JP7348489B2 - Physical property data prediction method and device Physical property data prediction device - Google Patents

Description

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

In recent years, technologies have been actively proposed that allow computers to perform machine learning to make various predictions from input data. It is also conceivable that the above technique can be applied to predicting the values of physical property data of a vulcanized rubber composition produced by blending a plurality of rubber materials, fillers, oils, etc. as raw materials.
BACKGROUND ART Conventionally, a vulcanized rubber composition has been prepared as a prototype by blending a plurality of rubber materials, fillers, oils, etc. through trial and error, and physical property data has been experimentally measured. For this reason, a large amount of data has been accumulated that links the formulation information of vulcanized rubber compositions with the values of physical property data. Using this accumulated data, it is possible to use machine learning on a computer to predict the values of physical property data.

For example, using a neural network method, we can learn the mapping relationship between a group of factors and a group of characteristics in experimental data such as design and formulation, and estimate characteristic values from factor conditions. A technique is known that provides a method for efficiently and easily determining the optimal value of factor data that produces the following (Patent Document 1).

Japanese Patent Application Publication No. 2003-58582

In learning the neural network in this technology, all prepared original training data is uniformly read and used for multiple pieces of training data. For example, as the plurality of pieces of learning data, a large number of accumulated experimental data in which information on the blending ratio of raw materials of the vulcanized rubber composition is linked to values of physical property data is used.

The compounding ratio in the accumulated vulcanized rubber composition is not only raw materials such as rubber materials whose compounding ratio is several 10% of the total mass of the vulcanized rubber composition, but also vulcanization accelerators and oil. It also includes raw materials with a blending ratio of less than 1%, such as plasticizers such as. In fact, even raw materials with a blending ratio of less than 1% may make a significant contribution to the value of physical property data. A prediction model using machine learning is expected to make a significant contribution to the value of physical property data and predict the value of physical property data, even for raw materials with a low blending ratio.
However, prediction models that are machine-trained using actual original data as learning data cannot fully meet the above expectations, and are still learning to make a small contribution to the physical property data values of raw materials with low blending ratios. It is often done.

Therefore, the present invention aims to accurately predict the physical property data of a vulcanized rubber composition produced from an unvulcanized rubber composition prepared by combining a plurality of preset raw materials. It is an object of the present invention to provide a physical property data prediction method and a physical property data prediction device that can predict the values of physical property data of a sulfur rubber composition.

One aspect of the present invention is a physical property data prediction method that causes a computer to predict physical property data values of a vulcanized rubber composition produced from an unvulcanized rubber composition blended by combining a plurality of preset raw materials. be. The prediction method is
The values of physical property data regarding a plurality of vulcanized rubber compositions produced by changing the blending ratio of a plurality of raw materials used in a vulcanized rubber composition are used as learning output data, and the raw materials in the vulcanized rubber composition are The values of the physical property data, the information on the blending ratio and a step of causing a predictive model in a computer to machine learn the relationship between the information on the processing conditions;
Input information on the blending ratio of constituent raw materials constituting the unvulcanized rubber composition before vulcanization of the vulcanized rubber composition to be predicted, and information on processing conditions when producing the vulcanized rubber composition to be predicted. and causing the prediction model to predict values of physical property data of the vulcanized rubber composition to be predicted using the received and inputted information on the blending ratio and information on the processing conditions.
The information on the mixing ratio used as the learning input data is the mixing ratio of the raw material in the original learning data prepared in advance, and the mixing ratio of the raw material in the plurality of vulcanized rubber compositions for each raw material. It is a value normalized between the maximum and minimum mixing ratios of
Information on the blending ratio of the constituent raw materials used when predicting the value of physical property data of the vulcanized rubber composition to be predicted is also a value normalized between the maximum blending ratio and the minimum blending ratio.

The plurality of vulcanized rubber compositions have a blending ratio such that the minimum blending ratio of raw materials in the same vulcanized rubber composition in the original learning data is 10% or less of the maximum blending ratio of the raw materials. Preferably, the rubber composition contains at least one vulcanized rubber composition having the following.

The input data for learning and the original data for learning include information on a blending ratio for each name of raw materials,
At least one of the names of the raw materials defined in the learning input data satisfies the condition that the material properties of the raw materials are the same as the names of a plurality of raw materials defined in the original learning data. It is a name that has been integrated into one name based on the determination of ,
At least one of the names of the component raw materials is a name that combines the names of multiple component raw materials into one name based on whether the material properties of the component raw materials satisfy the above conditions, and It is preferable that the blending ratio value in at least one of the names is the sum of the blending ratio values of a plurality of integrated constituent raw materials.

The material properties of the raw material or the constituent raw materials include a plurality of different raw material physical property values,
Preferably, the conditions include that the degree of approximation of the physical property values of the raw materials in the raw materials or the constituent raw materials is equal to or greater than a predetermined value.

The raw materials used for the original learning data and the constituent raw materials in the prediction target vulcanized rubber composition are a mixture of a plurality of chemical compositions,
The material property of the raw material or the constituent raw material is a composition ratio of a chemical composition,
Preferably, the condition includes that the degree of approximation of the composition ratio in the raw material or the constituent raw materials is equal to or greater than a predetermined value.

One aspect of the present invention is computer-generated physical property data that predicts values of physical property data of a vulcanized rubber composition produced from an unvulcanized rubber composition prepared by combining a plurality of preset raw materials. It is a prediction device. The physical property data prediction device is
The values of physical property data regarding a plurality of vulcanized rubber compositions produced by changing the blending ratio of a plurality of raw materials used in a vulcanized rubber composition are used as learning output data, and the raw materials in the vulcanized rubber composition are The values of the physical property data, the information on the blending ratio and a machine learning unit that causes a predictive model to perform machine learning on the relationship between the information on the processing conditions;
Input information on the blending ratio of constituent raw materials constituting the unvulcanized rubber composition before vulcanization of the vulcanized rubber composition to be predicted, and information on processing conditions when producing the vulcanized rubber composition to be predicted. a prediction unit that causes the prediction model to predict values of physical property data of the prediction target vulcanized rubber composition using the received and inputted information on the blending ratio and information on the processing conditions;
The blending ratio of the raw materials in the original learning data prepared in advance is determined for each raw material between the maximum blending ratio and the minimum blending ratio of the blending ratios of the raw materials in the plurality of vulcanized rubber compositions. A standardized value is created as information on the mixing ratio of the learning input data, and a value is further standardized using the input mixing ratio of the constituent raw materials using the maximum mixing ratio and the minimum mixing ratio. a pre-processing unit that creates the information as information on the mixing ratio of the constituent raw materials;
Equipped with

According to the above-described physical property data prediction method and physical property data prediction device, it is possible to have a computer accurately predict values of physical property data of a vulcanized rubber composition.

FIG. 2 is a diagram illustrating an example of a method for predicting physical property data using a computer according to an embodiment. FIG. 1 is a diagram illustrating an example of the main configuration of a physical property data prediction device according to an embodiment. FIG. 3 is a diagram showing an example of the blending ratio of raw materials before standardization used in the physical property data prediction method of one embodiment. FIG. 4 is a diagram showing an example of a result of normalizing the blending ratio shown in FIG. 3; FIG. 2 is a diagram showing an example of the composition of constituent raw materials in a prediction target vulcanized rubber composition used in the physical property data prediction method of one embodiment, and the composition of raw materials used in original data for learning.

Hereinafter, a physical property data prediction method and a physical property data prediction device of one embodiment will be described in detail.

The physical property data prediction method of one embodiment causes a computer to machine learn the association between physical property data values and factor data regarding a plurality of vulcanized rubber compositions, and uses a machine learned computer prediction model to This is a method for predicting the values of physical property data of sulfur rubber compositions.

FIG. 1 is a diagram showing an example of a method for predicting physical property data using a computer. FIG. 2 is a diagram illustrating an example of the main configuration of a physical property data prediction device according to an embodiment that performs a physical property data prediction method. The physical property data prediction device 100 is composed of a computer including a CPU 102 and a memory 104. The physical property data prediction device 100 forms at least a preprocessing unit 106, a machine learning unit 108, and a prediction unit 110 as software modules, which function by activating a program stored in the memory 104. The physical property data prediction device 100 is connected to a display (not shown), and is further connected to input operation devices including a mouse and a keyboard.

For machine learning, the preprocessing unit 106 of the physical property data prediction device 100 calculates the values of the physical property data of the vulcanized rubber composition and the blending ratio of each raw material constituting the vulcanized rubber composition, as shown in FIG. Original data for learning including a plurality of sets of vulcanized rubber compositions is obtained by combining the information on the vulcanized rubber composition and the information on the processing conditions when producing the vulcanized rubber composition (ST10 in FIG. 1). In other words, the original data for learning includes values of physical property data regarding multiple vulcanized rubber compositions produced by changing the blending ratio of multiple raw materials used for vulcanized rubber compositions, and This data includes a large number of sets of information on the blending ratio of each raw material in the vulcanized rubber composition and information on the processing conditions when producing this vulcanized rubber composition. Such original data for learning is stored and held in the memory 104. Alternatively, the original data for learning can be obtained from a data library that has accumulated and saved a large amount of experimental data that includes physical property data values obtained through experiments, in correspondence with information on vulcanized rubber compositions (compounding ratio and processing conditions). It may also be one that has been transferred to the data prediction device 100.

The information on the blending ratio includes the name of each raw material in the vulcanized rubber composition and the value of the blending ratio of each raw material. Processing conditions include, for example, the type of mixing machine that mixes raw materials containing unvulcanized rubber, the temperature during mixing, the mixing rotation speed of the mixing machine, the mixing pressure, and the amount of raw materials input to the mixing machine. or vulcanization conditions including at least one of vulcanization temperature and vulcanization time when vulcanizing the mixed unvulcanized rubber. Physical property data regarding the vulcanized rubber composition include, for example, rubber elasticity, tan δ, specific gravity, Mooney Vis viscosity, Mooney scorch, rheometer measurement results, Lübke JIS hardness, modulus, breaking strength, breaking energy, filler dispersity, rebound modulus, constant It includes at least one of strain test results, wear test results, fatigue test results, and crack growth characteristics.

Next, the preprocessing unit 106 calculates the blending ratio of each raw material in the original data for learning, for each raw material, between the maximum blending ratio and the minimum blending ratio of the blending ratios of raw materials in the plurality of vulcanized rubber compositions. Then, learning data is created that is a set of raw material information, processing condition information, and physical property data including the standardized values (ST12 in FIG. 1).

FIG. 3 is a diagram illustrating an example of the blending ratio of raw materials before standardization used in the physical property data prediction method of one embodiment. FIG. 4 is a diagram showing an example of the result of normalizing the blending ratio shown in FIG. 3. In FIGS. 3 and 4, information on processing conditions and information on values of physical property data are not shown.
In FIGS. 3 and 4, the vulcanized rubber composition is composed of raw materials A to H. Raw material A is, for example, SBR (styrene butadiene rubber), raw material B is, for example, carbon black, raw material C is, for example, zinc white, raw material D is, for example, stearic acid, raw material E is, for example, an antioxidant, raw material F is, for example, process oil, compounding agent G. is, for example, sulfur, and compounding agent H is, for example, a vulcanization accelerator.
In this way, the blending ratio of raw materials for multiple vulcanized rubber compositions is standardized by the maximum blending ratio and minimum blending ratio for each raw material, so the value of the maximum blending ratio is set to 1 by standardization, and the minimum blending ratio is When the value of the ratio is set to 0 by normalization, the value of the blending ratio is set in the range of 0 to 1. Therefore, even if the blending ratio of one vulcanized rubber composition is extremely small compared to the blending ratio of other raw materials, the value will be large.

Next, the machine learning unit 108 uses the values of the physical property data regarding the vulcanized rubber composition as output data for learning, and outputs information on the blending ratio, which is the standardized blending ratio of each raw material in the vulcanized rubber composition, and processing The prediction unit 110 in the computer calculates the relationship between the information on the mixing ratio, the information on the processing conditions, and the value of the physical property data using learning data that uses information on each condition as input data for learning. Machine learning is performed on the prediction model (ST14 in Figure 1). As the machine learning of the predictive model, for example, deep learning using a neural network is used. Additionally, a random forest using a tree structure can be used. As the model, a known model such as a convolutional neural network or a stag autoencoder can be used. In this way, the machine learning of the predictive model is completed.

The preprocessing unit 106 provides information on the blending ratio of constituent raw materials constituting the unvulcanized rubber composition before vulcanization of the vulcanized rubber composition to be predicted, and information on the blending ratio of the constituent raw materials constituting the unvulcanized rubber composition before vulcanization of the vulcanized rubber composition to be predicted, and When information on processing conditions is input, the blending ratio of the raw materials constituting the vulcanized rubber composition to be predicted is standardized according to the standardization process for each raw material described above (ST16 in FIG. 1). That is, the value of the blending ratio of the constituent raw materials of the vulcanized rubber composition to be predicted is also standardized using the above-mentioned maximum blending ratio and the above-mentioned minimum blending ratio. The value of the blending ratio of the constituent raw materials may deviate from the range of the above-mentioned maximum blending ratio and minimum blending ratio. In this case, the value of the normalized blending ratio is greater than 1 and smaller than 0. However, even with such a value, the prediction model can predict the value of the physical property data.

Next, the prediction unit 110 uses the values of the standardized blending ratios of the constituent raw materials of the vulcanized rubber composition to be predicted and the processing conditions for producing the vulcanized rubber composition to be predicted. The prediction model predicts the values of the physical property data of the vulcanized rubber composition (ST18 in FIG. 1).

In this way, when the predictive model performs machine learning, the learning data is not the original learning data itself, but the mixing ratio value used as the learning input data is the mixing ratio of the raw materials in the original data prepared in advance. is a value normalized for each raw material between the maximum blending ratio and the minimum blending ratio of the blending ratios of raw materials in a plurality of vulcanized rubber compositions. Further, the value of the blending ratio of the constituent raw materials used when predicting the value of physical property data of the vulcanized rubber composition to be predicted is also a value normalized using the above-mentioned maximum blending ratio and the above-mentioned minimum blending ratio. Therefore, even if the blending ratio of raw materials that greatly contribute to the value of physical property data is low, by normalizing the blending ratio of raw materials using the same raw material, the value of the blending ratio after specification can be set to a large value, and the physical property Data values can be predicted with high accuracy.

According to one embodiment, for the plurality of vulcanized rubber compositions used in the original data for learning, the minimum value of the blending ratio of raw materials in the same vulcanized rubber composition in the original data for learning is the blending ratio of raw materials. The composition may contain at least one vulcanized rubber composition having a blending ratio of 10% or less, further 1% or less of the maximum value of . Even in such a case, the minimum value of the blending ratio of the raw materials can be set to a large value by standardization, so that the values of the physical property data can be predicted with high accuracy.

In the above embodiment, since the blending ratio values are standardized within the same raw material, it is determined whether or not what is considered to be different raw materials should be considered to be originally the same raw material, and it is determined that they are the same raw material. In this case, it is preferable to combine the raw materials into one. Therefore, according to one embodiment, when the input data for learning and the original data for learning include information on the mixing ratio for each name of raw materials, at least one of the names of raw materials defined in the input data for learning is It is a name that integrates the names of multiple raw materials specified in the original data for use into one name based on the determination of whether or not the material characteristics of the raw materials satisfy the same condition, and at least one of the raw material names It is preferable that one blending ratio value is the sum of blending ratio values corresponding to the names of a plurality of integrated raw materials. In this case, at least one of the names of the constituent raw materials in the vulcanized rubber composition to be predicted may be changed to one name based on whether the material properties of the constituent raw materials satisfy the above-mentioned conditions. It is preferable that the name is an integrated name, and the value of the blending ratio in at least one of the names of the constituent raw materials is the sum of the values of the blending ratio corresponding to the names of the plurality of integrated constituent raw materials.

When the above-mentioned raw material characteristics include multiple types of raw material physical property values with different types, according to one embodiment, the above-mentioned condition of being the same is that the degree of approximation of the raw material physical property values of the raw materials or constituent raw materials is equal to or higher than a predetermined value. It is preferable to include. The fact that the degree of approximation is equal to or greater than a predetermined value includes, for example, that the ratio of raw material physical property values to be compared to determine whether they are the same or not is equal to or greater than a predetermined threshold, and when there are multiple raw material physical property values, This includes that the ratios of the physical property values of raw materials to be compared to determine whether they are the same are all greater than a predetermined threshold, and further includes that the correlation coefficient is greater than or equal to a predetermined threshold in place of the above ratios. .

In addition, when the raw materials used for the original data for learning and the constituent raw materials of the vulcanized rubber composition to be predicted are a mixture of multiple chemical compositions, the raw material characteristics may be the composition ratio of the chemical compositions. . In this case, it is preferable that the above-mentioned condition of being the same includes that the degree of approximation of the composition ratios of the raw materials or constituent raw materials is equal to or greater than a predetermined value. The fact that the degree of approximation is equal to or greater than a predetermined value includes, for example, that the ratio of composition ratios to be compared to determine whether they are the same or not is equal to or greater than a predetermined threshold; This includes that the ratios of the composition ratios to be compared are all equal to or greater than a predetermined threshold value, and further includes that the correlation coefficient is equal to or greater than a predetermined threshold value instead of the ratio.

FIG. 5 is a diagram showing an example of the composition of constituent raw material A in the vulcanized rubber composition to be predicted, and the composition of raw materials B and C used in the original data for learning. At this time, the correlation coefficient of the composition ratio of component raw material A and raw material 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 of component raw material A and raw material 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 ) ^(1/2) }. If the correlation coefficient between the composition ratio of component raw materials A and raw material B is larger than the correlation coefficient between the composition ratios of component raw materials A and raw material C, and the value is greater than or equal to a predetermined threshold, the name of component material A is Name raw material B. By classifying using the degree of approximation in this way, it is possible to perform accurate classification, and in turn, it is possible to accurately predict the values of physical property data of the vulcanized rubber composition.

(Example, comparative example)
The above-mentioned physical property data prediction method was performed using a vulcanized rubber composition as the tread rubber used in tires, and its effects were confirmed.
The blending ratio of the original learning data for tread rubber is the blending ratio of SBR as a rubber raw material, the blending ratio of one type of carbon black as a filler raw material, zinc white, stearic acid, anti-aging agent, process oil, sulfur, Contains the blending ratio of accelerator.
The original data for learning includes 1000 sets of information on the blending ratio of the vulcanized rubber composition, information on processing conditions, and values of physical property data.
The physical property data are the elastic modulus at 2% elongation at 20° C. and tan δ (loss tangent) at 20° C. and a frequency of 20 Hz.

As an example, learning input data in which blending ratio values were standardized was created, and a predictive model was subjected to machine learning. Information on the blending ratio of the two vulcanized rubber compositions and information on processing conditions were given to a machine-learned prediction model to calculate predicted values of physical property data.
On the other hand, as a comparative example, the original data for learning was used as input data for learning without standardizing the blending ratio, and the prediction model was subjected to machine learning. Information on the blending ratio and processing conditions of the two vulcanized rubber compositions used in the examples to be predicted were given to a machine-learned prediction model to calculate predicted values of physical property data. The predicted value of the physical property data was evaluated based on the ratio of the predicted value to the experimental value of the physical property data obtained by experiment. Since the physical property data includes the elastic modulus and tan δ, the evaluation value was the ratio of the elastic modulus and tan δ, whichever had a value significantly different from 1.

The evaluation values of the examples using the above two predicted vulcanized rubber compositions were 0.90 and 0.91. On the other hand, the evaluation values of the comparative example using the above two predicted vulcanized rubber compositions were 0.85 and 0.88.
This shows that the above physical property data prediction method can accurately predict the values of the physical property data of the vulcanized rubber composition.

Although the physical property data prediction method and the physical property data prediction device of the present invention have been described in detail above, the present invention is not limited to the above embodiments, and various improvements and changes may be made without departing from the spirit of the present invention. Of course it's good.

100 Physical property data prediction device 102 CPU
104 Memory 106 Preprocessing unit 108 Machine learning unit 110 Prediction unit

Claims (6)

A physical property data prediction method that causes a computer to predict physical property data values of a vulcanized rubber composition produced from an unvulcanized rubber composition blended by combining a plurality of preset raw materials, the method comprising:
The values of physical property data regarding a plurality of vulcanized rubber compositions produced by changing the blending ratio of a plurality of raw materials used in a vulcanized rubber composition are used as learning output data, and the raw materials in the vulcanized rubber composition are The values of the physical property data, the information on the blending ratio and a step of causing a predictive model in a computer to machine learn the relationship between the information on the processing conditions;
Input information on the blending ratio of constituent raw materials constituting the unvulcanized rubber composition before vulcanization of the vulcanized rubber composition to be predicted, and information on processing conditions when producing the vulcanized rubber composition to be predicted. a step of causing the prediction model to predict values of physical property data of the vulcanized rubber composition to be predicted using the received and inputted information on the blending ratio and information on the processing conditions,
The value of the blending ratio used as the input data for learning is the blending ratio of the raw material in the original learning data prepared in advance, for each raw material, the blending ratio of the raw material in the plurality of vulcanized rubber compositions. It is a value normalized between the maximum and minimum mixing ratios of
The value of the blending ratio of the constituent raw materials used when predicting the value of physical property data of the vulcanized rubber composition to be predicted is also a value normalized using the maximum blending ratio and the minimum blending ratio. Characteristic physical property data prediction method. The plurality of vulcanized rubber compositions have a blending ratio such that the minimum blending ratio of raw materials in the same vulcanized rubber composition in the original learning data is 10% or less of the maximum blending ratio of the raw materials. The physical property data prediction method according to claim 1, comprising at least one vulcanized rubber composition having the following. The input data for learning and the original data for learning include information on a blending ratio for each name of raw materials,
At least one of the names of the raw materials defined in the learning input data satisfies the condition that the material properties of the raw materials are the same as the names of a plurality of raw materials defined in the original learning data. It is a name that has been integrated into one name based on the determination of ,
At least one of the names of the component raw materials is a name that combines the names of multiple component raw materials into one name based on whether the material properties of the component raw materials satisfy the above conditions, and 3. The physical property data prediction method according to claim 1, wherein the blending ratio value for the at least one name is a sum of blending ratio values corresponding to names of a plurality of integrated constituent raw materials. The material properties of the raw material or the constituent raw materials include a plurality of different raw material physical property values,
4. The physical property data prediction method according to claim 3, wherein the condition includes that the degree of approximation of the raw material physical property values in the raw material or the constituent raw materials is a predetermined value or more. The raw materials used for the original learning data and the constituent raw materials in the prediction target vulcanized rubber composition are a mixture of a plurality of chemical compositions,
The material property of the raw material or the constituent raw material is a composition ratio of a chemical composition,
5. The physical property data prediction method according to claim 3, wherein the condition includes that the degree of approximation of the composition ratio in the raw material or the constituent raw materials is greater than or equal to a predetermined value. A physical property data prediction device configured with a computer that predicts physical property data values of a vulcanized rubber composition produced from an unvulcanized rubber composition compounded by combining a plurality of preset raw materials,
The values of physical property data regarding a plurality of vulcanized rubber compositions produced by changing the blending ratio of a plurality of raw materials used in a vulcanized rubber composition are used as learning output data, and the raw materials in the vulcanized rubber composition are The values of the physical property data, the information on the blending ratio and a machine learning unit that causes a predictive model to perform machine learning on the relationship between the information on the processing conditions;
Input information on the blending ratio of constituent raw materials constituting the unvulcanized rubber composition before vulcanization of the vulcanized rubber composition to be predicted, and information on processing conditions when producing the vulcanized rubber composition to be predicted. a prediction unit that causes the prediction model to predict values of physical property data of the prediction target vulcanized rubber composition using the received and inputted information on the blending ratio and information on the processing conditions;
The blending ratio of the raw materials in the original learning data prepared in advance is determined for each raw material between the maximum blending ratio and the minimum blending ratio of the blending ratios of the raw materials in the plurality of vulcanized rubber compositions. A standardized value is created as information on the mixing ratio of the learning input data, and a value is further standardized using the input mixing ratio of the constituent raw materials using the maximum mixing ratio and the minimum mixing ratio. a pre-processing unit that creates the information as information on the mixing ratio of the constituent raw materials;
A physical property data prediction device comprising:

Images