brightness agents, antibacterial agents, and light diffusing agents. The propylene resin composition may contain only one type of these components, or may contain two or more types of these components.

In explaining the composition proposal device according to this embodiment, a learning device and a prediction device will be explained.

<Learning device>
The learning device will be explained. Here, it is assumed that the propylene polymer (A) contained in the polymer composite material is a heterophasic propylene polymer material.

FIG. 2 is a block diagram showing a schematic configuration of the learning device. As shown in FIG. 2, the learning device 1 includes an acquisition section 11, a preprocessing section 12, a learning dataset creation section 13, a learning section 14, and a display section 15. The learning device 1 generates a learned model M1 that predicts physical properties of a polymer composite material containing a polymer.

The acquisition unit 11 acquires parameters including descriptors of the polymer composite material as explanatory variables, and acquires physical properties of the polymer composite material as objective variables.

Descriptors of polymer composite materials include mass ratios of constituent components contained in the polymer composite material, characteristic values and structural values for each type of constituent components, and the like.

The descriptor of a polymer composite material includes the mass ratio of constituent components included in the polymer composite material, and at least one of a characteristic value and a structural value for each type of constituent component.

The constituent components include one type of polymer as an essential component, and may also include components other than the one type of polymer as optional components. A plurality of one kind of polymer and a plurality of components other than one kind of polymer may be included. Since the constituent components include one kind of polymer, the characteristic values and structural values for each type of constituent component may be the characteristic values and structural values of one kind of polymer.

The mass ratio of the constituent components is the composition of the constituent components and their ratio. For example, the ratios of the heterophasic propylene polymer material, the polyolefin elastomer (B), and the filler (C) contained in the polymer composite material.

Characteristic values for each type of component include the intrinsic viscosity and MFR of the component.

The intrinsic viscosity of the constituent components includes the intrinsic viscosity of at least one type of polymer contained in the polymer composite material. The intrinsic viscosity of the constituent components is, for example, the intrinsic viscosity of the heterophasic propylene polymer material and the polyolefin elastomer (B) contained in the polymer composite material.

The intrinsic viscosity of the heterophasic propylene polymer material is the intrinsic viscosity of at least one of polymer (I) and polymer (II) constituting the heterophasic propylene polymer material. ) may be both of the limiting viscosity.

The structural values for each type of component include the molecular weight of the component, the content of the component, the composition ratio of the copolymer contained in the component, and the monomer unit ratio of the copolymer.

It is preferable that the structural value includes at least one of the content of the constituent components and the monomer unit ratio of the copolymer contained in the constituent components. The monomer unit ratio of the copolymer is the monomer unit ratio of the copolymer in a copolymer containing a monomer unit derived from ethylene (for example, a heterophasic propylene polymer material and a polyolefin elastomer (B), etc.). It can also be a ratio.

The molecular weight of the constituent components includes the molecular weight of the polymer. The molecular weight of the component is, for example, the molecular weight of the heterophasic propylene polymer material, polyolefin elastomer (B), etc. contained in the polymer composite material.

The molecular weight of the heterophasic propylene polymer material is the molecular weight of at least one of the polymer (I) and the polymer (II) contained in the heterophasic propylene polymer material, and the molecular weight of the polymer (I) and the polymer (II) contained in the heterophasic propylene polymer material. The molecular weight may be both, or the molecular weight of polymer (I) or polymer (II).

The content of the constituent components includes the content of polymer. The content of the constituent components is, for example, the content of the heterophasic propylene polymer material, the polyolefin elastomer (B), and the filler (C) contained in the polymer composite material.

The content of the heterophasic propylene polymer material is the content of at least one of polymer (I) and polymer (II) that constitute the heterophasic propylene polymer material, and even the content of polymer (II) alone good.

The monomer unit ratio of the copolymer contained in the constituent components is, for example, the amount of ethylene contained in the heterophasic propylene polymer material contained in the polymer composite material, and the amount of ethylene contained in the polyolefin elastomer (B). The amount of ethylene is the ratio of monomer units in a copolymer containing a monomer unit derived from ethylene, such as a heterophasic propylene polymer material and a polyolefin elastomer (B).

The amount of ethylene contained in the heterophasic propylene polymer material is the content of structural units derived from ethylene contained in at least one of polymer (I) and polymer (II) contained in the heterophasic propylene polymer material. , the content of structural units derived from ethylene contained in the polymer (II) may be sufficient.

The composition ratio of the copolymer contained in the constituent components is the composition ratio of the copolymer contained in the heterophasic propylene polymer material, the polyolefin elastomer (B), etc. contained in the polymer composite material. The composition ratio of the copolymer contained in the heterophasic propylene polymer material may be the ratio of polymer (I) to polymer (II).

When the component is a polymer, the characteristic values and structure values for each type of component preferably include parameters indicating the characteristics or structure of the polymer. Therefore, the characteristic values and structural values are parameters indicating the characteristics or structure of the polymeric telophasic propylene polymer material and polyolefin elastomer (B), such as the molecular weight and monomer unit ratio of the copolymer. It may include parameters etc. that are correlated with.

Examples of parameters correlated with molecular weight include intrinsic viscosity and melt flow rate (MFR).

Examples of parameters correlated with the monomer unit ratio of the copolymer include density and glass transition temperature.

It is preferable that at least one of the characteristic value and the structure value of the component includes an average value for each type of component, and the characteristic value for each type of component preferably includes the average characteristic value and the structure value for each type of component. Preferably, the values include average structural values.

That is, the molecular weight of the component, the content of the component, the intrinsic viscosity of the component, the monomer unit ratio of the copolymer contained in the component, and the composition ratio of the copolymer are each determined by the average molecular weight of the component. , the average content of the constituent components, the average intrinsic viscosity of the constituent components, the average monomer unit ratio of the copolymer contained in the constituent components, and the average ratio of the composition of the copolymer.

By using the average values of property values and structural values, it is possible to describe the polymer composite material used for learning whether it contains one polymer as a component or two or more polymers. Since the number of parameters including children can be made uniform, the prediction accuracy of the learned model M1 may be improved. For example, if there is a polymer composite material containing two propylene materials in the learning dataset, the characteristic values and structural values of the first component and the characteristic values and structural values of the second component are used to describe the polymer composite material. Used as a parameter that includes children. In this case, if a polymer composite material containing one propylene material exists in another training dataset, the second component of the material does not exist, so the property values and structural values of the second component are This results in missing data. If there is missing data in parameters including descriptors of polymer composite materials used for learning, the prediction accuracy of the obtained trained model will decrease. In such a case, the average value of the characteristic values and structural values of each component is used as a parameter including a descriptor of the polymer composite material. As a result, the number of parameters including corresponding descriptors can be made the same for a polymer composite material containing two propylene materials and a polymer composite material containing one propylene material, and no loss will occur in the learning data. Therefore, the prediction accuracy of the trained model improves.

The descriptor of the polymer composite material may include, as an additional descriptor, the cross term between the calculated characteristic value or structural value (average characteristic value or average structural value) of each of the above components and the mass ratio. Note that the characteristic value or structural value of each component may be their average characteristic value or average structural value. That is, the descriptor of the polymer composite material may include, as an additional descriptor, a value that is the sum of the products of the calculated characteristic values or structural values of each component and the mass ratio.

Note that the cross term is effective for expressing the interaction (so-called synergistic effect) between parameters including descriptors of the polymer composite material, which are explanatory variables. Although it is not necessarily necessary to include a cross term when using a nonlinear model, it is preferable to use a cross term because it may contribute to improving accuracy when the number of data in the learning data set is small. For the descriptors of polymer composite materials that create cross terms, you can select a combination of descriptors that is considered to be effective for the physical properties of the polymer composite material, which is the target variable, or you can try all patterns and make predictions. A combination with high accuracy may be selected.

Physical properties of polymer composite materials include moldability (fluidity), rigidity, impact resistance, gloss, and the like. Note that the moldability can be evaluated by MFR of the polymer composite material. Impact resistance can be evaluated by IZOD impact strength at room temperature or low temperature.

The preprocessing unit 12 preprocesses the parameters (explanatory variables) including the obtained descriptor of the polymer composite material and the physical properties of the polymer composite material (objective variables). Note that the learning device 1 does not need to include the preprocessing unit 12 when the acquisition unit 11 acquires the parameters including the descriptor of the polymer composite material and the physical properties of the polymer composite material that have been preprocessed in advance. good.

When the polymer composite materials acquired by the acquisition unit 11 overlap, the preprocessing unit 12 calculates the overlapping values of the parameters including the descriptor of the polymer composite material and the physical properties of the polymer composite material as an average value (median value). may be calculated.

The preprocessing unit 12 may calculate an average value of characteristic values and structural values for each type of polymer composite material as a parameter including a descriptor of the polymer composite material.

The learning data set creation unit 13 extracts the preprocessed parameters including the descriptor of the polymer composite material as an explanatory variable and the physical properties of the polymer composite material as an objective variable, respectively, and creates a learning data set. Add. The learning data set creation unit 13 creates a learning data set by linking the input parameters including the descriptor of the polymer composite material with the physical properties of the polymer composite material.

Note that if the preprocessing unit 12 does not preprocess the parameters including the descriptor of the polymer composite material and the physical properties of the polymer composite material, the learning data set creation unit 13 uses the acquisition unit 11 to acquire the parameters and the physical properties of the polymer composite material. Parameters (explanatory variables) including descriptors of the polymer composite material and physical properties of the polymer composite material (objective variables) may be extracted and added to the learning data set.

The learning unit 14 performs learning by learning using a learning data set in which parameters (explanatory variables) including descriptors of polymer composite materials are associated with physical properties (objective variables) of polymer composite materials. A completed model M1 is generated.

The trained model M1 is a trained model that has undergone machine learning in advance using a learning data set (learning data table) stored in a storage unit (not shown), and is stored in the storage unit. Learning results of the correspondence between parameters (explanatory variables) including descriptors of polymer composite materials and physical properties of polymer composite materials (objective variables), obtained by performing machine learning using a training dataset. applies. The trained model M1 uses parameters including the descriptor of the polymer composite material as input data as explanatory variables, output data as the physical properties of the polymer composite material as the objective variable, and inputs the parameters including the descriptor of the polymer composite material and the polymer This is a program for modeling the input-output relationship with the physical properties of molecular composite materials. Note that the trained model M1 may be expressed by a mathematical expression such as a function.

Among machine learning, it is preferable to apply a supervised learning algorithm to the trained model M1. Examples of supervised learning include linear regression, regularized regression, partial least squares regression, polynomial regression, kernel regression, logistic regression, and random forest. ), Gradient Boosting Regression Tree, Support Vector Machine (SVM), Neural Network, etc. As the neural network, deep learning in which the neural network has more layers than three layers can be used. Examples of the types of neural networks that can be used include a convolutional neural network (CNN), a recurrent neural network (RNN), and a general regression neural network (General Regression Neural Network). Can be done.

The display unit 15 displays information on the training dataset used in learning the trained model M1, information regarding the trained model M1, and the like.

The learning device 1 outputs one characteristic of the polymer composite material using one learned model M1. For this reason, it is preferable that the learning device 1 has a learned model M1 corresponding to the physical properties of each polymer composite material for each physical property of the polymer composite material to be output.

In this way, since the learning device 1 includes the learning section 14, it is possible to generate the learned model M1 that predicts the physical properties of the polymer composite material. The trained model M1 generated by the learning device 1 predicts the physical properties of a polymer composite material from input parameters including a descriptor of the polymer composite material.

In addition, since the learning device 1 can generate a trained model M1, using the learning device 1 can reduce the burden and time required to predict the physical properties of a polymer composite material from parameters including descriptors of the polymer composite material. It is possible to reduce the amount of In other words, when manufacturing a polymer composite material, various polymers are actually used in experiments in order to manufacture a polymer composite material that satisfies arbitrary properties such as formability depending on the application of the polymer composite material. Polymer composite materials containing molecules are manufactured and their physical properties are measured. By performing these steps, the type of polymer and composition of the composition that are effective for manufacturing the polymer composite material are determined, which requires a great deal of labor and requires a lot of effort to prepare various materials. The cost burden was large. The learning device 1 according to the present embodiment is used to predict the physical properties of a polymer composite material from parameters including descriptors of the polymer composite material, thereby reducing the burden on the prediction of the physical properties of the polymer composite material. It can be done efficiently. Therefore, the learning device 1 can reduce energy consumption when predicting the physical properties of a polymer composite material.

<Prediction device>
The prediction device will be explained. FIG. 3 is a system configuration diagram showing the configuration of the prediction device. As shown in FIG. 3, the prediction device 2 includes an acquisition section 21, a preprocessing section 22, a trained model M2, a prediction section 23, and a display section 24. The prediction device 2 predicts the physical properties of a polymer composite material containing a polymer.

The acquisition unit 21 acquires parameters including descriptors of the polymer composite material to be predicted as explanatory variables. The parameters including the descriptor of the polymer composite material are the same as the parameters including the descriptor of the polymer composite material acquired by the above-described learning device 1, so the description thereof will be omitted.

The preprocessing unit 22 is similar to the preprocessing unit 12 of the learning device 1, so a description thereof will be omitted.

The trained model M2 is trained using a training data set prepared in advance in which parameters (explanatory variables) including descriptors of polymer composite materials are associated with physical properties of polymer composite materials (objective variables). It is what was done. The trained model M1 generated by the learning device 1 described above can be used as the trained model M2.

The prediction unit 23 inputs the parameters including the descriptor of the polymer composite material of the prediction target acquired by the acquisition unit 21 to the trained model M2, thereby calculating the height of the prediction target predicted by the learned model M2. Outputs the physical properties of molecular composite materials. That is, the prediction unit 23 predicts one type of physical property of the polymer composite material to be predicted, which corresponds to a parameter including a descriptor of the polymer composite material to be predicted, from one trained model M2.

Furthermore, by using the plurality of trained models M2, the prediction unit 23 predicts a group including physical properties of the polymer composite material to be predicted, which corresponds to parameters including descriptors of the polymer composite material to be predicted. It's fine. For example, the prediction unit 23 may predict predicted values of a plurality of physical properties of the polymer composite material to be predicted, corresponding to the mass ratio, characteristic values, structural values, etc. of the polymer composite material to be predicted. Here, predicted values for each physical property may be obtained from each trained model by inputting parameters that include the same descriptor, and predicted values for each physical property may be obtained from the same prediction target polymer composite material. It may also be passed through a separate pre-processing section.

Further, the prediction unit 23 may predict a group including physical properties of the polymer composite material to be predicted, corresponding to a plurality of parameters including descriptors of the polymer composite material to be predicted.

The display unit 24 displays the physical properties of the polymer composite material predicted by the trained model M2 as target variables.

In this way, the prediction device 2 includes the prediction unit 23, and the prediction unit 23 can predict the physical properties of the polymer composite material from the parameters including the descriptor of the polymer composite material using the trained model M2. Therefore, the prediction device 2 can predict the physical properties of the polymer composite material from input parameters including the descriptor of the polymer composite material.

Furthermore, by including the prediction unit 23, the prediction device 2 can reduce the burden and time required for predicting the physical properties of a polymer composite material from parameters including descriptors of the polymer composite material. The physical properties of materials can be predicted. Therefore, since the prediction device 2 can efficiently predict the physical properties of the polymer composite material, it is possible to reduce energy consumption when predicting the physical properties of the polymer composite material.

<Composition proposal device>
A composition proposal device according to this embodiment will be explained. FIG. 4 is a block diagram showing the configuration of the composition proposal device according to this embodiment. As shown in FIG. 4, the composition proposal device 3 includes an acquisition section 31, a preprocessing section 32, a prediction section 33, a learned model M3, a setting section 34, a comparison section 35, a determination section 36, an optimization section 37, and a display section. 38. The composition proposal device 3 proposes a composition of a polymer composite material formed from a composition containing at least a polymer as a constituent component.

The acquisition unit 31 acquires a parameter including a descriptor of a virtual polymer composite material as a virtually set descriptor (virtual descriptor) as an explanatory variable.

The virtual descriptor first input to the acquisition unit 31 may be a randomly acquired descriptor, for example, a randomly acquired mass ratio, characteristic value, and structure of the constituent components included in the polymer composite material. Values etc. may be used. Further, the first input virtual descriptors may include a plurality of groups corresponding to a plurality of virtual polymer composite materials.

The descriptor of the polymer composite material is the same as the descriptor of the polymer composite material acquired by the learning device 1 described above, so the description thereof will be omitted.

The preprocessing unit 32 performs preprocessing on the parameters including the descriptor of the virtual polymer composite material acquired by the acquisition unit 31. Note that the composition proposal device 3 does not need to include the preprocessing unit 32 when the obtaining unit 31 obtains parameters including a descriptor of a virtual polymer composite material that has been preprocessed in advance.

When the parameters including the descriptor of the virtual polymer composite material acquired by the acquisition unit 31 overlap, the preprocessing unit 32 determines that the overlapping values of the parameters including the descriptor of the virtual polymer composite material are averaged ( The median value) may be calculated.

The preprocessing unit 32 may calculate an average value of characteristic values and structural values for each type of polymer composite material as a parameter including a descriptor of the virtual polymer composite material.

The prediction unit 33 inputs the parameters including the descriptor of the virtual polymer composite material acquired by the preprocessing unit 32 to the learned model M3, and predicts the physical properties of the virtual polymer composite material.

The prediction unit 33 may predict a group including predicted values of physical properties of the virtual polymer composite material corresponding to parameters including a descriptor of the virtual polymer composite material. The group including the physical properties of the virtual polymer composite material is a group of predicted values of the physical properties of the virtual polymer composite material predicted by the plurality of learned models M3. For example, the prediction unit 33 predicts a group of predicted values of physical properties of the virtual polymer composite material corresponding to the mass ratio, characteristic values, structural values, etc. of the virtual polymer composite material, using the plurality of trained models M3. You may do so.

The trained model M3 is a trained model that is trained using a training data set in which parameters including a descriptor of a polymer composite material are associated with physical properties of the polymer composite material, and is different from the trained model M1. Since they are similar, details will be omitted.

The setting unit 34 sets target values of physical properties of the virtual polymer composite material.

The target value is appropriately set depending on the use of the polymer composite material, etc.

The comparison unit 35 compares the physical properties of the virtual polymer composite material predicted by the prediction unit 33 with target values of the physical properties of the virtual polymer composite material.

The determining unit 36 determines whether it is necessary to optimize parameters including the descriptor of the virtual polymer composite material.

The determination unit 36 determines whether or not iterative optimization has been performed a predetermined number of times.

The predetermined number of times may be set as appropriate depending on the type, composition, etc. of the polymer composite material used.

Further, the determination unit 36 simultaneously determines whether the difference between the physical properties of the virtual polymer composite material predicted by the prediction unit 33 and the target value of the physical properties of the virtual polymer composite material is within a predetermined range. You may judge.

Note that the predetermined range may be set as appropriate depending on the type, composition, etc. of the polymer composite material used. For example, the predetermined range may be a range in which the physical properties of the virtual polymer composite material are the same as the target values, or a range in which the error between the physical properties of the virtual polymer composite material and the target value is several percent or less. Furthermore, when there are a plurality of physical properties of the virtual polymer composite material, the error between at least one of the physical properties of the virtual polymer composite material and the target value may be within a range of several percent or less.

The optimization unit 37 uses descriptors (for example, mass ratio, property values, and structure) of the virtual polymer composite material so that the physical properties of the virtual polymer composite material predicted by the prediction unit 33 approach the target values. Optimize by changing parameters, including values (values, etc.). The optimization unit 37 includes the descriptor of the virtual polymer composite material by changing the parameters including the descriptor of the virtual polymer composite material every time the physical properties of the virtual polymer composite material are predicted. Optimize parameters.

That is, the optimization unit 37 optimizes the parameters including the descriptor of the virtual polymer composite material so that the physical properties of the virtual polymer composite material change in a desirable direction.

As described above, the constituent components include a polymer as an essential component, and may include a component other than the polymer as an optional component, and may include one or more of both the polymer and the component other than the polymer. As a descriptor of a polymer composite material, for example, when optimizing the types of constituent components contained in a polymer composite material and the mass ratio of the constituent components, propylene-based polymers and It is preferable to optimize the type of polyolefin elastomer and the mass ratio of the propylene polymer, polyolefin elastomer, and other constituent components.

The optimization unit 37 inputs the optimized descriptor of the virtual polymer composite material to the acquisition unit 31 again as a virtual descriptor.

The optimization unit 37 may use, for example, a library included in Anaconda (registered trademark), which is software distributed by Anaconda, Inc. in the United States. Anaconda (registered trademark) includes a group of libraries used in Python (registered trademark) and machine learning. The optimization unit 37 uses Optuna (registered trademark), a Python library, to describe a virtual polymer composite material with good physical properties in order to optimize parameters including descriptors of the virtual polymer composite material. Can suggest parameters that include children.

That is, the optimization unit 37 determines a desirable mass ratio of the virtual polymer composite material upon optimization from parameters including descriptors of the virtual polymer composite material and predicted values of the physical properties of the virtual polymer composite material. , characteristic values, structure values, etc., can be proposed using the Python library Optuna (registered trademark).

The optimization unit 37 is not limited to changing a parameter including a descriptor of a virtual polymer composite material to a parameter including a descriptor of a polymer composite material recorded in the learning data set. The optimization unit 37 may change the parameters to include a descriptor of the polymer composite material that is not recorded in the learning data set.

That is, the optimization unit 37 converts the descriptor of the virtual polymer composite material into polymers recorded in the learning data set based on the physical properties of the virtual polymer composite material predicted by the prediction unit 33. It may be changed to the mass ratio, characteristic value, structural value, etc. of the composite material, or it may be changed to the mass ratio, characteristic value, structural value, etc. of the polymer composite material that are not recorded in the learning data set.

The optimization unit 37 also sets parameters including a descriptor of a specific polymer composite material, regardless of whether the parameters include a descriptor of a virtual polymer composite material or not, regardless of whether the parameters include a descriptor of a specific polymer composite material. You can change it to avoid it.

That is, the optimization unit 37 uses specific mass ratios, characteristic values, structural values, etc. of the polymer composite material as descriptors of the specific polymer composite material, regardless of whether or not they are recorded in the learning data set. It may be changed to avoid this.

The optimization unit 37 generates a descriptor of a virtual polymer composite material in order to realize parameters including a descriptor including a desirable mass ratio, characteristic values, structural values, etc. of the virtual polymer composite material at the time of optimization. It is preferable to optimize and propose using an algorithm.

Examples of algorithms include random search, mathematical optimization processing, etc., but among them, mathematical optimization processing is preferable because it performs optimization more efficiently.

As the mathematical optimization process, a genetic algorithm, base optimization, TPE (Tree-structured Parzen Estimator), etc. can be used. Among these, the genetic algorithm is preferable from the viewpoint of the balance between the calculation speed per processing and the efficiency of optimizing the predicted values of the properties of the polymer composite material.

The optimization unit 37 optimizes the parameters including the descriptor of the virtual polymer composite material from the predicted values of the physical properties of the virtual polymer composite material predicted by the prediction unit 33. A group of compositions (Pareto-optimal solutions) of polymer composite materials may be proposed. That is, the optimization unit 37 may propose a plurality of composition combinations of the virtual polymer composite material based on the physical properties of the virtual polymer composite material predicted by the prediction unit 33.

Even when the polymer composite material has two or more physical properties, the optimization unit 37 may propose a group of compositions (Pareto optimal solutions) of the polymer composite material for the two or more physical properties.

The following method is available as a method for optimizing parameters including descriptors of a virtual polymer composite material.

The first optimization method is to optimize parameters including descriptors of a virtual polymer composite material so as to maximize or minimize the physical properties of the virtual polymer composite material in the desired direction. There are methods that propose a set of descriptors for composite materials.

This optimization method preferably optimizes parameters including descriptors of the virtual polymer composite material so as to maximize the physical properties of the virtual polymer composite material in a desired direction.

In this optimization method, the optimization unit 37 specifies the physical properties of the polymer composite material to be optimized and the direction of optimization (maximization or minimization). The descriptors of the polymer composite material can be optimized so that the physical properties of the polymer composite material are good.

For example, the optimization unit 37 predicts the physical properties of the polymer composite material corresponding to the composition R from the proposed composition R of the polymer composite material via the acquisition unit 31, the preprocessing unit 32, and the prediction unit 33. Obtain a group of values P. At this time, the optimization unit 37 repeats the above operation to determine the virtual composition Rj (j=1, 2...n) of the polymer composite material and the polymer corresponding to the composition of the polymer composite material. A pair with a group Pj (j=1, 2...n) of predicted values of physical properties of the composite material is obtained. Then, the optimization unit 37 calculates the physical properties of the polymer composite material from the pair of the obtained virtual composition Rj of the polymer composite material and the group Pj of predicted values of the physical properties of the polymer composite material corresponding to the composition. It is possible to select a group of compositions of polymer composite materials for which the predicted value of is good.

For example, a genetic algorithm is used as the mathematical optimization process. In this case, the acquisition unit 31 first (randomly) inputs a plurality of virtual compositions Rj (j=1, 2...n) and a group of predicted values of the corresponding physical properties Pj (j=1 , 2...n) as the first generation. Then, from a subset of the virtual compositions Rj in which the values of the group Pj of predicted values are desirable (that is, maximized or minimized in a desirable direction), parameters including descriptors of the polymer composite material are rearranged among the virtual compositions. A crossover operation and a mutation operation are performed to randomly replace some of the parameters including the descriptor of the polymer composite material. Thereby, the optimization unit 37 obtains a group of compositions of polymer composite materials that provide good predicted values of physical properties of the polymer composite material. The above operation is repeated using the pair of the composition group obtained in this way and the corresponding predicted value group as the second generation. As a result, a group of compositions of polymer composite materials with better predicted values of physical properties of the polymer composite material can be obtained.

Using the above method, the optimization unit 37 can propose a group of descriptors of a polymer composite material in which the physical properties of the polymer composite material are maximized or minimized in a desired direction as a group of virtual descriptors.

Therefore, the optimization unit 37 uses the above method to describe a polymer composite material in which the physical properties of the polymer composite material are maximized or minimized in a desirable direction among the polymer composite materials for which many combinations are possible. Can suggest groups of children.

The second optimization method is to propose a group of descriptors for a virtual polymer composite material so that the physical properties of the virtual polymer composite material approach desired values.

In this optimization method, the optimization unit 37 specifies physical property items of the polymer composite material to be optimized and their target values as objective functions for the optimization. The composition of the polymer composite material can be optimized so that the physical properties of the composite material are good.

When the target value of the physical property k of the polymer composite material is Ok and the predicted value of the physical property of the polymer composite material corresponding to the composition Rj of the polymer composite material is Pk,j, the optimization unit 37 calculates the following values for optimization: The objective function fk of is determined by the following equation (1).
fk=|Ok-Pk,j|/|Ok| ...(1)

Then, the optimization unit 37 optimizes the composition so as to minimize the objective function fk using a method similar to the first optimization method described above.

Using the method described above, the optimization unit 37 can propose a group of descriptors for a polymer composite material that brings the physical properties of the polymer composite material closer to desired values (target values) as a group of virtual descriptors.

Even when the polymer composite material has two or more physical properties, the optimization unit 37 uses the method of optimizing parameters including the descriptor of the virtual polymer composite material in the same way as described above. We can propose a group of descriptors.

That is, the optimization unit 37 generates virtual properties predicted by inputting parameters including descriptors of the virtual polymer composite material into the trained model M3 for each of two or more physical properties of the polymer composite material. Parameters including descriptors of a virtual polymer composite material are optimized so that two or more physical properties of the polymer composite material are maximized or minimized, respectively, and a group of descriptors of the polymer composite material is You may make suggestions.

The optimization unit 37 also generates virtual properties predicted by inputting parameters including descriptors of the virtual polymer composite material into the learned model M3 for each of the two or more physical properties of the polymer composite material. Parameters including descriptors of the virtual polymer composite material are optimized so that two or more physical properties of the polymer composite material approach the respective target values of the two or more physical properties, and the parameters of the polymer composite material are optimized. A group of descriptors may be proposed.

The display unit 38 displays information regarding parameters including descriptors of the virtual polymer composite material. The descriptor of the polymer composite material may be a descriptor of the polymer composite material that has been proposed by the optimization unit 37 and re-determined by the determination unit 36 one or more times, or may be a descriptor of the polymer composite material that has been proposed by the optimization unit 37 one or more times and determined again by the determination unit 36. It may also be a descriptor for polymer composite materials. In the determination unit 36, if the physical properties of the polymer composite material predicted by the prediction unit 33 are within a predetermined range of target values, the descriptor of the polymer composite material acquired by the acquisition unit 31 is first displayed. You may do so.

The display unit 38 displays, for example, a group of descriptors for polymer composite materials proposed by the optimization unit 37 in a list of descriptors for polymer composite materials, such as a map, a ranking format, an optimal combination order, etc. It's fine.

In this way, the composition proposal device 3 includes a prediction section 33, a setting section 34, a comparison section 35, a determination section 36, and an optimization section 37. The composition proposal device 3 compares the physical properties of the virtual polymer composite material predicted by the prediction unit 33 with the target values set by the setting unit 34 in the comparison unit 35, and determines the properties of the virtual polymer composite material in the determination unit 36. It is determined whether the physical properties of the composite material are within a predetermined range of target values. In the composition proposal device 3, the physical properties of the virtual polymer composite material predicted by inputting parameters including descriptors of the virtual polymer composite material into the trained model M3 are set to target values in the optimization unit 37. Optimize the parameters including the descriptor of the virtual polymer composite material so that it approaches .

The composition proposal device 3 causes the optimization unit 37 to evaluate the virtual polymer composite material in the optimization unit 37 until the determination unit 36 determines that the physical properties of the virtual polymer composite material obtained by the prediction unit 33 are within a predetermined range of target values. The operation of optimizing parameters including descriptors of the virtual polymer composite material can be repeated so that the physical properties of the polymer composite material approach target values.

Thereby, the composition proposal device 3 can propose parameters including a descriptor of the polymer composite material from the physical properties of the virtual polymer composite material. Therefore, the composition proposal device 3 can propose parameters including a descriptor of the polymer composite material that can be set and obtain desired properties of the polymer composite material.

Furthermore, by including the optimization unit 37, the composition proposal device 3 can reduce the burden required when optimizing the descriptor of a virtual polymer composite material so as to satisfy the desired properties of the virtual polymer composite material. It is possible to reduce the amount of damage and time required. Therefore, the composition proposal device 3 can efficiently propose parameters including descriptors of polymer composite materials that satisfy desired characteristics, thereby reducing energy consumption required for selecting polymer composite materials.

(Other aspects of composition proposal device)
Other aspects of the composition proposal device according to this embodiment will be described. FIG. 5 is a block diagram showing another configuration of the composition proposal device according to this embodiment. As shown in FIG. 5, the composition proposal device 4 is a device that proposes the composition of a polymer composite material without using the trained model M3 in the composition proposal device 3 shown in FIG. That is, the composition proposal device 4 includes an acquisition section 41 , a preprocessing section 42 , a setting section 43 , a comparison section 44 , a determination section 45 , an optimization section 46 , and a display section 47 .

The acquisition unit 41 acquires a parameter including a descriptor of a virtual polymer composite material as a virtually set descriptor (virtual descriptor) as an explanatory variable. The descriptor of the polymer composite material is the same as the descriptor of the polymer composite material acquired by the acquisition unit 31 of the composition proposal device 3 described above, so the details will be omitted.

The preprocessing unit 42 performs preprocessing on the parameters including the virtual descriptor acquired by the acquisition unit 41. Note that the composition proposal device 3 does not need to include the preprocessing unit 42 when the obtaining unit 41 obtains parameters including a descriptor of a virtual polymer composite material that has been preprocessed in advance. The preprocessing unit 42 is similar to the preprocessing unit 42 of the composition proposal device 3 described above, so the details will be omitted.

The setting unit 43 sets target values of parameters including descriptors of the virtual polymer composite material.

The target value is appropriately set depending on the use of the polymer composite material, etc.

The comparison unit 44 compares the parameter including the descriptor of the virtual polymer composite material acquired by the preprocessing unit 32 with the target value of the parameter including the descriptor of the polymer composite material set by the setting unit 43. do.

The determination unit 45 determines whether the difference between the parameter including the descriptor of the virtual polymer composite material acquired by the preprocessing unit 32 and the target value set by the setting unit 43 is within a predetermined range. . The determination method can be performed in the same manner as the determination method in the determination unit 36 of the composition proposal device 3 described above.

The optimization unit 46 uses parameters including descriptors (for example, mass ratio, characteristic values, structural values, etc.) of the virtual polymer composite material acquired by the preprocessing unit 32 and target values set by the setting unit 43. From this, parameters including the descriptor of the virtual polymer composite material are changed and optimized so that the descriptor of the virtual polymer composite material approaches the target value.

The optimization unit 46 calculates a descriptor for the virtual polymer composite material from the parameters including the descriptor for the virtual polymer composite material obtained by the preprocessing unit 32 and the target value set by the setting unit 43. When the included parameters are optimized, a group of optimized virtual polymer composite material compositions (Pareto optimal solutions) may be proposed.

The optimization method can be performed in the same way as the optimization method performed by the optimization unit 37 of the composition proposal device 3 described above, so the details will be omitted.

The optimization unit 46 inputs the optimized descriptor of the virtual polymer composite material to the acquisition unit 41 again as a virtual descriptor.

The display unit 47 displays information regarding parameters including virtual descriptors of the polymer composite material. The descriptor of a polymer composite material may be a descriptor of a polymer composite material that has been proposed by the optimization unit 46 and determined again by the determination unit 45 one or more times, or may be a descriptor of a polymer composite material that has been proposed by the optimization unit 46 one or more times and determined again by the determination unit 45. It may also be a descriptor for polymer composite materials. In the determination unit 45, if the physical properties of the polymer composite material compared by the comparison unit 44 are within the predetermined range of the target value, the descriptor of the virtual polymer composite material first acquired by the acquisition unit 41 is determined. may be displayed. The display section 47 may display information in the same manner as the display section 38 of the composition proposal device 3 described above.

In this way, the composition proposal device 4 includes the setting section 43 , the setting section 43 , the determining section 45 , and the optimizing section 46 . In the composition proposal device 4, the setting unit 43 compares the descriptor of the virtual polymer composite material with the target value set by the setting unit 43, and the determining unit 45 determines whether the descriptor of the virtual polymer composite material is It is determined whether the target value is within a predetermined range. The composition proposal device 4 optimizes the parameters including the descriptor of the virtual polymer composite material in the optimization unit 46 so that the descriptor of the virtual polymer composite material approaches the target value. Optimize parameters including descriptors of virtual polymer composites.

The composition proposal device 4 causes the optimization unit 46 to modify the description of the virtual polymer composite material until the determination unit 45 determines that the descriptor of the virtual polymer composite material is within a predetermined range of the target value. The operation of optimizing the parameters including the descriptor of the virtual polymer composite material can be repeated so that the child approaches the target value. Thereby, the composition proposal device 4 can propose parameters including a descriptor of the virtual polymer composite material.

If the parameters including the descriptor of the polymer composite material are approximately equivalent to the set target value, the virtual polymer composite material has physical properties that are approximately equivalent to the polymer composite material set as the target value. Can be done. Therefore, the composition proposal device 4 can propose parameters including a descriptor of a virtual polymer composite material having desired physical properties.

Further, by including the optimization unit 46, the composition proposal device 4 can reduce the burden and time required to optimize parameters including descriptors of a virtual polymer composite material. Therefore, the composition proposal device 4 can efficiently propose parameters including a descriptor of a polymer composite material that is approximately equivalent to a parameter including a descriptor of a polymer composite material that satisfies the desired characteristics, so that the selection of a polymer composite material is facilitated. The energy consumption required for this can be reduced.

Note that the learning device 1, prediction device 2, and composition proposal devices 3 and 4 described above may be configured as a learning system, a prediction system, and a composition proposal system. That is, the learning device 1 is a single device such as a PC (Personal Computer) that has each component inside the device, but one or more of each component is placed outside the device and connected via a network. It's okay.

For example, the training data set may be provided on the cloud. In this case, the learning device 1 is configured as a learning system by a learning data set connected via a network.

Similarly, in the prediction device 2, one or more of the components may be placed outside the device and connected via a network.

Similarly, in the composition proposing devices 3 and 4, one or more of each component may be placed outside the devices and connected via a network. For example, the composition proposal data set may be provided on the cloud. In this case, the composition proposal device 3 is configured as a composition proposal system by a composition proposal data set connected via a network.

<Hardware configuration of composition proposal device>
Next, an example of the hardware configuration of the learning device 1, the prediction device 2, and the composition proposal devices 3 and 4 will be described. FIG. 5 is a block diagram showing the hardware configuration of the learning device 1, the prediction device 2, and the composition proposal devices 3 and 4. As shown in FIG. 6, the learning device 1, the prediction device 2, and the composition proposal devices 3 and 4 are composed of information processing devices (computers), and physically include a CPU (Central Processing Unit) that is an arithmetic processing unit. ) 101, a RAM (Random Access Memory) 102 and a ROM (Read Only Memory) 103 which are main storage devices, an input device 104 which is an input device, an output device 105, a communication module 106, an auxiliary storage device 107 such as a hard disk, etc. It can be configured as a computer system. These are interconnected by a bus 108. Note that the output device 105 and the auxiliary storage device 107 may be provided externally.

The CPU 101 controls the overall operation of the learning device 1, the prediction device 2, and the composition proposal devices 3 and 4, and performs various information processing. The CPU 101 performs learning, prediction, and composition proposal by executing, for example, a learning method, a prediction method, and a composition proposal method, or a learning program, a prediction program, and a composition proposal program, which will be described later, stored in the ROM 103 or the auxiliary storage device 107. be able to.

The RAM 102 is used as a work area for the CPU 101 and may include a nonvolatile RAM that stores main control parameters and information.

The ROM 103 stores basic input/output programs and the like. The catalyst selection program may be stored in ROM 103.

The input device 104 is an input device such as a keyboard, a mouse, an operation button, a touch panel, a display screen, etc., receives information input by a user as an instruction signal, and outputs the instruction signal to the CPU 101.

The output device 105 is a display device such as a monitor display, a speaker, a printing device such as a printer, or the like. In the output device 105, information such as catalyst selection results is displayed on a display device such as a monitor display, and the displayed screen is updated in response to an input operation via the input device 104 or the communication module 106.

The communication module 106 is a data transmitting/receiving device such as a network card, and functions as a communication interface that takes in information from an external data recording server or the like and outputs analysis information to other electronic devices.

The auxiliary storage device 107 is a storage device such as an SSD (Solid State Drive) and an HDD (Hard Disk Drive). Store data, files, etc.

Each function of the learning device 1, the prediction device 2, and the composition proposal devices 3 and 4 is executed by the CPU 101 by loading predetermined computer software (including a catalyst selection program) from the main storage device such as the RAM 102 or the auxiliary storage device 107. This is realized by reading and writing data in the main memory device such as the RAM 102 or the auxiliary memory device 107, and operating the input device 104, output device 105, and communication module 106.

Therefore, each part of the learning device 1, the prediction device 2, and the composition proposal devices 3 and 4 shown in FIGS. By executing stored predetermined computer software (including a learning program, a prediction program, and a composition proposal program), software and hardware are realized in cooperation.

The learning program, prediction program, and composition proposal program can be stored, for example, in the main storage device or auxiliary storage device 107 included in the computer. Furthermore, the learning program, prediction program, and composition proposal program are stored on a computer connected to a communication line such as the Internet, and provided by downloading part or all of the catalyst selection program via the communication line. Good too. Furthermore, the learning program, the prediction program, and the composition proposal program may be configured to be provided or distributed via a communication line.

The learning program, the prediction program, and the composition proposal program are stored in part or in whole in a portable storage medium such as an optical disk such as a CD-ROM or a DVD-ROM, or a semiconductor memory such as a flash memory, and are then stored in a computer. It may be recorded (including installation) within.

<Learning method>
Next, the learning method will be explained. The learning method is a learning method in which parameters (explanatory variables) including descriptors of polymer composite materials are associated with physical properties of polymer composite materials (objective variables) in a learning device 1 having a configuration as shown in FIG. This is a method of generating a trained model that predicts the physical properties of a polymer composite material containing polymers using a dataset of

FIG. 7 is a flowchart explaining the learning method. As shown in FIG. 7, the acquisition unit 11 acquires parameters including a descriptor of the polymer composite material as explanatory variables, and acquires the physical properties of the polymer composite material as a target variable (acquisition step: step S11).

Next, the preprocessing unit 12 preprocesses the parameters including the obtained descriptor of the polymer composite material and the physical properties of the polymer composite material (preprocessing step: step S12).

Next, the learning data set creation unit 13 extracts the preprocessed parameters including the descriptor of the polymer composite material as an explanatory variable and the physical properties of the polymer composite material as an objective variable. (Learning dataset creation step: Step S13).

The learning data set creation unit 13 creates a learning data set by linking the input parameters including the descriptor of the polymer composite material with the physical properties of the polymer composite material.

Next, the learning unit 14 generates a trained model M1 using a learning data set in which parameters including descriptors of the polymer composite material are associated with physical properties of the polymer composite material (learning step: Step S14).

The learning unit 14 determines the physical properties of the polymer composite material linked to the parameters including the descriptor of the polymer composite material, in response to the input of the parameters including the descriptor of the polymer composite material included in the learning data set. The trained model M1 is generated so that the output matches the following.

Next, the display unit 15 displays information on learning data used in learning the trained model M1 and information regarding the trained model M1 (display step: step S15).

By including the learning step (step S13), the learning method can generate a trained model M1 that predicts the physical properties of the polymer composite material. Therefore, the learning method according to the present embodiment uses the trained model M1 generated in the learning process (step S13) to determine the polymer composite material from the input parameters including the polymer composite descriptor. It can be trained to predict physical properties.

Further, the learning method includes a learning step (step S13), and a learned model M1 can be generated in the learning step (step S13). Therefore, if the learning method according to the present embodiment is used, the parameters including the descriptors of the polymer composite material can be used to predict the physical properties of the polymer composite material, and the learning method required for predicting the physical properties of the polymer composite material can be It is possible to reduce the burden and time. Therefore, by using the learning method according to this embodiment, it is possible to reduce energy consumption when predicting the physical properties of a polymer composite material.

<Prediction method>
Next, a prediction method will be explained. The prediction method is a method in which the physical properties of the polymer composite material to be predicted are predicted using the learned model M2 in the prediction device 2 having the configuration shown in FIG.

FIG. 8 is a flowchart illustrating the prediction method. As shown in FIG. 8, the acquisition unit 21 acquires parameters including descriptors of the polymer composite material to be predicted as explanatory variables (acquisition step: step S21).

Next, the preprocessing unit 12 preprocesses the parameters including the obtained descriptor of the polymer composite material (preprocessing step: step S22).

Next, the prediction unit 23 inputs the preprocessed parameters including the descriptor of the polymer composite material to be predicted to the trained model M2, thereby obtaining the predicted value of the predicted target by the trained model M2. The physical properties of the polymer composite material are output (prediction step: step S24).

Next, the display unit 24 displays information regarding the physical properties of the polymer composite material predicted by the trained model M2 in the prediction step S23 (display step: step S24).

The prediction method includes a prediction step (step S24), so that the learned model M2 can predict the physical properties of the polymer composite material as a target variable from parameters including the descriptor of the polymer composite material. Therefore, the prediction method according to the present embodiment can predict the physical properties of a polymer composite material from input parameters including a descriptor of the polymer composite material.

Furthermore, by including the prediction step (step S24), the prediction method reduces the burden and time required for predicting the physical properties of a polymer composite material from parameters including descriptors of the polymer composite material. The physical properties of molecular composite materials can be predicted. Therefore, since the learning method according to the present embodiment can efficiently predict the physical properties of a polymer composite material, it is possible to reduce energy consumption when predicting the physical properties of a polymer composite material.

<Composition proposal method>
Next, a composition proposal method according to this embodiment will be explained. The composition proposal method according to this embodiment is a method of proposing a composition of a polymer composite material using a trained model M3 in a composition proposal device 3 having a configuration as shown in FIG.

FIG. 9 is a flowchart illustrating the composition proposal method according to this embodiment. As shown in FIG. 9, the acquisition unit 31 acquires parameters including a descriptor of a virtual polymer composite material as a virtually set descriptor (virtual descriptor) as an explanatory variable (acquisition step: step S31 ).

For a polymer composite material containing a heterophasic propylene polymer material, a heterophasic propylene polymer material, a polyolefin elastomer, a filler (for example, talc, etc.), etc. are each randomly selected from a list of usable raw materials, and The mass ratio of the constituent components contained in the polymer composite material is randomly set. As a result, a descriptor (eg, virtual mass ratio, characteristic value, structural value, etc.) R1 of the virtual polymer composite material is obtained. However, when the descriptor includes a mass ratio, the total amount of each component in the mass ratio may be 100 parts by mass. At this time, each component may be of one type or may be of multiple types.

Next, the preprocessing unit 32 preprocesses the parameters including the descriptor of the virtual polymer composite material acquired in the acquisition process (step S31) (preprocessing process: step S32). Note that the pretreatment step (step S32) may be performed as necessary.

Next, the prediction unit 33 inputs the parameters including the descriptor of the virtual polymer composite material preprocessed in the preprocessing step (step S32) to the learned model M3, and The physical properties of are predicted as objective variables (prediction step: step S33).

In the prediction step (step S33), a group P1 of predicted values of physical properties of the polymer composite material corresponding to parameters including the descriptor R1 of the virtual polymer composite material is predicted using the plurality of trained models M3. It's fine.

Next, the setting unit 34 sets target values for the physical properties of the virtual polymer composite material (target value setting step for physical properties: step S34).

As mentioned above, the target value is appropriately set depending on the use of the polymer composite material, etc.

Next, the comparison unit 35 compares the physical properties of the virtual polymer composite material predicted in the prediction step (step S33) with the physical properties of the virtual polymer composite material set in the physical property target value setting step (step S34). The physical properties are compared with target values (comparison step: step S35).

Next, the determination unit 36 determines whether optimization of the parameters including the descriptor of the virtual polymer composite material acquired in the acquisition step (step S31) has been performed a predetermined number of times (first determination step :Step S36).

If optimization of the parameters including the descriptor of the virtual polymer composite material has been performed a predetermined number of times (step S36: Yes), the determination unit 36 changes the parameters including the descriptor of the virtual polymer composite material. Decide that it is not necessary.

Next, the determination unit 36 determines whether the difference between the physical properties of the virtual polymer composite material and the target value is within a predetermined range (second determination step: step S37). Note that in the composition proposal method according to the present embodiment shown in FIG. Step S37) may be omitted.

If the difference between the physical properties of the virtual polymer composite material and the target value is within a predetermined range (step S37: Yes), the determination unit 36 determines that it is necessary to change the parameters including the descriptor of the virtual polymer composite material. It is determined that there is no.

The parameters including the descriptor of the virtual polymer composite material acquired in the acquisition step (step S31) are not changed and are maintained.

Next, the display unit 38 displays information regarding parameters including descriptors of the virtual polymer composite material (display step: step S38).

On the other hand, if the optimization of the parameters including the descriptor of the virtual polymer composite material has not been performed the predetermined number of times (step S36: No), or if the physical properties of the virtual polymer composite material differ from the target values by a predetermined value. If it is not within the range (step S37: No), the determination unit 36 determines that it is necessary to change the parameter including the descriptor of the virtual polymer composite material.

Next, the optimization unit 37 operates so that the physical properties of the virtual polymer composite material predicted in the prediction step (step S32) approach the target values set in the physical property target value setting step (step S33). Parameters including descriptors (for example, mass ratio, characteristic values, structural values, etc.) of the virtual polymer composite material are changed and optimized to become optimal parameters (optimization step: step S39).

In the optimization process (step S39), the optimization unit 37 uses the descriptor of the virtual polymer composite material acquired in the acquisition process (step S31) and the virtual height predicted in the prediction process (step S33). Optimize the parameters including descriptors of the virtual polymer composite material so that the physical properties of the virtual polymer composite material change in the desired direction (closer to the target value) based on the predicted values of the physical properties of the molecular composite material. do.

In the optimization process (step S39), a description of the virtual polymer composite material is created in order to realize parameters including desirable descriptors (within a predetermined range from the target value) of the virtual polymer composite material when optimized. Preferably, the children are optimized using algorithms such as random search and mathematical optimization techniques.

In the optimization process (step S39), parameters including descriptors of the virtual polymer composite material are optimized from the predicted values of the physical properties of the virtual polymer composite material predicted in the prediction process (step S33). , a group of optimized virtual polymer composite compositions (Pareto-optimal solutions) may be proposed.

Even when the polymer composite material has two or more physical properties, the optimization unit 37 may propose a group of compositions (Pareto optimal solution) for the polymer composite material with respect to the two or more physical properties.

In the optimization step (step S39), the optimization unit 37 maximizes or minimizes the physical properties of the virtual polymer composite material in a desired direction as a method of optimizing parameters including descriptors of the virtual polymer composite material. A method may be used in which parameters including descriptors of a virtual polymer composite material are optimized and a group of descriptors of the polymer composite material is proposed so that the descriptors of the polymer composite material are optimized. Furthermore, as another optimization method, a method of proposing a group of descriptors for a polymer composite material may be used so that the physical properties of a virtual polymer composite material approach desirable values.

In the optimization step (step S39), the optimization unit 37 optimizes the parameters including the descriptor of the polymer composite material, and then the process moves to the acquisition step (step S31) again. Then, the acquisition unit 31 inputs parameters including the optimized descriptor of the virtual polymer composite material as a virtual descriptor to the acquisition unit 31 as explanatory variables, and performs the same process as described above.

Then, in the first determination step (step S36), optimization of parameters including descriptors of the virtual polymer composite material is performed a predetermined number of times, and in the second determination step (step S37), optimization of the parameters including the descriptor of the virtual polymer composite material is performed. In the optimization step (step S39), optimization of the descriptor of the virtual polymer composite material is repeated until it is determined that the difference between the parameter including the descriptor and the target value is within a predetermined range.

Finally, in the first determination step (step S36), optimization of the parameters including the descriptor of the virtual polymer composite material is performed a predetermined number of times, and in the second determination step (step S37), the optimization of the parameters including the descriptor of the virtual polymer composite material is After it is determined that the difference between the parameter including the descriptor of the composite material and the target value is within a predetermined range, the display unit 38 displays the proposed virtual height in the optimization process (step S39), as described above. Information regarding parameters including a descriptor of the molecular composite material is displayed (display step: step S38).

The composition proposal method according to the present embodiment includes a prediction step (step S33), a characteristic target value setting step (step S34), a comparison step (step S35), a first determination step (step S35), and a second determination step (step S37) and an optimization step (step S39). In the composition proposal method, in the first determination step (step S36), it is determined whether or not the parameters including the descriptor of the virtual polymer composite material acquired in the acquisition step (step S31) have been optimized a predetermined number of times. judge. In the composition proposal method, in the optimization step (step S39), the physical properties of the virtual polymer composite material predicted by inputting parameters including the descriptor of the virtual polymer composite material into the trained model M3 are as follows. Parameters including descriptors of the virtual polymer composite are optimized to approach the target values.

In the composition proposal method, in the first determination step (step S35), it is determined that optimization of parameters including the descriptor of the virtual polymer composite material obtained in the prediction step (step S32) has been performed a predetermined number of times. In the optimization step (step S39), the operation of optimizing the parameters including the descriptor of the virtual polymer composite material can be repeated so that the physical properties of the virtual polymer composite material approach the target values. .

In addition, in the composition proposal method, in the comparison step (step S35), the physical properties of the virtual polymer composite material predicted in the prediction step (step S32) are compared to the target values set in the characteristic target value setting step (step S33). In comparison, in the second determination step (step S37), it can be determined whether the physical properties of the virtual polymer composite material are within a predetermined range of target values. In the composition proposal method, in the optimization step (step S39), the physical properties of the virtual polymer composite material predicted by inputting parameters including the descriptor of the virtual polymer composite material into the trained model M3 are as follows. Parameters including descriptors of the virtual polymer composite are optimized to approach the target values.

In the composition proposal method, in the second determination step (step S37), the optimal In the step S39, an operation for optimizing parameters including descriptors of the virtual polymer composite material can be repeated so that the physical properties of the virtual polymer composite material approach target values.

Thereby, the composition proposal method according to the present embodiment can propose parameters including a descriptor of the polymer composite material from the physical properties of the virtual polymer composite material. Therefore, the composition proposal method according to the present embodiment can propose parameters including a descriptor of the polymer composite material that can be set and obtain the desired properties of the polymer composite material.

In addition, the composition proposal method according to the present embodiment includes an optimization step (step S39), so that the parameters including the descriptor of the virtual polymer composite material are adjusted to satisfy the desired characteristics of the virtual polymer composite material. It is possible to reduce the burden and time required when changing to Therefore, the composition proposal method according to the present embodiment can efficiently propose parameters including descriptors of polymer composite materials that satisfy desired characteristics, thereby reducing energy consumption required for selecting polymer composite materials. can.

Next, another example of the composition proposal method according to this embodiment will be described. The composition proposal method according to this embodiment is a method for proposing the composition of a polymer composite material without using the learned model M3 in the composition proposal device 4 having the configuration shown in FIG.

FIG. 10 is another flowchart illustrating the composition proposal method according to this embodiment. As shown in FIG. 10, the acquisition unit 41 acquires parameters that include a virtual descriptor of a polymer composite material as a virtually set descriptor (virtual descriptor) as an explanatory variable (acquisition process: step S41). The acquisition step (step S41) is similar to the acquisition step (step S31) of the composition proposal method shown in FIG.

Next, the preprocessing unit 42 preprocesses the parameters including the virtual descriptor of the polymer composite material obtained in the acquisition step (step S41) (preprocessing step: step S42). The pretreatment step (step S42) is similar to the pretreatment step (step S32) of the composition proposal method shown in FIG. Note that the pretreatment step (step S42) may be performed as necessary.

Next, the setting unit 43 sets a virtual target value of a parameter including a descriptor of the polymer composite material (parameter target value setting step: step S43).

Next, the setting unit 43 converts the parameters including the descriptor of the virtual polymer composite material preprocessed in the preprocessing step (step S42) into the virtual is compared with the target value of the parameter including the descriptor of the polymer composite material (comparison step: step S44).

Next, the determination unit 45 determines whether optimization of the parameters including the descriptor of the virtual polymer composite material acquired in the acquisition step (step S41) has been performed a predetermined number of times (first determination step :Step S45).

If optimization of the parameters including the descriptor of the virtual polymer composite material has been performed a predetermined number of times (step S45: Yes), the determination unit 45 changes the parameters including the descriptor of the virtual polymer composite material. Decide that it is not necessary.

Next, the determination unit 45 determines whether the difference between the parameter including the descriptor of the virtual polymer composite material and the target value is within a predetermined range (second determination step: step S46). Note that in the composition proposal method according to the present embodiment shown in FIG. Step S46) may be omitted.

If the difference between the parameter including the descriptor of the virtual polymer composite material and the target value of the parameter is within the predetermined range (step S46: Yes), the determination unit 45 determines the descriptor of the virtual polymer composite material. Determine that there is no need to change the included parameters.

The parameters including the descriptor of the virtual polymer composite material acquired in the acquisition step (step S41) are not changed and are maintained.

Next, the display unit 47 displays information regarding parameters including virtual descriptors of the polymer composite material (display step: step S47).

On the other hand, if the parameter including the descriptor of the virtual polymer composite material has not been optimized a predetermined number of times (step S45: No), or the parameter including the descriptor of the virtual polymer composite material and the target value If the difference is not within the predetermined range (step S46: No), the determination unit 45 determines that it is necessary to change the parameters including the virtual descriptor of the polymer composite material.

Next, the optimization unit 46 converts the parameters including the descriptor of the virtual polymer composite material preprocessed in the preprocessing step (step S42) to the target value set in the parameter target value setting step (step S43). In order to approach the optimal value, parameters including descriptors (e.g., mass ratio, property values, structural values, etc.) of the virtual polymer composite material are changed and optimized to become optimal parameters (optimization step: Step S48).

In the optimization process (step S48), the optimization unit 46 uses the descriptor of the virtual polymer composite material acquired in the acquisition process (step S41) and the target set in the parameter target value setting process (step S43). Parameters including descriptors of the virtual polymer composite material are optimized based on the values so that the physical properties of the virtual polymer composite material change in a desirable direction (a direction closer to the target value).

The optimization method used in the optimization step (step S48) is the same as the optimization method used in the optimization step (step S39) of the composition proposal method shown in FIG. 9, so the details will be omitted.

In the optimization step (step S48), the optimization unit 46 optimizes the parameters including the descriptor of the virtual polymer composite material, and then the process moves to the acquisition step (step S41) again. Then, the acquisition unit 41 inputs parameters including the optimized descriptor of the virtual polymer composite material as a virtual descriptor to the acquisition unit 41 as explanatory variables, and performs the same process as described above.

Then, in the first determination step (step S45), optimization of parameters including descriptors of the virtual polymer composite material is performed a predetermined number of times, and in the second determination step (step S46), optimization of the parameters including the descriptor of the virtual polymer composite material is performed. In the optimization step (step S48), optimization of the descriptor of the virtual polymer composite material is repeated until it is determined that the difference between the parameter including the descriptor and the target value is within a predetermined range.

Finally, in the first determination step (step S45), optimization of the parameters including the descriptor of the virtual polymer composite material is performed a predetermined number of times, and in the second determination step (step S46), the optimization of the parameters including the descriptor of the virtual polymer composite material is After it is determined that the difference between the parameter including the descriptor of the composite material and the target value is within a predetermined range, the display unit 47 displays the proposed virtual height in the optimization process (step S48), as described above. Information regarding parameters including a descriptor of the molecular composite material is displayed (display step: step S47).

The composition proposal method according to the present embodiment includes a parameter target value setting step (step S42), a comparison step (step S43), a first determination step (step S44), a second determination step (step S46), and an optimization step ( step S48). In the composition proposal method, in the first determination step (step S45), it is determined whether the parameters including the descriptor of the virtual polymer composite material acquired in the acquisition step (step S41) have been optimized a predetermined number of times. judge. In the composition proposal method, in the optimization step (step S48), the physical properties of the virtual polymer composite material predicted by inputting parameters including descriptors of the virtual polymer composite material into the learned model M3 are as follows. Parameters including descriptors of the virtual polymer composite are optimized to approach the target values.

In the composition proposal method, in the first determination step (step S44), it is determined that optimization of the parameters including the descriptor of the virtual polymer composite material acquired in the acquisition step (step S41) has been performed a predetermined number of times. In the optimization process (step S48), it is possible to repeat the operation of optimizing the parameters including the descriptor of the virtual polymer composite material so that the physical properties of the virtual polymer composite material approach the target values. .

In addition, in the comparison step (step S43), the composition proposal method uses the descriptor of the virtual polymer composite material acquired in the acquisition step (step S41) to the target value set in the parameter target value setting step (step S42). In the second determination step (step S46), it can be determined whether the descriptor of the virtual polymer composite material is within a predetermined range of the target value. In the composition proposal method, in the optimization step (step S48), the parameters including the descriptor of the virtual polymer composite material are optimized so that the parameters including the descriptor of the virtual polymer composite material approach the target values. do.

The composition proposal method continues through the optimization step (step S48) until it is determined in the second determination step (step S46) that the parameters including the descriptor of the virtual polymer composite material are within a predetermined range of target values. In this step, the operation of optimizing the parameters including the descriptor of the virtual polymer composite material can be repeated so that the parameters including the descriptor of the virtual polymer composite material approach the target values.

As a result, the composition proposal method according to the present embodiment can make the parameters including the descriptor of the polymer composite material substantially equivalent to the set target value. The virtual polymer composite material can have substantially the same physical properties as the polymer composite material set as the target value. Therefore, the composition proposal method according to the present embodiment can propose parameters including a descriptor of a polymer composite material that will yield a polymer composite material having desired characteristics.

Furthermore, the composition proposal method according to the present embodiment includes an optimization step (step S48), thereby reducing the burden and time required when optimizing parameters including descriptors of a virtual polymer composite material. can be achieved. Therefore, the composition proposal method according to the present embodiment can efficiently propose parameters including a descriptor of a polymer composite material that is approximately equivalent to a parameter including a descriptor of a polymer composite material that satisfies desired characteristics. It is possible to reduce energy consumption when selecting composite materials.

Hereinafter, the embodiments will be described in more detail with reference to Examples, but the embodiments are not limited to these Examples and Comparative Examples.

<Example 1>
(1) Generation of learning data set A learning data set used in a learning device to be described later was generated by the following procedure. In this example, a heterophasic propylene polymer material was used as the propylene polymer (A) contained in the polymer composite material.

(Obtaining properties of polypropylene raw materials)
The intrinsic viscosity of the polymer (I) contained in the heterophasic propylene polymer material (hereinafter referred to as "P part intrinsic viscosity"), which is a characteristic of the polypropylene raw material used in the training dataset, and the characteristic of the heterophasic propylene polymer material The intrinsic viscosity of the contained polymer (II) (hereinafter referred to as "EP part intrinsic viscosity") and the content of the polymer (II) in the heterophasic propylene polymer material in 100% by mass of the heterophasic propylene polymer material ( (hereinafter referred to as "EP content") (unit: mass %) and the content of structural units derived from ethylene contained in polymer (II) in 100 mass% of polymer (II) (hereinafter referred to as "EP content") (unit: mass %) ) (unit: mass %) was measured as follows. In addition, the intrinsic viscosity, which is a characteristic of polyolefin elastomers, was measured as follows.

(Intrinsic viscosity η (unit: dL/g))
Tetralin solution of propylene polymer powder (concentration: 0.1 g/dL, 0.2 g/dL, 0 After preparing three types of tetralin solutions (3 types of .5 g/dL), the reduced viscosity of the tetralin solution was measured at 135° C. using an Ubbelohde viscometer. The intrinsic viscosity of an olefin polymer can be calculated by plotting the reduced viscosity against the concentration according to the calculation method described in "Polymer Solutions, Polymer Experiments 11" (Kyoritsu Publishing Co., Ltd., 1982), page 491. It was determined by extrapolation to extrapolate to zero.

The EP part intrinsic viscosity was determined by the following method.
The intrinsic viscosity of the P portion actually measured by the above method was defined as [η] _I . The intrinsic viscosity of the heterophasic propylene polymer material was defined as [η] _Total . The EP content calculated by the method described below was defined as X _II . The EP part intrinsic viscosity [η] _II was calculated from the following formula.
[η] _II =100 {[η] _Total - [η] _I (100-X _II )/100}/X _II

(EP content)
The EP content X _II was determined by the following formula.
X _II = 100 {(1-(ΔHf) _Total )/(ΔHf) _I }
However, the heat of fusion (J/g) of the polymer (I) is (ΔHf) _I , and the heat of fusion (J/g) of the heterophasic propylene polymer material is (ΔHf) _Total .

(Amount of ethylene in EP)
The amount of ethylene (C ₂ ) _Total (unit: mass %) in the heterophasic propylene polymer material was determined by infrared absorption spectroscopy. The amount of ethylene in EP was calculated using the following formula.
Amount of ethylene in EP = 100 (C ₂ ) _Total /X _II

(Calculation of descriptors for polymer composite materials)
As a descriptor of the polymer composite material, the average structure value, mass ratio, and product of the average structure value and the mass ratio for each type of raw material of the polymer composite material were calculated as follows.

((Descriptor for heterophasic propylene polymeric material))
The respective characteristic values of the heterophasic propylene polymer material A-i (i = 1, 2...n) include the P part content Pcont, the P part intrinsic viscosity ηp,i, the EP part intrinsic viscosity ηep,i, and the EP content. Five descriptors were used: EPcont,i, and the amount of ethylene in the EP portion C2EP,i.

((Descriptor regarding polyolefin elastomer))
Three descriptors of intrinsic viscosity ηpoe,i and logarithmization MFRlogMFRpoe,i were used for each characteristic value of polyolefin elastomer B-i (i=1, 2...n).

((Descriptor regarding the sum of the mass ratios of other components))
The total value of the mass ratio was calculated for each of the inorganic filler, nucleating agent, and peroxide.

(Calculation of average structure value)
((Average structural values of descriptors for heterophasic propylene polymeric materials))
The average structural values of the five descriptors used as characteristic values of the heterophasic propylene polymer material A-i (i=1, 2...n) were calculated as shown below.
・av. Pcont=Σ_i(100-EPcont,i)×XAi/100
・av. EPcont=Σ_iEPcont, i×XAi/100
・av. ηp={Σ_iηp, i×(100−EPcont,i)×XAi/100}/av. Pcont
・av. ηep={Σ_i ηep, i×EPcont, i×XAi/100}/av. EPcont
・av. C2EP={Σ_i C2EP, i×EPcont,i×XAi/100}/av. EPcont

((Average structure value of descriptors for polyolefin elastomers))
The average structural value of the three descriptors used as the respective characteristic values of the polyolefin elastomer Bi (i=1, 2...n) was calculated.
・av. POEcont=Σ_iXBi
・av. ηpoe={Σ_iηpoe, i×POEcont, i×XBi/100}/av. POEcont
・av. logMFRpoe={Σ_i logMFRpoe, i×POEcont, i×XBi/100}/av. POEcont

((Amount of rubber component in the composition))
The amount of rubber component (av.RubberCont) in the composition was calculated from the following formula.
av. RubberCont=av. EPcont+av. POEcont

(cross term)
As an additional descriptor, the product of the calculated average structural value of each component and the mass ratio was calculated according to the following formula.
av. ηp×Pcont, av. ηep×EPcont, av. C2EP×EPcont, av. ηpoe×POEcont, av. logMFRpoe×POEcont

(Calculation of descriptors for polymer composite materials)
The calculation results of the descriptor of the polymer composite material used are shown. The polymer composite materials include heterophasic propylene polymer material A-1, heterophasic propylene polymer material A-2, polyolefin elastomer B-1, polyolefin elastomer B-2, talc C as an inorganic filler, and nucleating agent D. and peroxide E, the weight ratio of which is A-1:A-2:B-1:B-2:C:D:E=25:25:15:15:20:0.1:0 The composition descriptor was calculated to be .15. However, the total of A to C was 100 parts by mass. The characteristics of each raw material were as follows.
A-1: ηp = 1.0, ηep = 2.0, EP content = 30%, C2'/EP = 50%
A-2: ηp = 1.2, ηep = 6.0, EP content = 10%, C2'/EP = 30%
B-1: ηpoe=1.0, logMFR=1.00
B-2: ηpoe=1.2, logMFR=0.58

First, 12 types of descriptors shown below were calculated.
(A). Regarding the heterophasic propylene polymer material (the total of A to C was 100 parts by mass)
(A-1). Content of part P in the composition av. Pcont=(100-30)×25/100+(100-10)×25/100=40.0
(A-2). The content of EP part in the composition av. EPcont=30×25/100+10×25/100=10.0
(A-3). Average intrinsic viscosity of part P av. ηp={1.0×(100-30)×25/100+1.2×(100-10)×25/100}/40.0=1.1125
(A-4). Average intrinsic viscosity of the EP part av. ηep={2.0×30×25/100+6.0×10×25/100}/10.0=3.0
(A-5). Average C2'/EP av. C2EP={50×30×25/100+30×10×25/100}/10.0=45.0
(B). Regarding the polyolefin elastomer (the total of A to C was 100 parts by mass)
(B-1). Content of polyolefin elastomer in the composition av. POEcont=15+15=30.0
(B-2). Average intrinsic viscosity of polyolefin elastomer av. ηpoe=(1.0×15+1.2×15)/30=1.1
(B-3). Average log MFR of polyolefin elastomer av. logMFRpoe=((logMFR of B-1)×15+(logMFR of B-2)×15)/30=0.79
(C). Regarding the inorganic filler Content of inorganic filler in the composition fillerCont=20
(D). Regarding the nucleating agent, the content of the nucleating agent in the composition NuCont=0.1
(E). Content of peroxide in the composition NuCont=0.15 for peroxide
(F). Regarding the rubber component in the composition: Content of rubber component in the composition RubberCont=10.0+30.0=40.0

Subsequently, as additional descriptors, five types of cross terms consisting of the product of the content in the composition and the average structure value were calculated as follows.
(A). av. ηp×Pcont=1.1125×40=44.5
(B). av. ηep×EPcont=3.0×10=30
(C). av. C2EP×EPcont=45.0×10=450
(D). av. ηpoe×POEcont=1.1×30=33
(E). av. logMFRpoe×POEcont=0.79×30=23.7

(Obtaining physical properties of polymer composite materials)
A polymer composite material was obtained by twin-screw extrusion of the polymer composite material according to the method described in WO2020/149284. As physical property values of the polymer composite material, the flexural modulus of the injection molded product, IZOD impact strength at room temperature (normal temperature IZOD impact strength), IZOD impact strength at low temperature (low temperature IZOD impact strength), and MFR of the composite material were measured.

(Creation of training dataset)
A learning data set is created by associating the parameters including the calculated descriptor of the polymer composite with the physical properties of the polymer composite, such as flexural modulus, room temperature IZOD impact strength, low temperature IZOD impact strength, and MFR. did.

(Generation of trained model)
Then, a learned model M1 was generated using the learning device 1 shown in FIG.

(2) Learning device (2-1) Preprocessing unit In the generated learning data set, for data with duplicate parameters including composite material descriptors, the median value of the physical property values of the composite material is used as a representative value, and the Parameters including material descriptors were transformed into datasets with different data. Subsequently, the following standardization process was performed.

(2-2) Learning section Using the learning dataset preprocessed by the preprocessing section, learning was performed in the learning section to generate a trained model (learning step).

The trained models include a trained model for predicting flexural modulus of polymer composite materials, a trained model for predicting room temperature IZOD impact strength, a trained model for predicting low temperature IZOD impact strength, and a trained model for predicting MFR. A trained model was generated.

(Generation of trained model for bending elastic modulus prediction)
The preprocessing unit performs standardization processing on the number of training datasets shown in Table 1, divides the dataset into training dataset and validation dataset using the 5x5 double cross-validation method, and optimizes hyperparameters. We verified the prediction accuracy of the trained model.

Then, a trained model was constructed by support vector regression (SVR) using the regression equation construction data. A radial basis function was used as the kernel function. The generated trained model includes a learning program, and Python (registered trademark) was used as the programming language for the learning program. Additionally, the machine learning library Scikit-learn was used.

Figure 11 shows the predicted value (unit: MPa) of the bending elastic modulus predicted by the learned model constructed using the regression formula construction data using the actual measured value (unit: MPa) of the bending elastic modulus and the verification data. : MPa). Note that the determination coefficient R ² between the actual bending elastic modulus and the predicted bending elastic modulus was 0.96. The coefficient of determination ^R2 was obtained by a 5x5 double cross-validation method. The root mean square error (RMSE) between the actual and predicted intrinsic viscosity (η) is 90 MPa, and the optimal hyperparameters are C = 73 and ε = 0. .07, and γ was 0.009. Here, C is a regularization parameter, ε is a parameter indicating the width of the dead zone of support vector regression, and γ is a parameter indicating the spread in the radial basis function kernel.

(Generation of trained models for room temperature IZOD impact strength, low temperature IZOD impact strength and MFR, and prediction of room temperature IZOD impact strength, low temperature IZOD impact strength and MFR)
Room temperature IZOD impact strength, low temperature IZOD impact strength, and MFR were also calculated in the same manner as the flexural modulus. The respective prediction results are shown in FIGS. 12 to 14.

Figure 12 shows the actual measured value of the room temperature IZOD impact strength (unit: kJ/m ² ) and the room temperature IZOD impact strength predicted by the trained model constructed using the regression formula construction data using the verification data. represents the relationship with the predicted value (unit: kJ/m ² ). The coefficient of determination R ² is 0.72, the RMSE is 8.9 kJ/m ² , and the optimal hyperparameters are C 16, ε 0.06, and γ 0.01. Ta.

Figure 13 shows the actual measured value of low-temperature IZOD impact strength (unit: kJ/m ² ) and the low-temperature IZOD impact strength predicted by the trained model constructed using the regression formula construction data using the verification data. represents the relationship with the predicted value (unit: kJ/m ² ). The coefficient of determination R ² is 0.77, the RMSE is 0.6 kJ/m ² , and the optimal hyperparameters are C 7, ε 0.04, and γ 0.08. Ta.

FIG. 14 shows the relationship between the actual measured value of MFR and the predicted value of MFR predicted by a learned model constructed using regression formula construction data using verification data. Note that the MFR was logarithmized and output as logMFR. The coefficient of determination R ² was 0.95, the RMSE was 0.05, and the optimal hyperparameters were C 17, ε 0.002, and γ 0.05.

From FIGS. 11 to 14, it was confirmed that highly accurate trained models were obtained for the bending elastic modulus, room temperature IZOD impact strength, low temperature IZOD impact strength, and MFR.

(3) Prediction device Using the learned model generated by the learning device 1 shown in FIG. 2, the prediction device 2 shown in FIG. 3 was produced. The physical properties of polymer composite materials were predicted using the constructed prediction device. In order to compare with the predicted results, a molded article was prepared using the corresponding polymer composite material and evaluated. An example in which a polymer composite material was synthesized and evaluated experimentally was used as a synthesis example, and an example in which the physical properties of a polymer composite material were predicted using the trained model described above was used as a usage example.

(3-1) Synthesis example 1 and usage example 1
(Synthesis Example 1) Synthesis of polymer composite material A polymer composite material made of the following raw materials was actually synthesized, and its physical properties were measured. The synthesis and physical property measurements of the polymer composite material followed the method described in WO2020/149284.
A-3 (58 parts by mass): ηp = 0.79, ηep = 7.0, EP content = 11%, C2'/EP = 32%
B-3 (19 parts by mass): ηpoe=2.1, logMFR=-0.70
B-4 (8 parts by mass): ηpoe=1.5, logMFR=0.04
C-1 (15 parts by mass): Talc MW-UPN-TT-H (Hayashi Kasei)
Nucleating agent (0.1 part by mass): Adekastab NA-25 (Adeka)
The physical properties were as follows.
Flexural modulus: 1440MPa
Room temperature IZOD impact strength: 39kJ/ ^m2
Low temperature IZOD impact strength: 5.8kJ/ ^m2
Logarithmization MFR: 1.35 (actual value before logarithmization: 22.4g/10 minutes)

(Usage Example 1) Prediction of Properties of Polymer Composite Material Using the prediction device described above, the flexural modulus, normal temperature IZOD impact strength, low temperature IZOD impact strength, and MFR were predicted as physical property values of the polymer composite material. The predicted value of the flexural modulus is 1480 MPa, the predicted value of the room temperature IZOD impact strength is 44 kJ/ ^m2 , the predicted value of the low temperature IZOD impact strength is 5.5 kJ/ ^m2 , and the predicted value of the logarithmic MFR. The predicted value was 1.40 (actual value conversion: 25.1 g/10 minutes).

Note that parameters including descriptors of polymer composite materials to be predicted are input to the trained model, and parameters including descriptors of polymer composite materials are input to the trained model. It was calculated in the same way as the parameters including composite material descriptors.

<Example 2>
(1) Generation of training dataset This was performed in the same manner as in Example 1 above.
(2) Learning device It was conducted in the same manner as in Example 1 above.
(3) Prediction device It was performed in the same manner as in Example 1 above.
(4) Composition Proposal Device Using Optuna (registered trademark), a Python library for optimization, a composition of a polymer composite material with good properties was proposed. Below is an example (Example 2-1) in which a group of compositions that maximize or minimize the physical properties of a polymer composite material in a desirable direction is proposed, and a composition that brings the physical properties of a polymer composite material close to desirable values. An example (Example 2-2) in which a group of is proposed is shown below.

(Example 2-1) Method for proposing a group of compositions that maximizes or minimizes the physical properties of polymer composite materials to desired values Hypothetical composition Rj (j=1, 2...n) and composition A pair of predicted values of properties of the composite material corresponding to the group Pj (j=1, 2...n) was obtained as follows.
For a composition containing a heterophasic propylene polymer material, a polyolefin elastomer, and talc, heterophasic propylene polymer material A, polyolefin elastomer B, and talc C are each randomly selected from the list of available raw materials, and the mass ratio is determined. By setting randomly, an initial composition R1 was obtained. However, the total of heterophasic propylene polymer material A, polyolefin elastomer B, and talc C in R1 was 100 parts by mass. At this time, each raw material was one type or multiple types.
Parameters including descriptors for the initial composition R1 were calculated using a method similar to the method described in the above learning device, and a group P1 of predicted values of characteristics corresponding to the composition R1 was obtained using the above prediction device. . P1 is a group of predicted values of characteristics corresponding to the set trained model.
From the virtual composition R1 and the group P1 of predicted values of properties, a composition R2 having desirable properties was proposed using the optimization library Optuna. A genetic algorithm was used as the algorithm for the proposal.

As an objective function for optimization, by specifying the property item to be optimized and the direction of optimization (maximization or minimization), the composition can be optimized so that multiple property values are good. It became.

From the proposed composition R2, a group P2 of predicted values of the properties of the corresponding composite material was obtained.

By repeating the following, a virtual composition Rj (j=1, 2...n), a group of predicted values of composite material properties corresponding to the composition Pj (j=1, 2...n), and I got a pair of

From the pair of the obtained virtual composition Rj (j=1, 2...n) and the group Pj (j=1, 2...n) of predicted values of the properties of the composite material corresponding to the composition , a group of compositions with good predicted properties were selected.

For compositions containing a heterophasic propylene polymer material, one or two polyolefin elastomers, and talc, a genetic algorithm was used to determine the flexural modulus (unit: MPa) and room temperature IZOD impact strength (unit: kJ/m2 ⁾ . FIG. 14 shows the results of proposing a group of compositions that are optimized so that a plurality of properties consisting of As shown in FIG. 14, a group of compositions (Pareto optimal solutions, white circles in FIG. 14) were obtained that were optimized so that both the flexural modulus and the room-temperature IZOD impact strength were maximized.

Therefore, it has been confirmed that the composition proposal device can efficiently propose a group of compositions (Pareto optimal solutions) whose properties are maximized in the desired direction among compositions for which many combinations are possible.

(Example 2-2) Method for proposing a group of compositions that bring the properties of a composite material closer to desired values Setting the target value of the physical property k of the polymer composite material to Ok, predicting the physical properties of the polymer composite material corresponding to the composition Rj When the value is Pk,j, the objective function fk for optimization is determined as shown in the following equation (1).
fk=|Ok-Pk,j|/|Ok| ...(1)
(In the formula, |Ok-Pk,j| and |Ok| indicate the absolute values of Ok-Pk,j and Ok, respectively.)

By optimizing the composition to minimize the objective function fk for optimization using a method similar to the method described in (2-1) above, the physical properties of the polymer composite material are brought closer to desired values. It was confirmed that it is possible to propose a group of compositions for polymer composite materials.

Although the embodiments have been described as above, the embodiments are presented as examples, and the present invention is not limited to the embodiments described above. The embodiments described above can be implemented in various other forms, and various combinations, omissions, substitutions, changes, etc. can be made without departing from the gist of the invention. These embodiments and their modifications are included within the scope and gist of the invention, as well as within the scope of the invention described in the claims and its equivalents.

Note that aspects of the embodiment of the present invention are, for example, as follows.
<1> A composition proposal device that proposes the types and mass ratios of a type of polymer and components other than the type of polymer contained as constituent components in a polymer composite material,
a setting section for setting target values of physical properties of a virtual polymer composite material;
The descriptor of the virtual polymer composite material is added to a trained model trained using a training data set in which parameters including the descriptor of the polymer composite material are associated with physical properties of the polymer composite material. an optimization unit that changes parameters including a descriptor of the virtual polymer composite material so that physical properties of the virtual polymer composite material predicted by inputting parameters including a descriptor of the virtual polymer composite material approach the target value; and,
has
The optimization unit changes the parameters including the descriptor of the virtual polymer composite material each time the physical properties of the virtual polymer composite material are predicted. A composition proposal device that optimizes the type and mass ratio of the above-mentioned one type of polymer and components other than the above-mentioned one type of polymer, which are included as constituent components.
<2> The composition proposing device according to <1>, wherein the optimization unit changes so as to avoid a specific descriptor of the virtual polymer composite material.
<3> The optimization unit is configured to combine the one kind of polymer and the one contained in the virtual polymer composite material as the constituent components in order to realize parameters including a descriptor of the virtual polymer composite material. The composition proposal device according to <1> or <2>, which proposes the type and mass ratio of components other than a type of polymer by performing mathematical optimization processing.
<4> The composition proposing device according to <3>, wherein the mathematical optimization process is any one of a genetic algorithm, a base optimization, and a TPE.
<5> The optimization unit optimizes parameters including descriptors of the virtual polymer composite material so that physical properties of the virtual polymer composite material are maximized in a desired direction <1> to <4 ＞The composition proposal device according to any one of ＞.
<6> The optimization unit proposes a plurality of combinations of types and mass ratios of the one kind of polymer and components other than the one kind of polymer, which are included as the constituent components in the polymer composite material. <3> ~The composition proposal device according to any one of <5>.
<7> The optimization unit performs prediction by inputting parameters including descriptors of the virtual polymer composite material into the trained model generated for each of two or more properties of the polymer composite material. parameters including descriptors of the virtual polymer composite material such that each of the two or more properties of the virtual polymer composite material approaches the target value of each of the two or more properties. The composition proposal device according to any one of <1> to <6>, which is changed.
<8> The composition proposal device according to <7>, wherein the optimization unit proposes a Pareto solution for the two or more characteristics.
<9> The composition proposal device according to any one of <1> to <8>, wherein the mass ratio includes an average value for each type of the constituent components.
<10> The composition proposing device according to any one of <1> to <9>, wherein the polymer includes a copolymer containing two or more types of monomer units.
<11> The composition proposing device according to any one of <1> to <10>, wherein the polymer composite material includes a propylene polymer and a polyolefin elastomer.
<12> The composition proposing device according to <11>, wherein the propylene-based polymer includes a copolymer containing a monomer unit derived from propylene and a monomer unit derived from ethylene.

1 learning device 2 prediction device 3 composition proposal device 11, 21, 31, 41 acquisition unit 12, 22, 32, 42 preprocessing unit 13 learning dataset creation unit 14 learning unit 15, 24, 38, 47 display unit 23, 33 Prediction unit 34, 43 Setting unit 35, 44 Comparison unit 36, 45 Judgment unit 37, 46 Optimization unit M1, M2, M3 Learned model

Claims (14)

A composition proposal device that proposes the types and mass ratios of a type of polymer and components other than the type of polymer contained as constituent components in a polymer composite material, comprising:
a setting section for setting target values of physical properties of a virtual polymer composite material;
The descriptor of the virtual polymer composite material is added to a trained model trained using a training data set in which parameters including the descriptor of the polymer composite material are associated with physical properties of the polymer composite material. an optimization unit that changes parameters including a descriptor of the virtual polymer composite material so that physical properties of the virtual polymer composite material predicted by inputting parameters including a descriptor of the virtual polymer composite material approach the target value; and,
has
The optimization unit is configured to describe the virtual polymer composite material by changing parameters including a descriptor of the virtual polymer composite material each time the physical properties of the virtual polymer composite material are predicted. Genetic algorithm calculates the types and mass ratios of the type of polymer and components other than the type of polymer, which are included as the constituent components of the virtual polymer composite material, in order to realize parameters including children. Optimize by doing
The descriptor of the virtual polymer composite material includes a mass ratio, a characteristic value, and a structural value of the constituent components contained in the virtual polymer composite material,
The characteristic value and the structural value are average values for each type of the constituent component,
A composition in which the structural value is at least one selected from the group consisting of the content of the constituent components, the composition ratio of a copolymer contained in the constituent components, and the monomer unit ratio of the copolymer. Proposed device. A composition proposal device that proposes the types and mass ratios of a type of polymer and components other than the type of polymer contained as constituent components in a polymer composite material, comprising:
a setting section for setting target values of physical properties of a virtual polymer composite material;
The descriptor of the virtual polymer composite material is added to a trained model trained using a training data set in which parameters including the descriptor of the polymer composite material are associated with physical properties of the polymer composite material. an optimization unit that changes parameters including a descriptor of the virtual polymer composite material so that physical properties of the virtual polymer composite material predicted by inputting parameters including a descriptor of the virtual polymer composite material approach the target value; and,
has
The optimization unit is configured to describe the virtual polymer composite material by changing parameters including a descriptor of the virtual polymer composite material each time the physical properties of the virtual polymer composite material are predicted. Optimize parameters including children,
The descriptor of the virtual polymer composite material includes a mass ratio, a characteristic value, and a structural value of the constituent components contained in the virtual polymer composite material,
The characteristic value and the structural value are average values for each type of the constituent component,
A composition in which the structural value is at least one selected from the group consisting of the content of the constituent components, the composition ratio of a copolymer contained in the constituent components, and the monomer unit ratio of the copolymer. Proposed device. 3. The composition proposing device according to claim 2, wherein the physical properties of the polymer composite material are moldability, rigidity, impact resistance, and gloss. The optimization unit is configured to determine the type of polymer and the type of polymer included in the virtual polymer composite material as the constituent components in order to realize parameters including a descriptor of the virtual polymer composite material. 3. The composition proposing device according to claim 2, wherein the type and mass ratio of components other than molecules are proposed by performing mathematical optimization processing. The composition proposal device according to claim 4 , wherein the mathematical optimization process is one of a genetic algorithm, a base optimization, and a TPE. 3. The composition proposing device according to claim 1, wherein the optimization unit changes the parameters to avoid a specific descriptor of the virtual polymer composite material. The composition according to claim 1 or 2, wherein the optimization unit optimizes parameters including a descriptor of the virtual polymer composite material so that physical properties of the virtual polymer composite material are maximized in a desired direction. Proposed device. 3. The optimization unit according to claim 1 or 2, wherein the optimization unit proposes a plurality of combinations of types and mass ratios of the type of polymer and components other than the type of polymer, which are included as the constituent components in the polymer composite material. The composition proposal device described. The optimization unit inputs parameters including descriptors of the virtual polymer composite material into the learned model generated for each of the two or more properties of the polymer composite material, thereby making predictions. , changing parameters including descriptors of the virtual polymer composite material so that each of the two or more properties of the virtual polymer composite material approaches the target value of each of the two or more properties. The composition proposal device according to claim 1 or 2. The composition proposal device according to claim 9 , wherein the optimization unit proposes a Pareto solution for the two or more characteristics. The composition proposal device according to claim 1 or 2, wherein the mass ratio includes an average value for each type of the constituent components. The composition proposing device according to claim 1 or 2, wherein the polymer includes a copolymer containing two or more types of monomer units. The composition proposing device according to claim 1 or 2, wherein the polymer composite material includes a propylene polymer and a polyolefin elastomer. 14. The composition proposing device according to claim 13 , wherein the propylene-based polymer includes a copolymer containing a monomer unit derived from propylene and a monomer unit derived from ethylene.

Images