JP2020038493A - Method and device for predicting physical property data - Google Patents

Abstract

To allow for accurately and efficiently predicting physical property data of a vulcanized rubber composition produced from an unvulcanized rubber composition comprising a mix of multiple raw materials using a mixing ratio of each raw material.SOLUTION: A method disclosed herein comprises: making a prediction module of a computer machine-learn physical property data of multiple vulcanized rubber compositions using the physical property data, identification names of raw materials of the vulcanized rubber compositions, mixing ratios of the raw materials, and information on processing conditions; making the machine-learned prediction module predict physical property data of a target vulcanized rubber composition using names of constituent raw materials constituting an unvulcanized rubber composition of the target vulcanized rubber composition before being vulcanized, mixing ratios of the constituent raw materials, and processing conditions for a process of producing the target vulcanized rubber composition; and performing preprocessing for classifying each of the names of the constituent raw materials into one of the identification names that was used as learning input data on the basis of material properties of the constituent raw materials prior to proceeding with the prediction by the prediction module.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a physical property data prediction method and a physical property data prediction device for predicting physical property data of a vulcanized rubber composition.

In recent years, techniques for making computers perform machine learning and performing various predictions from input data have been actively proposed. For the vulcanized rubber composition prepared by blending a plurality of rubber materials, fillers, oils, and the like as raw materials, it is conceivable to apply the above technique to predicting physical property data.
BACKGROUND ART Conventionally, a vulcanized rubber composition has been prepared by mixing a plurality of rubber materials, fillers, oils, and the like by trial and error, and physical property data has been measured. For this reason, a large number of data in which the compounding information of the vulcanized rubber composition and the physical property data are linked. By utilizing this accumulated data, a computer can make machine learning to predict physical property data.

  For example, using a neural network technique, learn the mapping relationship between the group of factors and the group of characteristics of experimental data such as design and composition, estimate the characteristic values from the factor conditions, and apply it to any characteristic data. (Patent Document 1) is known which provides a method for efficiently and easily finding an optimum value of factor data for generating the data.

JP-A-2003-58582

In learning of a neural network in this technique, all prepared data are read uniformly and used for a plurality of learning data.
For example, learning of the neural network is performed using factor data representing the content ratio of each chemical composition in the rubber composition to be predicted. Since such factor data has a data structure in which raw materials composed of a plurality of chemical compositions are ungrouped, it is not preferable to use learning data for predicting characteristic data.

  Generally, a vulcanized rubber composition is prepared from an unvulcanized rubber composition in which raw materials containing a plurality of chemical compositions are combined and compounded. preferable.

  Thus, the present invention provides a computer for predicting the physical property data of a vulcanized rubber composition produced from an unvulcanized rubber composition prepared by combining a plurality of preset raw materials, with high efficiency and accuracy. An object of the present invention is to provide a physical property data prediction method and a physical property data prediction device capable of predicting physical property data (characteristic data) of a vulcanized rubber composition.

One embodiment of the present invention is a method for predicting physical property data in which a computer predicts physical property data of a vulcanized rubber composition produced from an unvulcanized rubber composition obtained by combining a plurality of preset raw materials. The prediction method is
A plurality of raw materials used for the vulcanized rubber composition are blended at a set blending ratio, and physical property data on a plurality of vulcanized rubber compositions produced by processing under the set processing conditions are used as learning output data. The identification data of each of the raw materials in the vulcanized rubber composition, the compounding ratio of each of the raw materials, and information of each of the processing conditions, using the learning data as learning input data, the physical property data, Machine learning a prediction module in a computer;
The names of constituent raw materials constituting the unvulcanized rubber composition before vulcanization of the predicted vulcanized rubber composition, the mixing ratio of the constituent raw materials, and the processing for producing the predicted vulcanized rubber composition Using processing conditions, the step of causing the prediction module to predict the physical property data of the vulcanized rubber composition to be predicted,
Before the step of causing the prediction module to predict the physical property data of the vulcanized rubber composition to be predicted,
Causing the computer to perform a pre-process of classifying the name of the constituent raw material into one of the identification names used as the learning input data, based on the material characteristics of the constituent raw material;
Causing the computer to input one of the classified identification names to the prediction module as a name corresponding to the constituent raw material for the prediction by the prediction module;
Is provided.

  The name of the constituent raw material, using raw material physical property data of the constituent raw material as the material property, based on a comparison result of raw material physical property data of the constituent raw material and raw material physical property data of the raw material used for the learning data, It is preferable to be classified into one of the identification names.

The raw material property data includes a plurality of different types of raw material property values,
The names of the constituent raw materials are classified into one of the identification names based on a correlation coefficient between the physical property values of the plurality of types of raw materials and the physical property values of the corresponding types of the raw materials used in the learning data. Is preferably performed.

The raw material used for the learning data and the constituent raw material in the predicted vulcanized rubber composition is a mixture of a plurality of chemical compositions,
In the pretreatment, it is preferable that a name of the constituent raw material is classified into one of the identification names by using a composition ratio of the chemical composition as the material property.

  The name of the constituent raw material, the composition ratio of the chemical composition in the constituent raw material, the identification name of the identification name based on the correlation coefficient between the composition ratio of the chemical composition in the raw material used for the learning data Preferably, they are classified into one.

The constituent raw materials and the raw materials used for the learning data have a main component chemical composition having the composition ratio of 80% or more,
The name of the constituent raw material is such that, among the raw materials used for the learning data, the difference between the composition ratio of the main component chemical composition of the constituent raw material and the composition ratio of the main component chemical composition of the raw material is 5%. It is preferable that the raw materials are classified into the identification names of the raw materials within.

The raw material and the constituent raw materials used for the learning data include a material whose shape is specified by a shape parameter,
In the preprocessing, it is preferable that a name of the constituent raw material is classified into one of the identification names of the raw material used for the learning data, using the shape parameter as the material characteristic.

  It is preferable that the shape parameters include shapes, shape sizes, or shape aspect ratios of the raw material and the constituent raw materials.

  The name of the constituent raw material is classified into one of the identification names of the raw material in which the value of the shape parameter of the constituent raw material matches within an allowable range the value of the shape parameter of the raw material used in the learning data. Is preferably performed.

The raw materials and the constituent raw materials used for the learning data include a plurality of synthetic rubbers that are polymers having different bonding modes and ratios of the bonding modes,
In the preprocessing, it is preferable that a name of the constituent raw material is classified into one of the identification names of the raw material used in the learning data, using a ratio of the bonding mode as the material characteristic.

  The name of the constituent raw material is one of the identification names based on a correlation coefficient between the ratio of the bond mode in the constituent raw material and the ratio of the bond mode in the raw material used for the learning data. Preferably, they are classified.

  It is preferable that the names of the constituent raw materials are classified into identification names of raw materials in which the correlation coefficient exceeds a predetermined value among the raw materials used for the learning data.

Another embodiment of the present invention is a physical property data prediction device for predicting physical property data of a vulcanized rubber composition produced from an unvulcanized rubber composition in which a plurality of preset raw materials are combined and compounded. The device is
Output data for learning is obtained by blending a plurality of raw materials used in a vulcanized rubber composition at a set blending ratio and processing them under set processing conditions. And an identification name of each of the raw materials, a blending ratio of each of the raw materials, and information of each of the processing conditions, using a learning data with learning input data as a prediction unit for machine learning the physical property data. ,
The names of constituent raw materials constituting the unvulcanized rubber composition before vulcanization of the predicted vulcanized rubber composition, the mixing ratio of the constituent raw materials, and the processing for producing the predicted vulcanized rubber composition And a processing condition, using a data operation unit that makes the prediction unit predict physical property data of the vulcanized rubber composition to be predicted,
The data operation unit includes:
A preprocessing unit that performs preprocessing of classifying the name of the constituent raw material into one of the identification names used as the learning input data by the prediction unit based on the material characteristics of the constituent raw material;
An input unit configured to input one of the identification names classified by the preprocessing unit to the prediction unit for prediction by the prediction unit, as a name corresponding to the constituent raw material;
Is provided.

  According to the physical property data prediction method and the physical property data predicting apparatus described above, when causing the computer to predict the physical property data of the vulcanized rubber composition, the physical property data (characteristic data) of the vulcanized rubber composition is efficiently and accurately calculated. Can be predicted.

(A) is a figure showing an example of a flow which makes a prediction module perform machine learning in a physical data prediction method of one embodiment, and (b) is a flow which predicts physical data using a prediction module which carried out machine learning. It is a figure showing an example of. It is a figure explaining each raw material used by the physical-property data prediction method of one Embodiment. It is a figure explaining composition of a physical property data prediction device of one embodiment.

  Hereinafter, a physical property data prediction method and a physical property data prediction apparatus according to an embodiment will be described in detail.

(Schematic explanation of physical property data prediction method)
Physical property data prediction method of one embodiment, the computer, the machine learning of the association between the physical property data and the factor data for a plurality of vulcanized rubber compositions, using a prediction module of the machine-learned computer, the vulcanized rubber to be predicted This is a method for predicting physical property data of a composition.
FIG. 1A is a diagram illustrating an example of a flow of causing a prediction module of a computer to perform machine learning, and FIG. 1B is an example of a flow of predicting physical property data using a prediction module that has performed machine learning. FIG. The computer forms at least the preprocessing module and the prediction module by activating the program called from the memory.

  As shown in FIG. 1 (a), the pre-processing module of the computer processes the raw material including the unvulcanized rubber for learning and processing from the raw material to produce a vulcanized rubber composition, as shown in FIG. 1 (a). For processing and physical property data of the produced vulcanized rubber composition are obtained (Step S10). Here, the raw materials include one commercially available material, a chemical composition alone composed of one chemical composition, or a material composed of a plurality of chemical compositions having a predetermined composition ratio. The blending information includes the name of each raw material in the vulcanized rubber composition and the blending ratio of each raw material. The processing conditions include, for example, the type of mixing machine for mixing the raw materials including unvulcanized rubber, the temperature at the time of mixing, the number of rotations of the mixing machine, the mixing pressure, and the input amount of the raw materials to the mixing machine. And a vulcanization condition including at least one vulcanization temperature and vulcanization time when vulcanizing the mixed unvulcanized rubber. Physical properties data of the vulcanized rubber composition include, for example, rubber elasticity, tan δ, specific gravity, Mooney viscosity, Mooney scorch, rheometer measurement result, Lupke JIS hardness, modulus, breaking strength, breaking energy, filler dispersion, coefficient of rebound resilience, constant It includes at least one of a strain test result, a wear test result, a fatigue test result, and crack growth characteristics.

  Next, the preprocessing module performs preprocessing of classifying the acquired names of the raw materials into identification names in order to use the obtained raw material names as learning input data based on the material characteristics of the raw materials (step S12). For example, one commercially available raw material is often a mixture of a plurality of chemical compositions in addition to a single chemical composition. For this reason, the pretreatment module integrates the names of the raw materials based on the degree of approximation of the composition ratios, using the composition ratios of the plurality of chemical compositions as the material characteristics. That is, similar raw materials are put together as one identification name. Further, the integration may be performed according to the degree of approximation between the physical property data of the raw materials. Therefore, the number of raw materials of the input data for learning is smaller than the number of acquired raw materials. For example, when the raw material α of Company A and the raw material β of Company B are similar in terms of the composition ratio, the name of the raw material β of Company B is integrated with the name of the raw material α of Company A. Raw materials include unvulcanized rubber as well as additional materials such as fillers and oils.

The prediction module performs machine learning using the learning input data obtained by the preprocessing and the physical property data (learning output data) of the vulcanized rubber composition (step S14). That is, the prediction module uses physical property data relating to a plurality of vulcanized rubber compositions as learning output data, and uses the vulcanized rubber composition created in the pre-processing, the identification name of each of the pre-processed raw materials, and the raw materials, respectively. The learning input data and the learning output data are associated by machine learning the physical property data using the learning data in which the mixing ratio and the respective processing condition information are used as the learning input data. As the machine learning of the model in the prediction module, for example, deep learning (deep learning) using a neural network is used. In addition, a random forest using a tree structure can be used. As the model, a known model such as a convolutional neural network or a stagged auto encoder can be used.
Thus, the prediction module completes the machine learning.

Next, the machine learning-based prediction module predicts physical property data based on the input data.
First, the pretreatment module is configured to name the constituent raw materials constituting the unvulcanized rubber composition before vulcanization of the predicted vulcanized rubber composition, the compounding ratio of the constituent raw materials, and the predicted vulcanized rubber composition. The input of processing conditions in the processing for manufacturing is received (step S20).
Upon receiving the input, the preprocessing module performs preprocessing for classifying the input constituent raw material names into one of the identification names used as the learning input data based on the material characteristics of the constituent raw materials (step S22). . At this time, the classification is performed, for example, using the composition ratio of the chemical composition of the constituent raw materials as a material characteristic, according to the degree of approximation between the composition ratio and the composition ratio of the name of the integrated raw material used as the learning input data. The classification is performed according to the degree of approximation between the physical property data of the constituent raw materials and the physical property data of the raw materials integrated as input data for learning.
As described above, the preprocessing module classifies the names of the constituent raw materials into one of the identification names of the raw materials used as the input data for learning, and inputs the names to the prediction module together with the mixing ratio and the processing conditions in the classified identification names. I do.
The prediction module predicts physical property data of the vulcanized rubber composition to be predicted using the input identification name, compounding ratio, and processing conditions (step S24).

  Thus, before predicting the physical property data in the prediction module, the name of the constituent raw material constituting the unvulcanized rubber composition before vulcanization of the vulcanized rubber composition to be predicted is based on the material properties of the constituent raw material. Performs preprocessing to classify into one of the identification names used as input data for learning, and inputs one of the classified identification names to the prediction module for prediction by the prediction module as a name corresponding to a constituent raw material. . For this reason, it is possible to efficiently and accurately predict physical property data (characteristic data) of the vulcanized rubber composition using the compounding ratio of each raw material.

In the pretreatment, when the names of the constituent raw materials are classified into one of the identification names using the composition ratio of the chemical composition as a material property, when the composition ratios are similar, different raw materials for each manufacturer that manufactured the raw materials Can be integrated into one name.
According to one embodiment, the names of the constituent raw materials are identified based on a correlation coefficient between the composition ratio of the chemical composition in the constituent raw materials and the composition ratio of the chemical composition in the raw materials used for the learning data. It is preferable to be classified into one of the following.

FIG. 2 is a diagram showing an example of the composition of the constituent raw material A and the compositions of the raw materials B and C corresponding to the identification names. In this case, the correlation coefficient of the composition ratios of the constituent raw material A and raw materials B is, (a1 · b1 + a2 · b2 + a3 · b3 + a4 · b4) / {(a1 2 + a2 2 + a3 2 + a4 2) (1/2) · (b1 2 + B2 ² + b3 ² + b4 ² ) ^(1/2) }, and the correlation coefficient of the composition ratio between the constituent raw materials A and C is (a1 · c1 + a2 · c2 + a3 · c3 + a4 · c4) / ｛(a1 ² + a2 ² + a3). ^{2 + a4 2) (1/2)} · (c1 2 + c2 2 + c3 2 + c4 2) is ^(1/2)}. If the correlation coefficient of the composition ratio of the constituent raw materials A and B is larger than the correlation coefficient of the composition ratio of the constituent raw materials A and C and the value exceeds a predetermined threshold, the name of the constituent material A is changed to B is the identification name. By performing classification using the correlation coefficient as described above, accurate classification can be performed, and thus, prediction of physical property data (characteristic data) of the vulcanized rubber composition can be performed efficiently and accurately. .

  When the constituent raw materials and the raw materials used for the learning data have a main component chemical composition having a composition ratio of 80% or more, the names of the constituent raw materials are the main constituent chemical components of the constituent raw materials among the raw materials used for the learning data. It is preferable that the composition ratio of the composition is classified into an identification name of the raw material in which a difference between the composition ratio of the main component chemical composition of the raw material corresponding to the identification name is within 5%. In this case as well, accurate classification can be performed, and prediction of the physical property data (characteristic data) of the vulcanized rubber composition can be performed efficiently and accurately.

Further, in the preprocessing of one embodiment, the raw material property data of the constituent raw materials are used as the material characteristics, and the names of the constituent raw materials are obtained by combining the raw material property data of the constituent raw materials and the raw material property data of the raw materials used in the learning data. It is also preferable that the information is classified into one of the identification names based on the comparison result.
A plurality of types of material property data are measured, and raw materials can be classified using the measured material property data. For example, when the raw material property data includes a plurality of different types of raw material property values, the names of the constituent raw materials are different between the plurality of types of raw material property values and the corresponding types of raw material property values in the raw materials used for the learning data. It is preferable that the data is classified into one of the identification names based on the correlation coefficient. By performing classification using the correlation coefficient as described above, accurate classification can be performed, and thus, prediction of physical property data (characteristic data) of the vulcanized rubber composition can be performed efficiently and accurately. .

  According to one embodiment, when the raw material and the constituent raw material used for the learning data include a material whose shape is specified by the shape parameter, in the preprocessing, the name of the constituent raw material is used by using the shape parameter as a material property, It is preferable to classify as one of the identification names of the raw materials used for the learning data. For example, carbon black may be used as a raw material corresponding to a constituent raw material and an identification name. The carbon black is composed of particles, and the size of the particles and the structure formed by the particles affect the degree of dispersion of the carbon black in the rubber. Therefore, when the difference between the values of the shape parameters is equal to or less than the predetermined value, they can be handled as the same raw material. In this regard, it is preferable to classify the names of the constituent raw materials into one of the identification names of the raw materials used for the learning data by using the shape parameters including the particle size and the structure as the material characteristics for classification.

The shape parameter preferably includes the shape, shape size, or shape aspect ratio of the raw material and constituent raw materials. For example, talc, which is one of the raw materials, is a plate-like filler. Since the aspect ratio (the aspect ratio of the plate shape) of such a plate-like filler affects the physical property data of rubber, the aspect ratio is preferably included as a shape parameter.
Therefore, it is preferable that the names of the constituent raw materials are classified into one of the identification names of the raw materials in which the values of the shape parameters of the constituent raw materials match the values of the shape parameters of the raw materials used in the learning data within an allowable range. .

According to an embodiment, when the raw materials and constituent raw materials used for the learning data include polymers (synthetic rubber) having different bonding modes and different bonding mode ratios, the pretreatment uses the bonding mode ratios as material characteristics. It is preferable to classify the names of the constituent raw materials into one of the identification names of the raw materials used for the learning data. For example, the diene rubber of synthetic rubber includes styrene butadiene rubber, isoprene rubber, and butadiene rubber. For example, styrene rubber and isoprene rubber include different bonding modes such as cis, trans, vinyl, and the like. By using the ratio of such a binding mode as a material property and classifying the names of the constituent raw materials into one of the identification names of the raw materials used in the learning data, the raw materials can be efficiently classified. Prediction of physical property data (characteristic data) of the vulcanized rubber composition can be performed efficiently and accurately.
For example, the names of the constituent raw materials are classified into one of the identification names based on the correlation coefficient between the ratio of the bonding style in the constituent raw materials and the ratio of the bonding style in the raw materials used for the learning data. Is preferred. The names of the constituent raw materials are classified into one of the identification names of the raw materials in which the correlation coefficient between the constituent materials and the raw materials is equal to or more than a predetermined value. By performing classification using the correlation coefficient as described above, accurate classification can be performed, and thus, prediction of physical property data (characteristic data) of the vulcanized rubber composition can be performed efficiently and accurately. . It is preferable that the names of the constituent raw materials are classified into the identification names of the constituent raw materials in which the correlation coefficient exceeds a predetermined threshold value.

FIG. 3 is a diagram illustrating an example of the configuration of the physical property data prediction device 100 that performs such a prediction method. The physical property data prediction device 100 includes a prediction unit 102 and a data operation unit 104.
The prediction unit 102 uses the physical property data of the plurality of vulcanized rubber compositions as learning output data, and identifies the identification name of each raw material, the blending ratio of each raw material, and the information of each processing condition as learning input data. It is configured to perform machine learning on physical property data using learning data to be executed. The plurality of vulcanized rubber compositions using the physical property data as the learning output data are obtained by blending a plurality of raw materials used for the vulcanized rubber composition at a set blending ratio and processing under the set processing conditions. It is made.

The prediction unit 102 is a prediction module formed in the computer.
The data operation unit 104 prepares the names of constituent raw materials constituting the unvulcanized rubber composition before vulcanization of the vulcanized rubber composition to be predicted, the mixing ratio of the constituent raw materials, and the vulcanized rubber composition to be predicted. The prediction unit 102 is configured to control the prediction unit 102 to cause the prediction unit 102 to predict the physical property data of the vulcanized rubber composition to be predicted using the processing conditions in the processing for performing the processing. The data operation unit 104 is a data operation module formed in the computer.

The data operation unit 104 includes a preprocessing unit 104a and an input unit 104b in detail.
The preprocessing unit 104a is configured to perform preprocessing for classifying the names of the constituent raw materials into one of the identification names used as input data for learning by the prediction unit based on the material characteristics of the constituent raw materials. The pre-processing unit 104a is a prediction module formed in the computer.
The input unit 104b is configured to input one of the identification names classified by the preprocessing unit 104a to the prediction unit 102 for prediction by the prediction unit 102 as a name corresponding to a constituent raw material. The input unit 104b is an input module formed in a computer.
The data operation unit 104 can integrate the names of a plurality of raw materials acquired by the preprocessing unit 104a as learning input data based on the material characteristics of the raw materials so that the prediction unit 102 performs machine learning.

When the actual physical property data is obtained by measurement etc. for the result of predicting the physical property data using the name of the constituent raw materials, the mixing ratio, and the processing conditions, the name of this constituent raw material, the mixing ratio, and the processing conditions are used for learning. Using the input data and the actual physical property data as learning output data, the prediction unit 102 may be made to perform machine learning.
Hereinafter, a more specific example will be described.

(Example of rubber material pretreatment)
In order to confirm the effect of the pretreatment using five types of rubber materials, the prediction accuracy of the physical properties of the vulcanized rubber composition to be predicted with or without the pretreatment was examined. The prediction accuracy was measured by actually measuring the physical properties of the vulcanized rubber composition produced, and examining how close the predicted result was to the measured result.
The five types of rubber materials are A1, A2, B1, C1, and D1 of company A to company D, and carbon black of company E is used as a filler. And accelerator F were used. The number of times of mixing was adjusted to two times.

Table 1 below shows the names of the rubber materials and the physical property data of the raw materials.
The rubber material is SBR (styrene-butadiene rubber) or butadiene rubber. In Table 1 below, “St” indicates the weight% of styrene based on the whole rubber material. “Cis”, “trans”, and “Vinyl” indicate the ratio (% by weight) of each binding mode (the sum of the values of “cis”, “trans”, and “Vinyl” becomes 100%) . “Tg” indicates a glass transition temperature, “MW” indicates a weight average molecular weight, and “MW / MN” indicates a value indicating a molecular weight distribution.

In Table 1 above, the trade names A1, B1, and C1 have almost the same weight percentage of “St”, the proportions of “cis”, “trans”, and “Vinyl”, and “Tg”, “MW”, and “MW / MN” also shows substantially the same physical property values. On the other hand, in the trade names D1 and A2, the ratios of “cis”, “trans”, and “Vinyl” are almost the same, and “Tg”, “MW”, and “MW / MN” show substantially the same physical property values. However, they are different from the ratios of the bonding modes of the trade names A1, B1, and C1 and the physical property values of “Tg”, “MW”, and “MW / MN”. For example, “Tg” differs by about 50 ° C.
For this reason, it is preferable to make the prediction module learn using the product names A1, B1, and C1 as one identification name. Further, it is preferable that the prediction module learns using the product names D1 and A2 as one identification name.
Whether or not to be combined into one identification landmark can be determined by calculating a correlation coefficient of a composition ratio or a physical property value, and can be integrated as one identification name according to the magnitude of the correlation coefficient.
Also in this example, the correlation coefficient of the three ratios of the binding mode and / or the three physical property values of “Tg”, “MW”, and “MW / MN” is calculated. A1, B1, and C1 can be put together as one identification name, and the product names D1 and A2 can be put together as one identification name.

Tables 2 and 3 below show examples of combinations of raw materials. In the case of producing a vulcanized rubber composition by such a combination of blending, a vulcanized rubber composition was produced as follows.
A compounding agent other than a vulcanizing compounding agent such as a vulcanization accelerator and a rubber material are heated to about 150 ° C. using a 1.7-liter closed Banbury mixer, mixed for 5 minutes, and then released. Cool to obtain a masterbatch. Further, sulfur and a vulcanization accelerator are mixed with the finally obtained master batch using the Banbury mixer to obtain an unvulcanized rubber composition. The obtained unvulcanized rubber composition is press-vulcanized in a predetermined mold at 140 ° C. to 190 ° C. for 10 minutes to 90 minutes to prepare a vulcanized rubber test piece. In the following tables, blank sections mean that there is no corresponding material.

  A prediction module that machine-learned 200 pieces of learning data without pre-processing the rubber material trade names A1, A2, and B1 to D1 was used to obtain a vulcanized rubber composition for each of Formulations 1 to 12 shown in Tables 2 and 3. The physical property data of the object was predicted. Physical property data are rubber hardness, breaking strength, breaking elongation, and tan δ (60 ° C.) of viscoelastic properties. The tan δ (60 ° C.) was determined for a test piece of the vulcanized rubber composition using a viscoelastic spectrometer (manufactured by Toyo Seiki Seisaku-sho, Ltd.) in accordance with JIS K6394: 2007, with an elongation and deformation ratio of 10% ± 2% and a vibration frequency. This is a value measured under the conditions of 20 Hz and a temperature of 60 ° C. The rubber hardness is a value obtained by measuring a test piece of a vulcanized rubber composition at a temperature of 20 ° C. using a durometer type A in accordance with JIS K6253.

On the other hand, the product names A1, B1, and C1 of the rubber material were made into one identification name (product name A1), and the product names D1 and A2 of the rubber material were made one identification name (product name D1) by the pre-processing.
Tables 4 and 5 below show formulations 1 to 12.

  A prediction module that machine-learned 200 learning data obtained by pre-processing the rubber material product names A1, A2, and B1 to D1 was added to each of the pre-processed formulations 1 to 12 shown in Tables 4 and 5. The physical property data of the vulcanized rubber composition was predicted. Physical property data are rubber hardness, breaking strength, breaking elongation, and tan δ (60 ° C.) of viscoelastic properties.

  Table 6 below shows the average value of the ratio of the prediction result of the physical property data without the pre-processing and the prediction result of the physical property data with the pre-processing to the actual measurement result. The closer the ratio is to 1.00, the closer the prediction result is to the actual measurement result.

  From Table 6, the ratio of the breaking strength, breaking elongation, and tan δ (60 ° C.) of the physical property data with the pre-treatment was changed to the ratio of the breaking strength, breaking elongation, and tan δ (60 ° C.) without the pre-treatment. In comparison, it turns out that it is close to 1.00.

(Example of pretreatment of filler)
In order to confirm the effect of the pretreatment using five types of fillers, the prediction accuracy of the physical properties of the vulcanized rubber composition to be predicted with or without the pretreatment was examined. The prediction accuracy was measured by actually measuring the physical properties of the vulcanized rubber composition produced, and examining how close the predicted result was to the measured result.
In the vulcanized rubber composition, the rubber material A1 (see Table 1) was fixed, and the composition of the five types of fillers was changed. As the filler, carbon blacks having brand names E1, E2, F1, G1, and G2 of companies E to G were used, and in addition, ingredients I to N and accelerator F were used.

Table 7 below shows the trade names of the fillers and their raw material property data.
Physical property values 1 to 4 are physical property values which are indicators of carbon black specific surface area and particle diameter such as nitrogen adsorption amount, oil absorption amount and coloring power of carbon black.

In Table 7, the trade names E1, F1, and G1 indicate substantially the same physical property values 1 to 4, and the trade names G2 and E2 indicate the substantially same physical property values 1 to 4. Therefore, it is preferable that the prediction module learns using the product names E1, F1, and G1 as one identification name. Further, it is preferable that the prediction module learns using the product names G2 and E2 as one identification name.
To determine whether to combine them into one, the correlation coefficients of the physical property values 1 to 4 are calculated, and the trade names E1, F1, and G1 are collected as one identification name, and the trade names G2, E2 As one identification name.

Tables 8 and 9 below show examples of combinations of raw material combinations. In the case of producing a vulcanized rubber composition by such a combination of blending, a vulcanized rubber composition was produced as follows.
A compounding agent other than a vulcanizing compounding agent such as a vulcanization accelerator and a rubber material are heated to about 150 ° C. using a 1.7-liter closed Banbury mixer, mixed for 5 minutes, and then released. Cool to obtain a masterbatch. Further, sulfur and a vulcanization accelerator are mixed with the finally obtained master batch using the Banbury mixer to obtain an unvulcanized rubber composition. The obtained unvulcanized rubber composition is press-vulcanized in a predetermined mold at 140 ° C. to 190 ° C. for 10 minutes to 90 minutes to prepare a vulcanized rubber test piece.

  The prediction module which machine-learned the learning data which did not pre-process the filler product names E1, E2, F1, G1, and G2 was used for the vulcanized rubber for each of Formulations 1 to 14 shown in Tables 8 and 9. The physical property data of the composition was predicted. Physical property data are rubber hardness, breaking strength, breaking elongation, and tan δ (60 ° C.) of viscoelastic properties.

On the other hand, the product names E1, F1, and G1 of the filler are one identification name (product name E1), and the product names E2 and G2 of the filler are one identification name (product name G2).
Tables 10 and 11 below show formulations 1 to 14.

  A prediction module that machine-learned learning data obtained by pre-processing the filler product names E1, E2, F1, G1, and G2 was vulcanized for each of the pre-processed formulations 1 to 14 shown in Tables 10 and 11. The physical property data of the rubber composition was predicted. Physical property data are rubber hardness, breaking strength, breaking elongation, and tan δ (60 ° C.) of viscoelastic properties. The tan δ (60 ° C.) was determined for a test piece of the vulcanized rubber composition using a viscoelastic spectrometer (manufactured by Toyo Seiki Seisaku-sho, Ltd.) in accordance with JIS K6394: 2007, with an elongation and deformation ratio of 10% ± 2% and a vibration frequency. This is a value measured under the conditions of 20 Hz and a temperature of 60 ° C. The rubber hardness is a value obtained by measuring a test piece of a vulcanized rubber composition at a temperature of 20 ° C. using a durometer type A in accordance with JIS K6253.

  Table 12 below shows the average of the ratio of the prediction result of the physical property data without pre-processing and the prediction result of the physical property data with pre-processing to the actual measurement result. The closer the ratio is to 1.00, the closer the prediction result is to the actual measurement result.

From Table 12, the ratio of the breaking strength, breaking elongation, and tan δ (60 ° C.) of the physical property data with the pretreatment to the ratio of the breaking strength, breaking elongation, and tan δ (60 ° C.) without the pretreatment are shown. In comparison, it turns out that it is close to 1.00.
From this, the effect of the pretreatment is clear.
Such a pre-processing may be performed by simultaneously performing pre-processing on two raw materials of different groups, such as a rubber material and a filler, and having the prediction module predict the same.

  As described above, when the computer predicts the physical property data of the vulcanized rubber composition, the physical property data (characteristic data) of the vulcanized rubber composition is efficiently and accurately used by using the mixing ratio or the physical property value of each raw material. Can be predicted.

  As described above, the physical property data prediction method and the physical property data prediction apparatus of the present invention have been described in detail. However, the present invention is not limited to the above embodiment, and various improvements and changes may be made without departing from the gist of the present invention. Of course it is good.

Reference Signs List 100 Physical property prediction device 102 Prediction unit 104 Data operation unit 104a Preprocessing unit 104b Input unit

Claims (13)

Physical property data of a vulcanized rubber composition produced from an unvulcanized rubber composition blended by combining a plurality of raw materials set in advance, a physical property data prediction method for making a computer predict,
A plurality of raw materials used for the vulcanized rubber composition are blended at a set blending ratio, and physical property data on a plurality of vulcanized rubber compositions produced by processing under the set processing conditions are used as learning output data. The identification data of each of the raw materials in the vulcanized rubber composition, the compounding ratio of each of the raw materials, and information of each of the processing conditions, using the learning data as learning input data, the physical property data, Machine learning a prediction module in a computer;
The names of constituent raw materials constituting the unvulcanized rubber composition before vulcanization of the predicted vulcanized rubber composition, the mixing ratio of the constituent raw materials, and the processing for producing the predicted vulcanized rubber composition Using processing conditions, the step of causing the prediction module to predict the physical property data of the vulcanized rubber composition to be predicted,
Before the step of causing the prediction module to predict the physical property data of the vulcanized rubber composition to be predicted,
Causing the computer to perform a pre-process of classifying the name of the constituent raw material into one of the identification names used as the learning input data, based on the material characteristics of the constituent raw material;
Causing the computer to input one of the classified identification names to the prediction module as a name corresponding to the constituent raw material for the prediction by the prediction module;
A physical property data prediction method comprising:   The name of the constituent raw material, using raw material physical property data of the constituent raw material as the material property, based on a comparison result of raw material physical property data of the constituent raw material and raw material physical property data of the raw material used for the learning data, The physical property data prediction method according to claim 1, wherein the physical property data is classified into one of the identification names. The raw material property data includes a plurality of different types of raw material property values,
The names of the constituent raw materials are classified into one of the identification names based on a correlation coefficient between the physical property values of the plurality of types of raw materials and the physical property values of the corresponding types of the raw materials used in the learning data. 3. The physical property data prediction method according to claim 2, wherein the method is performed. The raw material used for the learning data and the constituent raw material in the predicted vulcanized rubber composition is a mixture of a plurality of chemical compositions,
The physical property according to any one of claims 1 to 3, wherein, in the pretreatment, a name of the constituent raw material is classified into one of the identification names by using a composition ratio of the chemical composition as the material property. Data prediction method.   The name of the constituent raw material, the composition ratio of the chemical composition in the constituent raw material, the identification name of the identification name based on the correlation coefficient between the composition ratio of the chemical composition in the raw material used for the learning data The physical property data prediction method according to claim 4, wherein the physical property data is classified into one. The constituent raw materials and the raw materials used for the learning data have a main component chemical composition having the composition ratio of 80% or more,
The name of the constituent raw material is such that, among the raw materials used for the learning data, the difference between the composition ratio of the main component chemical composition of the constituent raw material and the composition ratio of the main component chemical composition of the raw material is 5%. The physical property data prediction method according to claim 4, wherein the physical property data is classified into identification names of raw materials falling within the range. The raw material and the constituent raw materials used for the learning data include a material whose shape is specified by a shape parameter,
7. The pre-processing according to claim 1, wherein the name of the constituent raw material is classified into one of the identification names of the raw material used for the learning data, using the shape parameter as the material characteristic. Item 2. The physical property data prediction method according to Item 1.   The physical property data prediction method according to claim 7, wherein the shape parameter includes a shape, a shape size, or an aspect ratio of the shape of the raw material and the constituent raw material.   The name of the constituent raw material is classified into one of the identification names of the raw material in which the value of the shape parameter of the constituent raw material matches within an allowable range the value of the shape parameter of the raw material used in the learning data. 9. The physical property data prediction method according to claim 7, wherein the method is performed. The raw materials and the constituent raw materials used for the learning data include a plurality of synthetic rubbers that are polymers having different bonding modes and ratios of the bonding modes,
The preprocessing, wherein the name of the constituent raw material is classified into one of the identification names of the raw material used for the learning data, using the ratio of the bonding mode as the material characteristic. The physical property data predicting method according to any one of the preceding claims.   The name of the constituent raw material is one of the identification names based on a correlation coefficient between the ratio of the bond mode in the constituent raw material and the ratio of the bond mode in the raw material used for the learning data. The physical property data prediction method according to claim 10, which is classified.   The name of the constituent raw material is classified as an identification name of a raw material in which the correlation coefficient exceeds a predetermined value among the raw materials used for the learning data. Physical property data prediction method. A physical property data prediction device that predicts physical property data of a vulcanized rubber composition produced from an unvulcanized rubber composition that is compounded by combining a plurality of raw materials set in advance,
Output data for learning is obtained by blending a plurality of raw materials used in a vulcanized rubber composition at a set blending ratio and processing them under set processing conditions. And an identification name of each of the raw materials, a blending ratio of each of the raw materials, and information of each of the processing conditions, using a learning data with learning input data as a prediction unit for machine learning the physical property data. ,
The names of constituent raw materials constituting the unvulcanized rubber composition before vulcanization of the predicted vulcanized rubber composition, the mixing ratio of the constituent raw materials, and the processing for producing the predicted vulcanized rubber composition And a processing condition, using a data operation unit that makes the prediction unit predict physical property data of the vulcanized rubber composition to be predicted,
The data operation unit includes:
A preprocessing unit that performs preprocessing of classifying the name of the constituent raw material into one of the identification names used as the learning input data by the prediction unit based on the material characteristics of the constituent raw material;
An input unit configured to input one of the identification names classified by the preprocessing unit to the prediction unit for prediction by the prediction unit, as a name corresponding to the constituent raw material;
A physical property data prediction device comprising:

Images