Physical property data prediction method and device, and factor data prediction method and device

Abstract

To precisely perform prediction when allowing a computer to predict values of physical property data of a vulcanized rubber composition produced by vulcanizing an unvulcanized rubber composition obtained by mixing multiple raw materials.SOLUTION: A physical property data prediction method includes the steps of: allowing a prediction model within the computer to perform machine learning by using learning data with values of physical property data regarding multiple vulcanized rubber compositions as learning output data and with information on a blend ratio of each of raw materials and information on processing conditions as learning input data; and allowing a prediction model to predict values of physical property data of a predicted vulcanized rubber composition by using information on a blend ratio of raw materials for constituting a predicted vulcanized rubber composition and information on processing conditions of a predicted vulcanized rubber composition. Information on processing conditions used for machine learning includes information on a heat history that a raw material receives in processing of the raw material for producing a vulcanized rubber composition.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a method and apparatus for predicting the values ​​of physical property data or factor data of a vulcanized rubber composition. [Background technology]

[0002] In recent years, there have been many proposals for technologies that use computer machine learning to make various predictions from input data. It is conceivable that the above technologies can be applied to predict the values ​​of physical property data of vulcanized rubber compositions produced by blending multiple rubber materials, fillers, oils, etc. as raw materials. Conventionally, vulcanized rubber compositions have been produced by blending multiple rubber materials, fillers, oils, etc. through trial and error, and their physical property data has been measured. As a result, a large amount of data linking the blending information of vulcanized rubber compositions with the values ​​of their physical property data has been accumulated. By utilizing this accumulated data and training a computer to perform machine learning, it is possible to predict the values ​​of the physical property data of vulcanized rubber compositions produced with new blends of raw materials.

[0003] For example, a method is known in which neural network techniques are used to learn the mapping relationship between a group of factors and a group of characteristics in experimental data such as design and composition, estimate characteristic values ​​from factor conditions, and efficiently and easily find the optimal value of the factor data that creates any characteristic data (Patent Document 1). [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Laid-Open No. 2003-58582 Summary of the Invention [Problem to be solved by the invention]

[0005] The values ​​of physical property data of vulcanized rubber compositions are significantly affected not only by the blending ratio of raw materials but also by the conditions (processing conditions) used when mixing and vulcanizing the raw materials. Therefore, it is believed that there is room for improving prediction accuracy by having a computer learn data that links blending information, physical property data values, and processing condition information. However, processing conditions include a wide variety of conditions, and the impact of each condition on the physical property data of vulcanized rubber compositions varies. Furthermore, it is difficult to preset parameters whose measured values ​​change over time, such as the temperature of rubber materials during processing, as processing conditions.

[0006] Therefore, an object of the present invention is to provide a physical property data prediction method and device, as well as a factor data prediction method and device, which are capable of efficiently performing machine learning by a computer and accurately predicting the physical property data values ​​or factor data of a vulcanized rubber composition produced by vulcanizing an unvulcanized rubber composition obtained by mixing a plurality of predetermined raw materials. [Means for solving the problem]

[0007] One aspect of the present invention is a physical property data prediction method, which includes the steps of: A physical property data prediction method for causing a computer to predict values ​​of physical property data of a vulcanized rubber composition produced by vulcanizing an unvulcanized rubber composition obtained by mixing a plurality of predetermined raw materials, comprising: a step of using learning data in which values ​​of physical property data relating to a plurality of vulcanized rubber compositions prepared using a plurality of raw materials are used as learning output data, and information on the compounding ratios of the respective raw materials in the vulcanized rubber compositions and information on processing conditions for preparing the vulcanized rubber compositions are used as learning input data to perform machine learning on a prediction model in a computer to determine the relationship between the physical property data, the compounding ratios, and the processing conditions; and a step of having the prediction model predict values ​​of physical property data of the prediction target vulcanized rubber composition using information on the compounding ratios of constituent raw materials constituting an unvulcanized rubber composition before vulcanization of the prediction target vulcanized rubber composition and information on processing conditions for producing the prediction target vulcanized rubber composition, The information on the processing conditions used in the machine learning is characterized by including information on the thermal history of the raw materials when the raw materials are processed to produce the vulcanized rubber composition.

[0008] Before the step of performing machine learning, A step of causing a computer to perform preprocessing to calculate a value of the thermal history of the raw material; It is preferable that the method further comprises the step of causing a computer to input the calculated value of the thermal history into the prediction model as information on the thermal history.

[0009] The thermal history value is preferably a value calculated using temperature data measured over time during processing of the raw material.

[0010] The thermal history value is preferably a value obtained based on the reaction rate of the reaction between the raw materials that occurs during processing of the raw materials.

[0011] The unvulcanized rubber composition is obtained through a plurality of mixing steps, The information on the thermal history is preferably the sum of the values ​​of the thermal history that the raw materials have received in each of the mixing steps.

[0012] It is preferable that the information on the processing conditions used in the machine learning further includes information on the power consumed by a mixing device that mixes the raw materials to obtain the unvulcanized rubber composition.

[0013] It is preferable that the learning data includes 5000 or more sets each including the information on the blending ratio, the information on the processing conditions, and the values ​​of the physical property data.

[0014] In the machine learning step, it is preferable that information on the test conditions performed on the vulcanized rubber composition when obtaining the values ​​of the physical property data is further used as the learning input data, and the values ​​of the physical property data of the vulcanized rubber composition obtained by performing tests according to the test conditions are used as the learning output data for the values ​​of the physical property data, and that machine learning is performed on the predictive model.

[0015] Another aspect of the present invention is a physical property data prediction device, which comprises: A physical property data prediction device configured by a computer that predicts values ​​of physical property data of a vulcanized rubber composition produced by vulcanizing an unvulcanized rubber composition obtained by mixing a plurality of predetermined raw materials, a machine learning unit that uses learning data in which values ​​of physical property data relating to a plurality of vulcanized rubber compositions prepared using a plurality of raw materials are used as learning output data, and information on the compounding ratios of the respective raw materials in the vulcanized rubber compositions and information on processing conditions for preparing the vulcanized rubber compositions are used as learning input data, and that performs machine learning on a prediction model in a computer to determine the relationship between the physical property data, the compounding ratios, and the processing conditions; a prediction unit having a prediction model that predicts values ​​of physical property data of the prediction target vulcanized rubber composition using information on the compounding ratios of constituent raw materials that constitute an unvulcanized rubber composition before vulcanization of the prediction target vulcanized rubber composition and information on processing conditions for producing the prediction target vulcanized rubber composition, The information on the processing conditions used in the machine learning is characterized by including information on the thermal history of the raw materials when the raw materials are processed to produce the vulcanized rubber composition.

[0016] Another aspect of the present invention is a factor data prediction method, the factor data prediction method comprising: A factor data prediction method for causing a computer to predict at least one of information on blending ratios of raw materials and information on processing conditions in a vulcanized rubber composition produced by vulcanizing an unvulcanized rubber composition obtained by mixing a plurality of predetermined raw materials, the method comprising: a step of performing machine learning on a predictive model in a computer to determine the relationship between the compounding ratios and the processing conditions and the physical property data, using learning data in which information on the compounding ratios of each of the plurality of raw materials in the vulcanized rubber composition and information on processing conditions for producing the vulcanized rubber composition are used as learning output data and values ​​of physical property data related to the plurality of vulcanized rubber compositions produced using the raw materials are used as learning input data; and using values ​​of physical property data of a vulcanized rubber composition to be predicted, the prediction model predicts at least one of information on the blending ratio of constituent raw materials constituting an unvulcanized rubber composition before vulcanization of the vulcanized rubber composition to be predicted, and information on processing conditions for producing the vulcanized rubber composition to be predicted, The information on the processing conditions used in the machine learning is characterized by including information on the thermal history of the raw materials when the raw materials are processed to produce the vulcanized rubber composition.

[0017] Another aspect of the present invention is a factor data prediction device, comprising: A factor data prediction device configured by a computer that predicts at least one of information on blending ratios of raw materials and information on processing conditions in a vulcanized rubber composition by vulcanizing an unvulcanized rubber composition obtained by mixing a plurality of predetermined raw materials, a machine learning unit that uses learning data in which information on the compounding ratios of each of a plurality of raw materials in the vulcanized rubber composition and information on processing conditions for producing the vulcanized rubber composition are used as learning output data, and values ​​of physical property data related to a plurality of vulcanized rubber compositions produced using the raw materials are used as learning input data, and that performs machine learning on a prediction model in a computer to determine the relationship between the compounding ratios, the processing conditions, and the physical property data; a prediction unit having a prediction model that predicts, using values ​​of physical property data of a prediction target vulcanized rubber composition, at least one of information on the blending ratio of constituent raw materials constituting an unvulcanized rubber composition before vulcanization of the prediction target vulcanized rubber composition and information on processing conditions for producing the prediction target vulcanized rubber composition, The information on the processing conditions used in the machine learning is characterized by including information on the thermal history of the raw materials when the raw materials are processed to produce the vulcanized rubber composition. [Brief explanation of the drawings]

[0018] [Figure 1] FIG. 1 is a diagram illustrating an example of a physical property data prediction method using a computer according to an embodiment. [Figure 2] FIG. 1 is a diagram illustrating an example of a main configuration of a physical property data prediction device according to an embodiment. [Figure 3] FIG. 10 is a diagram illustrating an example of a factor data prediction method performed by a computer according to an embodiment. [Figure 4] FIG. 1 is a diagram illustrating an example of a main configuration of a factor data prediction device according to an embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0019] (Physical property data prediction method and physical property data prediction device) A physical property data prediction method and a physical property data prediction device according to an embodiment will be described in detail below.

[0020] One embodiment of the physical property data prediction method involves having a computer learn the association between physical property data values ​​and factor data for a plurality of vulcanized rubber compositions through machine learning, and then using the machine-learned computer prediction model to predict the physical property data values ​​of the vulcanized rubber composition to be predicted. FIG. 1 is a diagram illustrating an example of a physical property data prediction method by 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 the physical property data prediction method. The physical property data prediction device 100 is configured as a computer including a CPU 102 and / or a GPU, and a memory 104. The physical property data prediction device 100 includes at least a preprocessing unit 106, a machine learning unit 108, and a prediction unit 110, which are implemented as software modules 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.

[0021] For machine learning, the preprocessing unit 106 acquires, as original data, information on the compounding ratios of raw materials, processing conditions for processing the raw materials to produce a vulcanized rubber composition, and physical property data of the produced vulcanized rubber composition, as shown in Fig. 1 (ST10 in Fig. 1). The compounding ratios, processing conditions, and physical property data are stored in memory 104. The compounding ratios, processing conditions, and physical property data may be transferred to the physical property data prediction device 100 from a data library that stores and stores a large amount of experimental data including values ​​of physical property data obtained by experiments, in association with information on the vulcanized rubber composition (compounding ratios and processing conditions).

[0022] The information on the compounding ratio includes the name of each raw material in the vulcanized rubber composition and the compounding ratio of each raw material.

[0023] The processing conditions include information on the thermal history of the raw materials during processing (at least one of mixing and vulcanization) of the raw materials to produce a vulcanized rubber composition. Preferably, the pre-processing unit 106 performs pre-processing to calculate the value of the thermal history before acquiring the processing conditions in step ST10. The value of the thermal history is calculated, for example, as described below. The processing conditions also include, as set conditions previously set before processing the raw materials, mixing conditions including at least one of the type of mixing machine that mixes the raw materials including the unvulcanized rubber, the temperature during mixing, the mixing rotation speed of the mixing machine, the mixing pressure, and the total amount of raw materials fed into the mixing machine, or vulcanization conditions including at least one of the vulcanization temperature and vulcanization time when vulcanizing the mixed unvulcanized rubber.

[0024] The physical property data of the vulcanized rubber composition includes, for example, at least one of rubber elasticity, tan δ, specific gravity, Mooney viscosity, Mooney scorch, rheometer measurement results, JIS hardness, modulus, breaking strength, breaking energy, filler dispersion, rebound elasticity coefficient, viscoelasticity test results, abrasion test results, fatigue test results, and crack growth characteristics.

[0025] Thereafter, the machine learning unit 108 uses the values ​​of the physical property data related to the vulcanized rubber composition as learning output data, and uses the learning data, which includes information on the compounding ratios of each raw material in the vulcanized rubber composition and information on each processing condition as learning input data, to machine-learn the relationship between the information on the compounding ratios and processing conditions and the values ​​of the physical property data in a prediction model in the prediction unit 110 in the computer (FIG. 1 ST12). Deep learning using a neural network, for example, can be used for machine learning the prediction model. Random forests using a tree structure can also be used. Well-known models such as convolutional neural networks and stag- ed autoencoders can also be used for the model. Thus, machine learning of the prediction model is completed.

[0026] Next, the prediction unit 110 uses information on the compounding ratios of the constituent raw materials constituting the unvulcanized rubber composition before vulcanization of the prediction target vulcanized rubber composition and the processing conditions for producing the prediction target vulcanized rubber composition to have the prediction model predict the physical property data of the prediction target vulcanized rubber composition (FIG. 1 ST14). The processing conditions include information on the thermal history of the constituent raw materials during processing (at least one of mixing and vulcanization) for producing the prediction target vulcanized rubber composition. Furthermore, the information on the processing conditions preferably includes conditions similar to the above-mentioned set conditions.

[0027] In this way, the learning data used by the prediction model for machine learning includes information on the thermal history of the raw materials during processing. The thermal history is a parameter indicating the cumulative heat generated in the raw materials during processing and represents the progress (amount of reaction) of the reaction between the raw materials. The progress of the reaction between the raw materials has a significant impact on the physical property data of the vulcanized rubber composition. Therefore, the prediction unit that has undergone machine learning can accurately predict the values ​​of the physical property data of the vulcanized rubber composition to be predicted. Furthermore, since information on the thermal history can be easily input as learning data into the prediction unit compared to waveform data obtained by measuring over time during processing, machine learning by the prediction unit can be performed efficiently. In other words, according to this embodiment, the prediction unit can efficiently perform machine learning and accurately predict the values ​​of the physical property data of the vulcanized rubber composition to be predicted.

[0028] The thermal history value is preferably calculated using temperature data measured over time during the processing of the raw materials. Such thermal history values ​​reflect the cumulative heat generated in the raw materials during processing, thereby contributing to improved prediction accuracy of the physical property data of the vulcanized rubber composition. Temperature data during mixing is, for example, time-lapse data (waveform data) of the temperatures of the unvulcanized rubber and compounding ingredients contained in the raw materials and mixed together, or the temperature inside the chamber of the mixing device that mixes the unvulcanized rubber and compounding ingredients. Temperature data during vulcanization is, for example, time-lapse data (waveform data) of the temperature of the unvulcanized rubber composition or the temperature of the vulcanization mold. The thermal history value is calculated as the integral of the temperatures measured during processing. For the thermal history during mixing, the integral spans from the start of the temperature rise due to mixing to the point at which the temperature reaches the end temperature (the point at which the material is released from the mixing device). For the thermal history during vulcanization, the integral spans from the start to the end of the vulcanization process using the vulcanization mold. When these time-lapse data (waveform data) include data for multiple batches, the preprocessing unit 106 extracts data for one batch.

[0029] Furthermore, it is also preferable that the thermal history value is calculated based on the reaction rate of the reaction between raw materials occurring during processing of the raw materials, instead of using temperature data. Such a thermal history value reflects the progress (amount of reaction) of the reaction between the raw materials, thereby contributing to improved accuracy in predicting the values ​​of the physical property data of the vulcanized rubber composition. The thermal history value is calculated, for example, using an integral rate equation or the method for calculating the thermal history amount (HHS) described in JP 2012-111878 A. Examples of reactions between raw materials include the reaction between silica and a silane coupling agent during mixing, and the reaction between unvulcanized rubber and sulfur (including the sulfur in the silane coupling agent) during vulcanization.

[0030] The thermal history information used in the learning data and the thermal history information input to the prediction model for predicting the values ​​of physical property data are preferably separated into information during mixing and information during vulcanization. Because the rate constants of the reactions occurring during mixing and the reactions occurring during vulcanization are different, separating the thermal history information during mixing and vulcanization increases the effect of improving the prediction accuracy of the values ​​of physical property data of the vulcanized rubber composition.

[0031] On the other hand, when an unvulcanized rubber composition is obtained through multiple mixing processes, the thermal history information used in the training data and the thermal history information input into the prediction model for predicting the values ​​of physical property data are preferably the sum of the values ​​of the thermal history of the raw materials in the mixing processes. That is, it is preferable that the information on the thermal history during mixing is consolidated into one. When preparing one batch of an unvulcanized rubber composition, the raw materials may be mixed and then released after mixing repeatedly. In this way, when an unvulcanized rubber composition is obtained through multiple mixing processes, by consolidating the information on the thermal history during mixing into one, the number of parameters required for learning or prediction can be reduced and the calculation speed can be increased.

[0032] The processing condition information used in machine learning preferably includes, in addition to thermal history information, information on the power consumed by a mixer that mixes raw materials to obtain an unvulcanized rubber composition. For example, the power information is calculated using time-series data (waveform data) of power consumption values ​​measured over time during mixing of the raw materials, and is calculated as the integral of the power values ​​measured during mixing. This value corresponds to the rotational speed of the rotor of the mixer that mixes the raw materials, and the rotational speed of the rotor corresponds to the progress of reaction and dispersion between the raw materials, thereby contributing to improved prediction accuracy of the physical property data values ​​of the vulcanized rubber composition. In this case, the processing condition information input into the prediction model for predicting the physical property data values ​​preferably includes, in addition to the thermal history information, information similar to the power consumption information.

[0033] From the viewpoint of improving the prediction accuracy of the physical property data values ​​of the vulcanized rubber composition by the prediction model, the training data preferably includes 5,000 or more sets each including information on the compounding ratios, information on processing conditions, and physical property data values, for example, 10,000 to 1,000,000 sets or 20,000 to 500,000 sets. Training data including such a large number of data sets can be preferably obtained by machine learning using deep learning with a neural network.

[0034] In the machine learning step, it is preferable to use information on the test conditions used for the vulcanized rubber composition when obtaining the physical property data as input data for learning, and use the physical property data values ​​of the vulcanized rubber composition obtained by performing tests under the test conditions as output data for learning to perform machine learning on the prediction model. Performing such machine learning allows the prediction model to output physical property data values ​​corresponding to the input test conditions, thereby improving the accuracy of predicting the physical property data values ​​under those test conditions. Without such machine learning, inputting untrained test conditions into the prediction model will not yield appropriate physical property data values. Examples of test conditions include varying the aging temperature and aging time in an aging test of a vulcanized rubber composition to examine the aging characteristics, such as the tensile strength, of the vulcanized rubber composition. Another example of test conditions is varying the frequency in a test to measure tan δ.

[0035] (Factor data prediction method and factor data prediction device) Next, a factor data prediction method and a factor data prediction device according to an embodiment will be described. In one embodiment, the factor data prediction method is a method of having a computer learn associations between factor data and values ​​of physical property data for a plurality of vulcanized rubber compositions through machine learning, and predicting factor data for a vulcanized rubber composition to be predicted using a prediction model learned by the computer through machine learning. The associations are the same as the associations between factor data and values ​​of physical property data for a plurality of vulcanized rubber compositions obtained by the machine learning of the physical property data prediction method and the physical property data prediction device. FIG. 3 is a diagram illustrating an example of a factor data prediction method by a computer. The factor data prediction method illustrated in FIG. 3 includes steps ST20, ST22, and ST24. FIG. 4 is a diagram illustrating an example of the main configuration of a factor data prediction device according to an embodiment that performs the factor data prediction method. The factor data prediction device 200 is configured by a computer including a CPU 202 and / or a GPU, and a memory 204. The factor data prediction device 200 includes at least a preprocessing unit 206, a machine learning unit 208, and a prediction unit 210, which are implemented as software modules by activating a program stored in the memory 204. The factor data prediction device 200 is connected to a display (not shown) and is further connected to input operation devices including a mouse and a keyboard. The factor data prediction device 200 is configured similarly to the physical property data prediction device 100 except for the following points: (1) information on the blending ratios and information on the processing conditions are used as learning output data instead of the values ​​of the physical property data, and the values ​​of the physical property data are used as learning input data instead of the values ​​of the blending ratios and information on the processing conditions; (2) instead of predicting the values ​​of the physical property data using the information on the blending ratios and information on the processing conditions, at least one of the information on the blending ratios and information on the processing conditions is predicted using the values ​​of the physical property data; and (3) instead of using information on the test conditions as learning input data and the values ​​of the physical property data obtained according to the test conditions as learning output data, the information on the test conditions is used as learning output data and the values ​​of the physical property data obtained according to the test conditions are used as learning input data. The factor data prediction method is configured similarly to the above-mentioned physical property data prediction method except for the points (1) to (3). An operator can operate the factor data prediction device 200 to set whether to predict the blending ratios or the processing conditions, whether to predict both, or whether to additionally predict the test conditions. According to this embodiment, at least one of the compounding ratios and processing conditions can be predicted as optimal factor data information for the constituent raw materials of the vulcanized rubber composition to be predicted from the values ​​of the physical property data of the vulcanized rubber composition to be predicted. In this embodiment, the learning data used when the prediction model performs machine learning includes information on the thermal history of the raw materials during processing, so that at least one of the compounding ratios and processing conditions information for the vulcanized rubber composition to be predicted can be predicted with high accuracy while efficiently performing machine learning by the prediction unit.

[0036] (Examples and Comparative Examples) The tread rubber used in tires was used as a vulcanized rubber composition, and the above-mentioned physical property data prediction method was carried out to confirm its effect. The compounding ratios of the original data for the tread rubber include: as rubber materials, compounding ratios of rubber materials containing at least butadiene rubber, styrene butadiene rubber, and natural rubber; as filler materials, compounding ratios of filler materials containing at least 40 types of carbon black with different specific surface areas and 10 types of silica with different specific surface areas; as silane coupling agents, compounding ratios of three types of silane coupling agents; as vulcanization accelerators, compounding ratios of 10 types of vulcanization accelerators; and as plasticizers, compounding ratios of 10 types of plasticizers containing at least oil and resin. The original data includes 30,000 sets of information on the compounding ratio and processing conditions of the vulcanized rubber composition and values ​​of the physical property data. The processing conditions include the type of mixing machine, vulcanization temperature and vulcanization time as set conditions. The physical property data includes tan δ (60°C) at a frequency of 20 Hz and rubber hardness measured with a durometer (in accordance with JIS K6253-3:2012, Type A, 20°C).

[0037] In Comparative Example 1, the original data was directly provided to the prediction model as learning data, and the relationship between the information on the blending ratio and processing conditions and the values ​​of the physical property data was machine-learned. In Comparative Example 2, in addition to the set conditions in the original data, data including the average temperature inside the chamber of the Banbury mixer measured while mixing the raw materials was used as learning data, and the prediction model was given machine learning to learn the relationship between the information on the blending ratio and processing conditions and the values ​​of the physical property data. On the other hand, in the examples, in addition to the set conditions as processing conditions in the original data, data further including the value of thermal history was provided as learning data to the prediction model, and the relationship between the information on the blending ratio and processing conditions and the value of the physical property data was machine-learned. The value of thermal history was calculated using the total amount of heat given for the reaction between silica and the silane coupling agent during mixing of the raw materials, according to the method described in JP 2012-111878 A. Deep learning using a neural network was used for machine learning of the prediction models in the comparative examples and examples.

[0038] The prediction models created in the comparative examples and examples were given information on the compounding ratios of the constituent raw materials constituting the unvulcanized rubber composition before vulcanization of the vulcanized rubber composition to be predicted, and information on the processing conditions for producing the vulcanized rubber composition to be predicted, to predict the values ​​of the physical property data (tan δ, rubber hardness) of the vulcanized rubber composition to be predicted. In each of the comparative examples 1 and 2 and the examples, the processing conditions given to the prediction models were the same as the processing conditions used for the training data. Meanwhile, vulcanized rubber compositions to be predicted were actually prepared, and the physical property data of the vulcanized rubber compositions to be predicted were actually measured to obtain actual measured values. There were three types of vulcanized rubber compositions to be predicted (rubber compositions A to C). In the tables below, the actual measured value is set to 100, and the predicted values ​​by the prediction model are shown as indices. The "average values" in Tables 1 to 3 below are simple average values ​​of the indices of the physical property data of rubber compositions A to C.

[0039] [Table 1]

[0040] [Table 2]

[0041] [Table 3]

[0042] It can be seen from Tables 1 to 3 that the "average values" of the Examples are closer to 100 than the "average values" of Comparative Examples 1 and 2. Therefore, it can be said that the physical property data prediction method of the present embodiment can predict the values ​​of the physical property data of a vulcanized rubber composition with higher accuracy than conventional methods.

[0043] The physical property data prediction method and physical property data prediction device, and the factor data prediction method and factor data prediction device of the present invention have been described above. However, the present invention is not limited to the above-described embodiments, and various improvements and modifications may be made without departing from the spirit and scope of the present invention. [Explanation of symbols]

[0044] 100 Physical property data prediction device 102, 204 CPU 104, 204 memory 106, 206 Pretreatment section 108, 208 Machine Learning Department 110, 210 Forecasting Department 200 Factor Data Prediction Device

Claims

1. A physical property data prediction method for causing a computer to predict values ​​of physical property data of a vulcanized rubber composition produced by vulcanizing an unvulcanized rubber composition obtained by mixing a plurality of predetermined raw materials, comprising: a step of using learning data in which values ​​of physical property data relating to a plurality of vulcanized rubber compositions prepared using a plurality of raw materials are used as learning output data, and information on the compounding ratios of the respective raw materials in the vulcanized rubber compositions and information on processing conditions for preparing the vulcanized rubber compositions are used as learning input data to perform machine learning on a prediction model in a computer to determine the relationship between the physical property data, the compounding ratios, and the processing conditions; and a step of having the prediction model predict values ​​of physical property data of the prediction target vulcanized rubber composition using information on the compounding ratios of constituent raw materials constituting an unvulcanized rubber composition before vulcanization of the prediction target vulcanized rubber composition and information on processing conditions for producing the prediction target vulcanized rubber composition, A physical property data prediction method characterized in that the information on processing conditions used in the machine learning includes information on the thermal history that the raw materials have undergone during processing of the raw materials to produce the vulcanized rubber composition, and indicates the cumulative heat generated in the raw materials during processing.

2. Before the step of performing machine learning, A step of causing a computer to perform preprocessing to calculate a value of the thermal history of the raw material; The physical property data prediction method according to claim 1 , further comprising the step of causing a computer to input the calculated value of the thermal history into the prediction model as information on the thermal history.

3. 3. The physical property data prediction method according to claim 2, wherein the thermal history value is a value calculated using temperature data measured over time during processing of the raw materials, and the temperature data includes temperature data during mixing of the raw materials and temperature data during vulcanization of the unvulcanized rubber composition.

4. The physical property data prediction method according to claim 2 , wherein the value of the thermal history is a value obtained based on a reaction rate of a reaction between the raw materials that occurs during processing of the raw materials.

5. The unvulcanized rubber composition is obtained through a plurality of mixing steps, The physical property data prediction method according to claim 1 , wherein the information on the thermal history is a sum of values ​​of the thermal history of the raw materials in each of the mixing steps.

6. 6. The physical property data prediction method according to claim 1, wherein the information on the processing conditions used in the machine learning further includes information on the power consumed by a mixing device that mixes the raw materials to obtain the unvulcanized rubber composition.

7. The physical property data prediction method according to claim 1 , wherein the learning data includes 5,000 or more sets each including information on the blending ratio, information on the processing conditions, and values ​​of the physical property data.

8. 8. The physical property data prediction method according to claim 1, wherein in the machine learning step, information on the test conditions performed on the vulcanized rubber composition when obtaining the values ​​of the physical property data is further used as the learning input data, and the physical property data values ​​of the vulcanized rubber composition obtained by performing tests according to the test conditions are used as the learning output data for the values ​​of the physical property data, and machine learning is performed on the prediction model.

9. A physical property data prediction method as described in claim 1 or 2, wherein the information on the processing conditions for producing the vulcanized rubber composition includes the vulcanization time and vulcanization temperature of the unvulcanized rubber composition, and the information included in the thermal history information is information different from the vulcanization time and the vulcanization temperature.

10. A physical property data prediction device configured by a computer that predicts values ​​of physical property data of a vulcanized rubber composition produced by vulcanizing an unvulcanized rubber composition obtained by mixing a plurality of predetermined raw materials, a machine learning unit that uses learning data in which values ​​of physical property data relating to a plurality of vulcanized rubber compositions prepared using a plurality of raw materials are used as learning output data, and information on the compounding ratios of the respective raw materials in the vulcanized rubber compositions and information on processing conditions for preparing the vulcanized rubber compositions are used as learning input data, and that performs machine learning on a prediction model in a computer to determine the relationship between the physical property data, the compounding ratios, and the processing conditions; a prediction unit having a prediction model that predicts values ​​of physical property data of the prediction target vulcanized rubber composition using information on the compounding ratios of constituent raw materials that constitute an unvulcanized rubber composition before vulcanization of the prediction target vulcanized rubber composition and information on processing conditions for producing the prediction target vulcanized rubber composition, The information on processing conditions used in the machine learning is information on the thermal history of the raw materials during processing to produce the vulcanized rubber composition, and includes a thermal history indicating the accumulation of heat generated in the raw materials during processing.

11. A preprocessing unit that performs preprocessing to calculate a value of the thermal history of the raw material before the prediction model undergoes machine learning, the machine learning unit performs machine learning on the prediction model using the calculated value of the thermal history as information on the thermal history; 11. The physical property data prediction device according to claim 10, wherein the thermal history value is a value calculated using temperature data measured over time during processing of the raw materials, and the temperature data includes temperature data during mixing of the raw materials and temperature data during vulcanization of the unvulcanized rubber composition.

12. A factor data prediction method for causing a computer to predict at least one of information on blending ratios of raw materials and information on processing conditions in a vulcanized rubber composition produced by vulcanizing an unvulcanized rubber composition obtained by mixing a plurality of predetermined raw materials, the method comprising: a step of performing machine learning on a predictive model in a computer to determine the relationship between the compounding ratios and the processing conditions and the physical property data, using learning data in which information on the compounding ratios of each of the plurality of raw materials in the vulcanized rubber composition and information on processing conditions for producing the vulcanized rubber composition are used as learning output data and values ​​of physical property data related to the plurality of vulcanized rubber compositions produced using the raw materials are used as learning input data; and using values ​​of physical property data of a vulcanized rubber composition to be predicted, the prediction model predicts at least one of information on the blending ratio of constituent raw materials constituting an unvulcanized rubber composition before vulcanization of the vulcanized rubber composition to be predicted, and information on processing conditions for producing the vulcanized rubber composition to be predicted, A factor data prediction method, characterized in that the information on the processing conditions used in the machine learning includes information on the thermal history that the raw materials have received during processing of the raw materials to produce the vulcanized rubber composition, and which indicates the accumulation of heat generated in the raw materials during processing.

13. Before the step of performing machine learning, A step of causing a computer to perform preprocessing to calculate a value of the thermal history of the raw material; and causing a computer to input the calculated value of the thermal history into the prediction model as information of the thermal history.

13. The factor data prediction method according to claim 12, wherein the value of the thermal history is calculated using temperature data measured over time during processing of the raw materials, and the temperature data includes temperature data during mixing of the raw materials and temperature data during vulcanization of the unvulcanized rubber composition.

14. A factor data prediction device configured by a computer that predicts at least one of information on blending ratios of raw materials and information on processing conditions in a vulcanized rubber composition by vulcanizing an unvulcanized rubber composition obtained by mixing a plurality of predetermined raw materials, a machine learning unit that uses learning data in which information on the compounding ratios of each of a plurality of raw materials in the vulcanized rubber composition and information on processing conditions for producing the vulcanized rubber composition are used as learning output data, and values ​​of physical property data related to a plurality of vulcanized rubber compositions produced using the raw materials are used as learning input data, and that performs machine learning on a prediction model in a computer to determine the relationship between the compounding ratios, the processing conditions, and the physical property data; a prediction unit having a prediction model that predicts, using values ​​of physical property data of a prediction target vulcanized rubber composition, at least one of information on the blending ratio of constituent raw materials constituting an unvulcanized rubber composition before vulcanization of the prediction target vulcanized rubber composition and information on processing conditions for producing the prediction target vulcanized rubber composition, The factor data prediction device is characterized in that the information on the processing conditions used in the machine learning includes information on the thermal history that the raw materials have received during processing of the raw materials to produce the vulcanized rubber composition, and indicates the accumulation of heat generated in the raw materials during processing.

15. A pre-processing unit is further provided that performs pre-processing to calculate a value of the thermal history of the raw material before the prediction model is subjected to machine learning, the machine learning unit performs machine learning on the prediction model using the calculated value of the thermal history as information on the thermal history; 15. The factor data prediction device according to claim 14, wherein the value of the thermal history is a value calculated using temperature data measured over time during processing of the raw materials, and the temperature data includes temperature data during mixing of the raw materials and temperature data during vulcanization of the unvulcanized rubber composition.

Images