Unvulcanized kneaded rubber processing method and system

The method and system utilize unsupervised machine learning to classify unvulcanized kneaded rubber into groups with similar characteristics, enabling efficient production of rubber products by applying standardized processing conditions, addressing the challenge of varying rubber compositions.

JP2024168205APending Publication Date: 2024-12-05THE YOKOHAMA RUBBER CO LTD
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
JP2023084684
Authority / Receiving Office
JP · JP
Patent Type
Applications
Current Assignee / Owner
Filing Date
2023-05-23
Publication Date
2024-12-05

AI Technical Summary

Technical Problem

Existing methods struggle to efficiently process unvulcanized kneaded rubber to desired specifications due to variations in rubber characteristics, requiring different processing conditions for each type, which is difficult to determine and implement.

Method used

A method and system using unsupervised machine learning to classify unvulcanized kneaded rubber into groups with similar characteristics, allowing for the application of standardized processing conditions through a computing device, and a predictive model to determine suitable processing parameters for each group.

Benefits of technology

Enables efficient production of unvulcanized kneaded rubber to desired specifications by simplifying the determination of processing conditions, ensuring consistent quality across varying rubber compositions, and accommodating new compositions through supervised machine learning.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 2024168205000001_ABST
    Figure 2024168205000001_ABST
Patent Text Reader

Abstract

To provide a processing and a system which simply determine a processing condition suitable for a feature amount of a large number of kinds of unvulcanized kneaded rubber, and can efficiently manufactured processed unvulcanized kneaded rubber.SOLUTION: A preparation step S100 creates a classification data set 10a obtained by classifying a data set 10 input to a calculation device 4 into a plurality of groups by clustering based on a feature amount of a target feature 11 selected from the data set 10, constructs a prediction model 20 by machine learning using the classification data set 10a as teacher data, and inputs preset data 30 setting a reference processing condition capable of processing unvulcanized kneaded rubber into a non-defective for each group by a processing device 3 to the calculation device 4. A processing step S200 predicts a group into which the unvulcanized kneaded rubber R is classified by the calculation device 4 using input data D1 input to the calculation device 4 and the predication model 20, and processes the unvulcanized kneaded rubber R under a reference processing condition set to the predicted group by the processing device 3.SELECTED DRAWING: Figure 1
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present invention relates to a method and system for processing unvulcanized mixed rubber, and more specifically, to a method and system for processing unvulcanized mixed rubber that can easily determine processing conditions suitable for the characteristic quantities of each rubber for many types of unvulcanized mixed rubber having different rubber characteristics (rubber physical property values), and efficiently produce unvulcanized mixed rubber processed to desired specifications. [Background technology]

[0002] The unvulcanized mixed rubber that has been mixed in a mixer is processed to the desired specifications by processing equipment such as an extruder, a rolling machine, and a vulcanizer. In each processing step, appropriate processing conditions are set to process the unvulcanized mixed rubber to the desired specifications (see, for example, Patent Document 1). In the invention described in Patent Document 1, in order to maintain the dimensions and shape of the extruded unvulcanized rubber, a speed change pattern for the extrusion speed and the conveyor transport speed is determined according to the characteristics of the unvulcanized rubber, including its physical properties and environmental conditions.

[0003] In the manufacturing sites of rubber products such as tires, many kinds of unvulcanized kneaded rubbers with different formulations or different kneading conditions even with the same formulation are used, so it is difficult to process each unvulcanized kneaded rubber to the desired specifications by simply determining the speed change pattern of the extrusion speed and the conveyor transport speed and processing it. If the characteristics of the unvulcanized kneaded rubbers are similar, they can be processed to roughly the same quality by using the same processing conditions. Therefore, there is room for improvement in easily determining processing conditions suitable for the characteristics of each rubber for many kinds of unvulcanized kneaded rubbers with different rubber characteristics, and efficiently manufacturing unvulcanized kneaded rubbers processed to the desired specifications. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Publication No. 7-60821 Summary of the Invention [Problem to be solved by the invention]

[0005] An object of the present invention is to provide a method and system for processing unvulcanized mixed rubber that can easily determine processing conditions suitable for the characteristic quantities of each of many types of unvulcanized mixed rubber having different rubber characteristics, and efficiently produce unvulcanized mixed rubber processed to desired specifications. [Means for solving the problem]

[0006] The processing method of unvulcanized kneaded rubber of the present invention for achieving the above object comprises feeding unvulcanized kneaded rubber that has been subjected to a kneading process by a kneader into a processing device, controlling the processing device by a calculation device, and processing the unvulcanized kneaded rubber by the processing device under processing conditions inputted to the calculation device, the method comprising the steps of: in a preparatory stage, for a large number of types of unvulcanized kneaded rubber having different formulations, acquiring feature amounts indicating the characteristics of the large number of types of features before the unvulcanized kneaded rubber is fed into the processing device, and inputting the acquired feature amounts into the calculation device as a data set; selecting a plurality of types of the features from the data set as target features; and using the calculation device to perform classification by unsupervised machine learning based on the feature amounts of the respective target features to create a classification data set in which the data set is classified into a specified predetermined number of groups; and using the feature amounts of the respective target features as explanatory variables, The method is characterized in that a predictive model is constructed using machine learning with the arithmetic device and the classification dataset as training data, with the group as the objective variable, and preset data for each of the groups, which are set in advance as standard processing conditions under which the unvulcanized kneaded rubber can be processed into a good product by the processing device, are input into the arithmetic device, and when the unvulcanized kneaded rubber is processed by the processing device after completion of the preparation stage, feature amounts indicating the target features of each of the unvulcanized kneaded rubber input to the processing device are obtained and input into the arithmetic device as input data, the arithmetic device predicts a group into which the unvulcanized kneaded rubber input to the processing device will be classified using the input data and the predictive model, and the unvulcanized kneaded rubber input to the processing device is processed by the processing device under the standard processing conditions set for the predicted group.

[0007] The processing system for unvulcanized kneaded rubber of the present invention includes a processing device into which unvulcanized kneaded rubber that has been subjected to a kneading process by a kneader is input, and a calculation device that controls the processing device, and the unvulcanized kneaded rubber is processed by the processing device under processing conditions input to the calculation device. In the processing system for unvulcanized kneaded rubber, feature amounts indicating the characteristics of many types of unvulcanized kneaded rubber with different formulations are acquired for many types of features before the unvulcanized kneaded rubber is input as a dataset to the calculation device, and classification is performed by unsupervised machine learning using the calculation device based on the feature amounts of multiple types of features selected as target features from the dataset, thereby creating a classification dataset in which the dataset is classified into a specified number of groups, and the feature amounts of each of the target features are used as explanatory variables to perform classification. A predictive model is constructed by machine learning using the arithmetic device with the classification dataset as training data, with a predetermined number of the groups as objective variables, and standard processing conditions for processing the unvulcanized kneaded rubber into a good product by the processing device are set in advance for each of the groups and input to the arithmetic device, and when the unvulcanized kneaded rubber is processed by the processing device, feature amounts indicating the target features of each of the unvulcanized kneaded rubber input to the processing device are acquired and input to the arithmetic device as input data, and a group into which the unvulcanized kneaded rubber input to the processing device is classified is predicted from the multiple groups using the input data and the predictive model, and the unvulcanized kneaded rubber input to the processing device is processed by the processing device under the standard processing conditions set for the predicted group. Effect of the Invention

[0008] According to the present invention, since the unvulcanized kneaded rubbers belonging to the same group are similar in feature values, they can be processed to roughly the same quality under the same processing conditions. Reference processing conditions suitable for each group are set in advance, and the prediction model is also constructed. Therefore, when processing the unvulcanized kneaded rubber with a processing device, the processing conditions suitable for the unvulcanized kneaded rubber having each of the input data can be easily determined as the preset reference processing conditions using a calculation device by simply inputting the input data of the unvulcanized kneaded rubber to be fed into the processing device. As a result, even if the feature values ​​are variously different, unvulcanized kneaded rubber processed to the desired specifications can be efficiently manufactured.

[0009] In addition, the prediction of the group is made using the input data and a prediction model constructed by supervised machine learning. Therefore, even if the compounding of the unvulcanized kneaded rubber is new, if the input data is available, the group into which the unvulcanized kneaded rubber will be classified can be predicted, and the unvulcanized kneaded rubber can be processed under the standard processing conditions suitable for the predicted group. [Brief description of the drawings]

[0010] [Figure 1] FIG. 1 is an explanatory diagram illustrating an embodiment of a processing system for unvulcanized kneaded rubber. [Diagram 2] FIG. 1 is an explanatory diagram illustrating an example of a data set. [Diagram 3] FIG. 1 is a flow diagram illustrating the steps of an embodiment of a method for processing unvulcanized kneaded rubber. [Figure 4] FIG. 11 is an explanatory diagram illustrating a result of a principal component analysis. [Diagram 5] FIG. 11 is an explanatory diagram illustrating a result of clustering. [Figure 6] FIG. 1 is an explanatory diagram illustrating a classification data set. [Figure 7] FIG. 4 is an explanatory diagram illustrating an example of preset data. [Figure 8] FIG. 4 is an explanatory diagram illustrating an example of input data. [Figure 9] FIG. 11 is an explanatory diagram illustrating a prediction result using a prediction model. [Figure 10] FIG. 13 is an explanatory diagram illustrating a data set in a modified example. [Figure 11] FIG. 11 is a flow chart illustrating the procedure of a modified example of a method for processing unvulcanized kneaded rubber. [Figure 12] FIG. 13 is a graph illustrating the results of a time series analysis. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0011] Hereinafter, a method and system for processing unvulcanized kneaded rubber according to the present invention will be described based on the embodiments shown in the drawings.

[0012] An embodiment of a processing system 1 for unvulcanized kneaded rubber illustrated in FIG. 1 is used to perform a processing method for unvulcanized kneaded rubber. The processing system 1 and the processing method are used to process unvulcanized kneaded rubber R that has been subjected to a kneading process by a kneader 2 into a desired specification by a processing device (extruder) 3. That is, in this embodiment, as described later, according to the procedure illustrated in FIG. 3, a data set 10 illustrated in FIG. 2 is input to a calculation device 4 as a preparation stage (preparation stage S100 described later), and a prediction model 20 and preset data 30 illustrated in FIG. 7 are created in advance. When processing unvulcanized kneaded rubber R by the processing device 3 (processing step S200 described later), input data D1 acquired by a data acquisition device 5 before inputting unvulcanized kneaded rubber R into the processing device 3 are input to the calculation device 4, and unvulcanized kneaded rubber R is processed under reference processing conditions previously set by the calculation device 4 using the prediction model 20 and the preset data 30.

[0013] First, the processing system 1 will be described in detail.

[0014] The processing system 1 includes a processing device 3, a calculation device 4, and various data acquisition devices 5. The processing system 1 is incorporated into a manufacturing line for rubber products. The processing step of the unvulcanized kneaded rubber R by the processing device 3 of the processing system 1 is performed after the kneading step by the kneader 2. The unvulcanized kneaded rubber R is used to manufacture various rubber products such as rubber tires, conveyor belts, and hoses.

[0015] Examples of the processing device 3 include various known processing devices such as an extruder, a rolling machine, and a vulcanizing device that process the unvulcanized kneaded rubber R. In this embodiment, an extruder is used as the processing device 3. The unvulcanized kneaded rubber R extruded by the extruder is transported to the next process by a transport conveyor. The processing device 3 is controlled by a calculation device 4, and the processing conditions in the processing device 3 are input to the calculation device 4. The processing conditions are, for example, target values ​​(sometimes target selections) of the control parameters of the processing device 3. When the processing device 3 is an extruder, examples of the processing conditions include the screw rotation speed [rpm], the conveyor speed [m / min] of the transport conveyor, the type of extrusion die shape, the screw temperature [°C], and the cylinder temperature [°C]. The extrusion die shape is a target selection. When the processing device 3 is a rolling machine, examples of the processing conditions include the conveyor speed [m / min], the cooling drum temperature [°C], the gap between the rolls [mm], the roll speed [rpm], and the cut-out width [m]. When the processing device 3 is a vulcanizing device, the processing conditions are exemplified by vulcanization time [min], vulcanization temperature [°C], and vulcanization pressure [MPa]. The processing conditions are not limited to the exemplified control parameters, and other control parameters can be appropriately selected. For example, when the processing device 3 is an extruder, the processing conditions may include a cooling drum temperature and a cutout width. In this embodiment, an extruder is used as the processing device 3.

[0016] The arithmetic device 4 may be any of various known computers. The arithmetic device 4 has a central processing unit (CPU), a main storage unit (memory), an auxiliary storage unit (e.g., HDD), an input unit (keyboard, mouse), and an output unit (display). A predetermined program, a data set 10, a classification data set 10a, a prediction model 20, and preset data 30 are stored in the auxiliary storage unit.

[0017] When the data set 10 is input from the input unit in the preparation stage S100 before the unvulcanized kneaded rubber R is processed by the processing device 3, the calculation device 4 executes a predetermined program to carry out each data processing instructed by the program, thereby creating a prediction model 20 and preset data 30. When the calculation device 4 executes the processing step S200 after the completion of the preparation stage S100, the calculation device 4 executes each data processing instructed by the program, thereby controlling the processing device 3.

[0018] The various data acquisition devices 5 acquire feature quantities (rubber physical property values) of the unvulcanized kneaded rubber R that has been subjected to the kneading process. More specifically, the various data acquisition devices 5 acquire feature quantities of each of the target features 11 selected in step (S120) in the procedure of the processing method for unvulcanized kneaded rubber illustrated in Fig. 3, which will be described later. As the various data acquisition devices 5, various known measuring devices such as a Mooney viscometer, a capillary rheometer, a plasticity meter, a plastometer, a mass meter, a specific gravity meter, and a specific heat measuring device can be used.

[0019] Next, the data set 10 will be described in detail.

[0020] The data set 10 illustrated in Fig. 2 may be input to the calculation device 4 (stored in the auxiliary storage unit of the calculation device 4) at the time of the preparation stage S100. In the preparation stage S100, for many types of unvulcanized kneaded rubber (sample A) with different blends, feature amounts indicating the characteristics of the many types of features are acquired for the many types of features before the samples are fed into the processing device 3, and the feature amounts are input to the calculation device as the data set 10 together with a blend label that specifies the blend of the sample A. The data set 10 has feature amounts (hereinafter, feature data) of many types of features (G'_0.28, G'_300, , G''_0.28, G''_300, , tan δ_0.28, tan δ_300, ) for each of the unvulcanized kneaded rubber samples A (A1, A2, . . . , An) listed in the leftmost column of the table in Fig. 2.

[0021] The data set 10 may be created by preparing many types of unvulcanized mixed rubber samples A that have been mixed by the mixer 2 with different compositions, and acquiring feature quantities indicating the characteristics of each sample using the data acquisition device 5 before the samples are fed into the processing device 3. Manufacturers of rubber products accumulate large amounts of data on many types of feature quantities for unvulcanized mixed rubber R with various compositions. Data accumulated by manufacturers includes laboratory data such as measured values ​​in actual rubber product production lines, results of many experiments and tests in research and development, and computer simulation results. This accumulated data can be used as the data set 10.

[0022] A compounding label is used for the compounding of sample A. The compounding label may be anything that can specify the compounding of the unvulcanized kneaded rubber R. In more detail, it is sufficient that the label can distinguish the compounding of the unvulcanized kneaded rubber R from other compounds, so it does not have to be actual compounding data and may be a simple mark or symbol. Therefore, the same compounding label means that the compounding is the same. The compounding refers to the type of components of the unvulcanized kneaded rubber R and the amount of each component. The data set 10 may have a compounding label or a compounding label and actual compounding data.

[0023] Even for the same blended label, at least some of the characteristic quantities will be slightly different due to the degree of blending by the blender 2 and changes over time after blending. When the blends are similar, the respective characteristic quantities will generally be similar. The number (n) of samples A is huge, for example, 50 or more. The number of samples A for the same blended label is, for example, 5 or more.

[0024] The feature data is a group of data (a collection of feature values ​​for each sample A) that includes feature values ​​that indicate the characteristics of sample A of unvulcanized kneaded rubber R for many types of features. In the table of FIG. 2, some of the features and feature values ​​are omitted by "...".

[0025] The features in the feature data are classified into a feature group having multiple features that indicate the viscoelastic properties of sample A, a feature group having multiple features that indicate the viscous properties, and a feature group having multiple features that indicate the torque properties obtained by rotational vibration. Specifically, the types of features are dynamic shear modulus (G), dynamic storage shear modulus (G'), dynamic loss shear modulus (G'', complex modulus (G * ), creep compliance (J), loss creep compliance (J'), complex creep compliance (J * ), loss tangent (tanδ), kinematic viscosity (η'), apparent kinematic viscosity (η''), complex kinematic viscosity (η * ), elastic torque (S'), viscous torque (S''), and complex torque (S * ), Mooney viscosity (M), scorch time (T5, T10, T30, T50, T95), etc. Also, the weight, specific gravity, heat capacity, etc. of sample A can be used as one type of characteristic.

[0026] The feature quantities of some features in the data set 10 may be obtained only in one specific deformation region, but are preferably obtained for each of a plurality of deformation regions with different degrees of deformation. The plurality of deformation regions are, for example, a small deformation region with a strain of 10% or less and a large deformation region with a strain of several tens of percent or more. The plurality of deformation regions may further include an intermediate deformation region between the small deformation region and the large deformation region. The small deformation region can be set arbitrarily in a range of strain of 10% or less, for example, 1% or less, or a range of 0.1% to 0.5%. The large deformation region can be set arbitrarily in a range of strain of several tens of percent or more, for example, 20% or more, or a range of 50% to 1250%. In FIG. 2, for example, for the dynamic loss shear modulus G', which is a characteristic type, there are a dynamic loss shear modulus G'_0.28 in a small deformation region with a strain of 0.28% and a dynamic loss shear modulus G'_300 ​​in a large deformation region with a strain of 300%. In this way, each deformation region is treated as an individual feature of the same type in the data set 10. In Fig. 2, the deformation regions are common to each feature type at 0.28% and 300%, but a desired deformation region may be set for each feature type. For example, for the kinematic viscosity η', a strain of 0.14% may be used as the small deformation region, and a strain of 1250% may be used as the large deformation region.

[0027] Depending on the sample A, some of the feature quantities may be missing. When a feature quantity is missing, it is advisable to complement the missing feature quantity. Various known methods can be used to complement the missing feature quantity. For example, when a sample A with some missing feature quantities exists in the data set 10, the feature quantity is complemented using the non-missing feature quantity of a sample A with the same blend label, or the feature quantity is complemented using a representative value of the non-missing feature quantity of a plurality of samples A with the same blend label.

[0028] Furthermore, when the range of feature values ​​(numeric values) varies significantly for each feature (each column) in the data set 10, it is advisable to scale (normalize or standardize) each feature value. Various known methods can be used for scaling, such as the Min-Max method, a method using the mean and standard deviation, RobustScaler (dividing the deviation from the median by the interquartile range), and a method using RMSE (root mean square error). By scaling the feature data of the data set 10, the weights of the values ​​of the feature values ​​of each feature can be made roughly equal, which is advantageous in avoiding biased data processing, such as emphasizing only one feature in the preparation stage S100 described below.

[0029] Each feature amount is acquired by various data acquisition devices 5 during the time when the unvulcanized kneaded rubber sample A is fed into the processing device 3 after the kneading process by the kneader 2. The timing for acquiring each feature amount by the data acquisition device 5 is preferably immediately before feeding into the processing device 3, considering the change in physical properties due to the change over time from the kneading process, but it is difficult to acquire all of the many feature amounts immediately before feeding into the processing device 3. Therefore, it is preferable to use the same timing for each feature, and to unify the degree of the change in physical properties due to the change over time for each feature. In an actual production line in which the processing system 1 is incorporated, each feature amount is acquired sequentially by each data acquisition device 5, and the timing is approximately the same for each feature. When using laboratory data as the sample A, it is preferable to actually measure the acquisition timing (the time elapsed since the sample was discharged from the kneader 2) by each data acquisition device 5 in the production line, and to consider each measured acquisition timing. As a result, the acquisition timing of the feature amount differs between different features, but the feature amount is acquired at approximately the same acquisition timing for the same feature. Therefore, the degree of change over time after the kneading process will be roughly the same for feature amounts of the same characteristic, which is advantageous in suppressing the influence of fluctuations in feature amounts due to changes over time.

[0030] Next, the method for processing the unvulcanized kneaded rubber will be described in detail.

[0031] As shown in FIG. 3, the procedure of the processing method of unvulcanized kneaded rubber is to execute a preparation step S100 before the processing step S200. In the preparation step S100, a data set 10 is input to the calculation device 4 (S110), and a prediction model 20 and preset data 30 are created (S120 to S150). Next, in the processing step S200 in which the unvulcanized kneaded rubber R is processed by the processing device 3, input data D1 of the unvulcanized kneaded rubber R to be input to the processing device 3 is input to the calculation device 4 (S210), and the unvulcanized kneaded rubber R is processed under the reference processing conditions specified using the input data D1, the prediction model 20, and the preset data 30 (S220, S230). The contents of each step (S110 to S150, S210 to S230) will be described in detail below.

[0032] In step (S110), the data set 10 is input to the calculation device 4. This step can be performed at any timing before the start of the preparation stage S100.

[0033] In step (S120), a plurality of types of features are selected as target features 11 from among many types of features of the feature data of the data set 10. Specifically, the arithmetic device 4 executes data processing for outputting a selection screen for the target features 11 to the output unit, and the target features 11 are selected on the selection screen by the input unit.

[0034] If the number of types of feature types of features selected as the target feature 11 is small, information necessary for clustering will be missing, while if the number of types is too large, the spherical concentration phenomenon will occur or combinations with high correlation coefficients will occur. Therefore, it is desirable to select an appropriate number of feature types as the target feature 11, taking into consideration the improvement of accuracy and the excessive influence of specific characteristics due to information duplication. The number of types of feature types of the target feature 11 is preferably 4 or more, and the upper limit is, for example, 12. For example, an appropriate target feature 11 is selected for each use (part of a rubber product) of the unvulcanized kneaded rubber R to be processed. Note that the number of types of feature types of the target feature 11 is not limited to this when performing feature reduction such as principal component analysis in a later process.

[0035] Although no particular constraints need to be placed on the selection of the target features 11, placing constraints is advantageous for improving the accuracy of the clustering in step (S130). Therefore, it is desirable to use constraint conditions that restrict the types of features to be selected, and select features that satisfy the constraint conditions as the target features 11.

[0036] As a constraint condition, the target feature 11 is limited to each feature corresponding to each characteristic. For example, at least two feature groups are selected from a feature group (a collection of multiple features) indicating viscoelastic characteristics, a feature group indicating viscosity characteristics, and a feature group indicating torque characteristics obtained by rotational vibration, and at least one feature from each of the selected feature groups is selected as the target feature 11. More preferably, all of these feature groups are selected, and at least one feature from each of the selected feature groups is selected as the target feature 11. In this way, by selecting features with different characteristics, rather than only the same characteristic, as the target feature 11, it is advantageous to grasp the physical properties of the unvulcanized kneaded rubber R from multiple angles.

[0037] As another constraint condition, when the feature type selected as the target feature 11 has feature amounts for a plurality of deformation regions with different degrees of deformation in the data set 10, the features in each deformation region (individual features of the same type of feature) are selected as the target feature 11. For example, when the dynamic storage shear modulus G' is selected as the target feature 11, each feature (individual features of the same type of feature) of the dynamic storage shear modulus G'_0.28 in the small deformation region and the dynamic storage shear modulus G'_300 ​​in the large deformation region are selected as the target feature 11. By selecting the target feature 11 in this way, the behavior of the unvulcanized kneaded rubber during processing by the processing device 3 can be grasped more accurately.

[0038] The constraint conditions are not limited to the above. For example, in addition to the features selected from the characteristics of the unvulcanized kneaded rubber R, features whose feature amount does not change depending on the deformation region (weight, specific gravity, heat capacity, etc.) may be selected as the target feature 11. Also, when each compounding label and actual compounding data are described in the data set 10, the compounding label and actual compounding data may be selected as the target feature 11. Selecting the compounding label and actual compounding data as the target feature 11 and performing clustering is advantageous for classifying those with more similar features into the same group.

[0039] The target features 11 include, for example, the dynamic storage shear modulus (G'), the dynamic loss shear modulus (G''), the elastic torque (S'), the viscous torque (S''), the loss creep compliance (J'), and the complex creep compliance (J'') obtained by rotational vibration in two deformation regions, namely, a small deformation region with a strain of 0.28% and a large deformation region with a strain of 300%. * ), kinematic viscosity (η'), apparent kinematic viscosity (η''), and loss tangent (tan δ) are selected. That is, as the target feature 11, for example, a total of 18 types of features are selected by combining 9 types of feature types and 2 types of deformation regions.

[0040] In step (S130), the computing device 4 executes data processing to classify the data set 10 into a specified number of groups (G1 to G4) using classification (clustering) by unsupervised machine learning based on the feature quantities of each selected target feature 11. For clustering, various known hierarchical clustering methods such as Ward's method, group average method, shortest distance method, and longest distance method, and various known non-hierarchical clustering methods such as k-means can be used. The predetermined number (the number of classifications by clustering) can be arbitrarily selected within a range that is less than the number of types of the compounded labels of sample A. When the number of types of the compounded labels of sample A is 15, the number of classifications by clustering is, for example, 3 or more and less than 15.

[0041] In this step (S130), it is preferable to perform classification (clustering) by unsupervised machine learning based on the principal component scores of each sample A obtained by performing dimensionality reduction by principal component analysis on each target feature 11. If the number of target features 11 selected in the above step (S120) is, for example, four or more, it is preferable to perform data processing by known principal component analysis by the calculation device 4 to reduce the dimensionality of the data.

[0042] FIG. 4 shows an example of the result of data processing of principal component analysis. In FIG. 4, the selected target features 11 are regarded as 18 types of parameters, and the first principal component axis PC1 of two axes created from the 18 types of parameters is set as the horizontal axis, and the other second principal component axis PC2 is set as the vertical axis. The white dots in FIG. 4 are plots of the principal component scores of each sample A (A1, . . . , An) of the data set 10. Note that the reduction of the dimensions of data by principal component analysis is not limited to reduction from multi-dimensions to two-dimensions, and may be reduction from multi-dimensions to three-dimensions, for example.

[0043] Figure 5 shows an example of the clustering result. In Figure 5, the number of classifications by clustering is four, and the principal component scores of each sample A are classified into each group (G1 to G4). Since it is not possible to express each group by color or the like in Figure 5, each classified group is shown by a dashed line.

[0044] Each sample A belonging to the same group has similar feature values ​​of the target feature 11. Therefore, each sample A belonging to the same group can be processed to approximately the same quality when processed under the same processing conditions. Thus, in this step (S130), a set of samples A that can be processed to approximately the same quality when processed under the same processing conditions is grouped according to the similarity of each feature value of the target feature 11.

[0045] The feature values ​​of the target features 11 are similar between samples A with the same compounding label. Therefore, in clustering, it is preferable that samples A with the same compounding label are classified into the same group. In order to classify the dataset 10 into multiple groups by the same compounding label in clustering, the compounding label or actual compounding data is selected as the target feature 11. In addition, when the clustering result is confirmed and the samples A with the same compounding label are not classified into the same group, it is preferable to correct the data so that the samples A are classified into the same group. For example, the data correction is performed by selecting a group with the most samples A with the same compounding label from among multiple groups as the group to which the samples A with the same compounding label belong, and classifying the samples A that belong to groups other than the selected group into the selected group. In this way, it is preferable that the samples A with the same compounding label are classified into the same group in clustering.

[0046] The classification dataset 10a illustrated in Fig. 6 is a table showing the results of the clustering in Fig. 5. The classification dataset 10a is obtained by classifying the samples A in the dataset 10 into a plurality of groups, and includes feature quantities of the target features 11 for each of the classified samples A.

[0047] In step (S140), the arithmetic device 4 executes data processing to construct a prediction model 20 by machine learning using the classification data set 10a as training data. The prediction model 20 is a type of computer program constructed by machine learning using training data and predicting the group (G1 to G4) into which the unvulcanized kneaded rubber R is classified. The prediction model 20 is constructed by supervised machine learning in which the feature amount of each target feature 11 in the classification data set 10a is used as an explanatory variable and each group (G1 to G4) is used as a target variable. Examples of supervised machine learning include various well-known supervised machine learning such as k-nearest neighbor method (k-nn method), random forest, and deep neural network (DNN).

[0048] In one example of the step (S140) using the k-nearest neighbor method as supervised machine learning, a hyperparameter (number of nearest neighbor samples) k is set. The hyperparameter k can be any positive integer. A more appropriate value may be set for the hyperparameter k using a known validation method such as hold-out validation or cross validation. For example, 50% to 80% of the total number of samples A in the classification data set 10a are used as provisional teacher data, and the remaining samples excluding the provisional teacher data from all samples A are used as test data. A prediction is made for the test data while changing the value of the hyperparameter k, and the accuracy rate of the prediction is calculated. The value of the hyperparameter k is determined to be the one that has the highest accuracy rate among the calculated accuracy rates. Note that, in the step (S140) using the k-nearest neighbor method without a learning phase (training phase), the hyperparameter k is set, but in the step (S140) using other supervised machine learning, a learning phase (training phase) may be performed.

[0049] In step (S150), standard processing conditions for each group that can process the unvulcanized kneaded rubber sample A into a non-defective product by the processing device 3 are preset and input to the calculation device 4 to create preset data 30. Specifically, the calculation device 4 executes data processing to output a setting screen for the processing conditions to the output unit, and the processing conditions are set on the setting screen by the input unit.

[0050] A good product means that the sample A processed by the processing device 3 is in a good state. Whether the product is good or not is determined, for example, by whether the dimensions, shape, surface condition, etc. of the processed sample A fall within a predetermined allowable range using these as indicators. If it falls within the allowable range, it is a good product, and if it does not fall within the allowable range, it is a defective product.

[0051] The reference processing conditions are set based on the characteristic quantities of the samples A belonging to each group. By grasping the characteristic quantities of each group, a person skilled in the art can set the reference processing conditions suitable for each group by referring to the experience gained in the manufacturing process of rubber products, various known literature, the accumulation of a large number of experimental and test data, and the accumulation of computer simulation results. The reference processing conditions are set for each group by utilizing such knowledge of the person skilled in the art. For example, when the samples A1 and A2 belonging to the group G1 are processed by the processing device 3 under the reference processing conditions set for the group G1, each of the samples A1 and A2 is processed into a good product. In this way, the reference processing conditions are set so that any sample A belonging to the set group will be a good product when processed by the processing device 3.

[0052] In the preset data 30 illustrated in Fig. 7, reference processing conditions are set for each of the groups (G1 to G4) listed in the leftmost column of the table in Fig. 7. In the table in Fig. 7, some of the more specific control parameters of the reference processing conditions and some of the target values ​​of the control parameters are omitted by "...". When an extruder is used as the processing device 3, the control parameters set as the reference processing conditions are, for example, the screw rotation speed, conveyor speed, extrusion die shape, screw temperature, and cylinder temperature.

[0053] By carrying out the above preparation step S100, the prediction model 20 and the preset data 30 are created and stored in the auxiliary storage unit of the calculation device 4. Each step (S110 to S150) of the preparation step S100 can be performed at any desired timing regardless of the processing step S200 in which the unvulcanized kneaded rubber R is actually processed by the processing device 3. The prediction model 20 and the preset data 30 created in the preparation step S100 do not need to be re-created unless there is a change in the data related to the prediction model 20 and the preset data 30, such as adding a sample A to the data set 10 or changing the target feature 11 selected in step (S120). In other words, since the prediction model 20 and the preset data 30 created once can be used as they are any number of times, if the prediction model 20 and the preset data 30 have been created, the preparation step S100 can be omitted.

[0054] Next, the processing step S200 will be described in detail.

[0055] In step (S210), before the unvulcanized kneaded rubber R that has been subjected to the kneading process by the kneader 2 is fed into the processing device 3, the various data acquisition devices 5 acquire feature amounts of the respective target features 11 related to the unvulcanized kneaded rubber R, and the feature amounts of the respective target features 11 acquired by the various data acquisition devices 5 are input as input data D1 to the arithmetic device 4. The input data D1 may be input directly from the various data acquisition devices 5 to the arithmetic device 4, or may be input to the arithmetic device 4 via an input unit.

[0056] The input data D1 illustrated in Fig. 8 is acquired by various data acquisition devices 5 and input to the calculation device 4. The input data D1 accumulates feature amounts of each target feature 11 of the unvulcanized kneaded rubber R. In the table of Fig. 8, specific feature amounts are omitted by "···".

[0057] The target features 11 in the input data D1 are the same as the target features 11 selected in step (S120) of the preparation stage S100 described above. The input data D1 may be subjected to completion of missing features and scaling of each feature, similar to the data set 10 in FIG. 2 described above. The timing of acquiring the input data D1 by the various data acquisition devices 5 may be the same as the timing of acquiring the features in the data set 10. By matching the acquisition timing of the various data acquisition devices 5 with the acquisition timing of the data set 10, the changes over time after the kneading process between the features in the data set 10 and the features acquired in this step (S210) become roughly the same, which is advantageous in eliminating the influence of fluctuations in the features due to changes over time.

[0058] In step (S220), the calculation device 4 executes data processing to predict a group into which the unvulcanized kneaded rubber R is classified, using the input data D1 and the prediction model 20. Specifically, when the prediction model 20 is constructed using the k-nearest neighbor method, the group into which the unvulcanized kneaded rubber R is classified is determined by majority vote based on the breakdown of the k nearest neighbor samples. For group determination in the k-nearest neighbor method, the Euclidean distance is usually used, but various known distances such as the Manhattan distance and the Mahalanobis distance can also be used.

[0059] Fig. 9 shows an example of a prediction result using the prediction model 20. In Fig. 9, the group into which the unvulcanized kneaded rubber R is classified is predicted to be group G2. If the group into which the unvulcanized kneaded rubber R is classified cannot be determined by majority decision in the k-nearest neighbor method, the graph illustrated in Fig. 9 may be output to the output section of the calculation device 4, and an operator may determine the group into which the unvulcanized kneaded rubber R is classified based on the graph, and input the determined group into the input section.

[0060] In step (S230), the calculation device 4 executes data processing to control the processing device 3 under the reference processing conditions set for the predicted group, and the unvulcanized kneaded rubber R is processed under the reference processing conditions. After starting the reference processing conditions in step (S230), it is also possible to adjust the processing conditions at an appropriate timing.

[0061] As described above, according to this embodiment, since the features of the unvulcanized kneaded rubbers R belonging to the same group classified by unsupervised machine learning are similar to each other, the reference processing conditions that can process the unvulcanized kneaded rubbers R into a good product are set in advance for each group by the processing device 3. In addition, the input data D1 of the unvulcanized kneaded rubber R and the prediction model 20 are used to accurately predict a group having a feature similar to the input data D1. Therefore, when the unvulcanized kneaded rubber R is processed by the processing device 3, the processing conditions suitable for the unvulcanized kneaded rubber R having each input data D1 can be easily determined as the reference processing conditions set in advance by using the calculation device 4 by simply inputting the input data D1 of the unvulcanized kneaded rubber R to be input to the processing device 3. As a result, even if the feature values ​​are variously different, the unvulcanized kneaded rubber R processed to the desired specifications can be efficiently manufactured.

[0062] In addition, the input data D1 and a prediction model 20 constructed by supervised machine learning are used to predict the group. Therefore, even if the compounding data of the unvulcanized kneaded rubber R is new compounding data that is not existing, the group to which the unvulcanized kneaded rubber R belongs can be predicted, and the unvulcanized kneaded rubber R can be processed under standard processing conditions suitable for the predicted group.

[0063] The set reference processing conditions can be used as a guideline for processing the unvulcanized kneaded rubber R into a good product. By fine-tuning the processing conditions, taking into consideration the surrounding environment and changes in physical properties over time from the kneading process, using the reference processing conditions obtained by inputting the input data D1 as a reference, the unvulcanized kneaded rubber R can be processed by the processing device 3 under more appropriate processing conditions in accordance with the site. In other words, since the reference processing conditions can be easily determined, unvulcanized kneaded rubber R processed to the desired specifications can be efficiently manufactured without being affected by the level of skill of the worker.

[0064] Next, modified examples of the embodiment of the method and system for processing unvulcanized kneaded rubber will be described.

[0065] FIG. 10 shows an example of a data set 10B used in a modified example. Data set 10B has an acquisition time added to data set 10 in FIG. 2 described above. The acquisition time is data indicating the time when the processing step S200 was performed by the processing device 3 for each sample A (the time when the feature quantities of many types of features were acquired). In the table in FIG. 10, the acquisition time is abbreviated by "···", but in reality, the date (YYYY / MM / DD) is listed. The acquisition time can be only the year and month, or the year and first half / second half, etc. Data set 10B also has a blending label (C1, C2, ···, Cm) as a feature.

[0066] Fig. 11 shows an example of the procedure of the unvulcanized rubber processing method in the modified example. In the modified example, other steps (S310, S320) are added to the procedure in Fig. 3 described above. That is, in the procedure of the modified example, the input data D1 inputted is corrected by the calculation device 4 (S310, S320). Therefore, in the modified example, the calculation device 4 predicts the group into which the unvulcanized kneaded rubber R is classified using the corrected input data D1 and the prediction model 20 (S220).

[0067] In step (S310), the arithmetic unit 4 executes data processing to perform a time series analysis of the feature quantities of each of the target features 11 based on the respective acquisition times of the many samples A belonging to the same group. For the time series analysis (time series prediction), various known time series analysis methods can be used, such as the X11 decomposition method, additive or multiplicative seasonal smoothing method, the autoregressive integrated moving average (ARIMA) model, the seasonal autoregressive integrated moving average (SARIMA) model, and the season-trend decomposition (STL decomposition). Since the kneading process by the kneader 2 is affected by seasonal changes in the temperature of the surrounding environment, it is preferable to use a method other than a non-seasonal method, such as the moving average method or the exponential smoothing method, which is suitable for data that does not include seasonal factors, for the time series analysis.

[0068] The group used for the time series analysis may be selected arbitrarily, but the more the number of samples A classified in the classification data set 10a of FIG. 6 described above, the more data there is in the time series analysis, which is advantageous for more accurately grasping the trend of the variation of the feature amount in the time series. In addition, the more the compositions of the many samples A used for the time series analysis are similar to each other, the more the influence on the variation of the feature amount due to the difference in composition is eliminated, which is advantageous for more accurately grasping the trend of the variation of the feature amount in the time series. Therefore, it is preferable to use samples A having the same composition label among the many samples A classified into the same group as the many samples A used for the time series analysis. In this step (S310), a group with the largest number of samples A having the same composition label is selected from the multiple groups (G1 to G4), and the feature amount of each target feature 11 for the samples A classified into the selected group is analyzed in time series based on the acquisition time.

[0069] FIG. 12 shows an example of the results of a time series analysis of the amount of variation in the dynamic storage shear modulus G'_300 ​​in the large deformation region for sample A having a blending label Cm classified into group G2. The open circles in FIG. 12 are observed values, plotted at the corresponding positions of the amount of variation in the dynamic storage shear modulus G'_300 ​​in the large deformation region for each sample A and the date. The open squares in FIG. 12 are plotted at the corresponding positions of the amount of variation due to seasonal variation obtained from the observed values ​​by time series analysis and the date. The black circles in FIG. 12 are plotted at the corresponding positions of the amount of variation due to trend variation (long-term variation) obtained by removing seasonal variation and irregular variation (noise) from the observed values ​​and the date.

[0070] The thick dashed line in Fig. 12 is a straight line approximation of the black circle points, and indicates the predicted trend variation. In Fig. 12, the acquisition time of the kneading process and processing process S200 of the unvulcanized kneaded rubber R is indicated by Db. The black star points on the thick dashed line at the acquisition time Db indicate the trend variation amount in the feature amount of the unvulcanized kneaded rubber R. The trend variation amount reflects the influence of the change in the degree of workmanship in the kneading process by the kneader 2, and the cause of this is, for example, aging deterioration of the kneader 2.

[0071] The white star points in FIG. 12 indicate the seasonal variation in the feature amount of the unvulcanized kneaded rubber R. In the group of white square points in FIG. 12, the number of data is small to accurately predict the seasonal variation, and the seasonal variation indicates whether the degree of influence of the seasonal variation on the seasonal average is positive or negative. Therefore, the seasonal variation is not taken into consideration in the modified example (seasonal variation amount=0), but the seasonal variation amount can be obtained depending on the data set 10B. In order to accurately predict the seasonal variation and obtain the seasonal variation amount in the feature amount of the unvulcanized kneaded rubber R, for example, the data set 10B in which data is collected in units of multiple years is used. With such a data set 10B, the periodic variation in the feature amount due to the seasonal variation can be predicted and the seasonal variation amount in the feature amount of the unvulcanized kneaded rubber R can be obtained. The obtained seasonal variation amount reflects the influence of changes in the surrounding environment such as temperature and humidity due to the season.

[0072] In step (S320), the calculation device 4 executes data processing to correct the input data D1 based on the result of the time series analysis. In this correction of the input data D1, the influence of seasonal and trend variations on the variation of the input data D1 is eliminated, so that the input data D1 can be compared with the data set 10B under approximately the same conditions. Specifically, the calculation device 4 calculates a correction value based on the seasonal and trend variations in the feature quantities of the unvulcanized kneaded rubber R, and corrects each feature quantity of the input data D1 with the calculated correction value. The correction value of the seasonal variation is a value that makes the seasonal variation obtained as a result of the time series analysis zero. The correction value of the trend variation is the difference between the trend variation obtained as a result of the time series analysis and the average value of the trend variation in the time series analysis (the representative value of the black circle in FIG. 12). Therefore, the input data D1 is corrected by the sum of the correction value of the seasonal variation and the correction value of the trend variation.

[0073] As described above, according to the modified example, in step (S220), the calculation device 4 executes data processing to predict the group into which the unvulcanized kneaded rubber R is classified, using the corrected input data D1 and the prediction model 20. The corrected input data D1 used for predicting the group is a value that can be considered to have been acquired under approximately the same conditions as the data set 10B, with the influence of seasonal and trend fluctuations on the fluctuations of the input data D1 eliminated. This is therefore advantageous in improving the accuracy of prediction by the prediction model 20 of the group into which the unvulcanized kneaded rubber R is classified.

[0074] In the above step (S120), a predetermined feature in the feature data of the data set 10 can also be selected as the target feature 11. That is, the feature to be selected as the target feature 11 is stored in the calculation device 4, and in the preparation stage S100, data processing is executed to select the stored target feature 11 by the calculation device 4. This allows the above step (S120) to be automatically performed by the calculation device 4.

[0075] In the above-described embodiment and its modified examples, in order to respond to changes in the manufacturing environment such as the development of unvulcanized kneaded rubber with a new compounding that does not exist and the aging of devices such as the kneader 2, it is desirable to periodically expand the data sets 10 and 10B and update the prediction model 20 and the preset data 30. The frequency of expansion and updating can be set arbitrarily, but is, for example, at a predetermined period such as monthly or semi-annual.

[0076] Although an embodiment of the present invention has been described above, the method and system for processing unvulcanized kneaded rubber of the present invention are not limited to the specific embodiment, and various modifications and variations are possible within the scope of the gist of the present invention.

[0077] The present disclosure includes the following inventions. Invention (1): A method for processing unvulcanized kneaded rubber, which has been subjected to a kneading process using a kneader, is fed into a processing device, which is controlled by a computing device, and the unvulcanized kneaded rubber is processed by the processing device under processing conditions inputted to the computing device, In a preparatory step, feature amounts indicating the characteristics of many types of unvulcanized mixed rubber having different formulations are obtained for many types of features before the unvulcanized mixed rubber is fed into the processing device, and the feature amounts are input to the arithmetic device as a data set. A plurality of types of the features are selected from the data set as target features, and classification is performed by unsupervised machine learning using the arithmetic device based on the feature amounts of the respective target features, thereby creating a classification data set in which the data set is classified into a specified number of groups. constructing a prediction model by machine learning using the arithmetic device with the feature amount of each of the target features as explanatory variables and a predetermined number of the classified groups as objective variables, and using the classification data set as training data; Preset data is input to the arithmetic unit, which is a standard processing condition for processing the unvulcanized kneaded rubber into a good product by the processing device for each of the groups; A method for processing unvulcanized kneaded rubber, in which, when processing the unvulcanized kneaded rubber using the processing device after completion of the preparatory stage, feature quantities indicating the target features of each of the unvulcanized kneaded rubber fed into the processing device are obtained and input as input data to the arithmetic device, and the arithmetic device uses the input data and the predictive model to predict a group from among a plurality of groups into which the unvulcanized kneaded rubber fed into the processing device will be classified, and the unvulcanized kneaded rubber fed into the processing device is processed by the processing device under the standard processing conditions set for the predicted group. Invention (2): The data set is added with the acquisition time when each of the features was acquired, The processing method for unvulcanized kneaded rubber described in invention (1) comprises the steps of: performing a time series analysis of the feature quantities of a specified target feature based on the acquisition times of the unvulcanized kneaded rubbers classified into the same group; and, when predicting the group, using the input data corrected based on the results of the time series analysis. Invention (3): A method for processing unvulcanized kneaded rubber according to invention (1) or (2), in which the classification by unsupervised machine learning is performed based on the principal component scores of multiple types of the unvulcanized kneaded rubber obtained by performing dimensionality reduction by principal component analysis on each of the target features. Invention (4): A method for processing unvulcanized kneaded rubber according to any one of inventions (1) to (3), in which the dataset contains features having the feature amount in each of a plurality of deformation regions having the same feature type but different degrees of deformation, and when the feature type having the feature amount in each of the plurality of deformation regions is selected as the target feature, each of the deformation regions is treated as an individual feature and selected as the target feature. Invention (5): A method for processing unvulcanized kneaded rubber according to any one of inventions (1) to (4), in which the selection of the target feature comprises selecting at least two feature groups from a feature group having a plurality of the features indicating the viscoelastic properties of the unvulcanized kneaded rubber, a feature group having a plurality of the features indicating the viscoelastic properties, and a feature group having a plurality of the features indicating the torque properties obtained by rotational vibration, and selecting at least one of the features from each of the selected feature groups as the target feature. Invention (6): A method for processing unvulcanized mixed rubber according to any one of inventions (1) to (5), wherein the dataset has a compounding label that identifies the compounding of the unvulcanized mixed rubber as the feature, and the compounding label is selected as the target feature. Invention (7): A method for processing unvulcanized kneaded rubber according to any one of Inventions (1) to (6), wherein, when the processing device is an extruder, the processing conditions are screw temperature, conveyor speed, extrusion die shape, screw temperature, and cylinder temperature; when the processing device is a rolling machine, the processing conditions are conveyor speed, cooling drum temperature, roll gap, roll speed, and cut-out width; and when the processing device is a vulcanizing device, the processing conditions are vulcanizing temperature, vulcanizing pressure, and vulcanizing time. Invention (8): A processing system for unvulcanized kneaded rubber, comprising a processing device into which unvulcanized kneaded rubber that has been kneaded by a kneader is input, and a calculation device that controls the processing device, the unvulcanized kneaded rubber being processed by the processing device under processing conditions input to the calculation device, In the arithmetic device, feature amounts indicating the characteristics of many types of unvulcanized mixed rubber having different formulations are acquired for many types of features before the unvulcanized mixed rubber is fed into the processing device, and the feature amounts are input as a dataset, and classification is performed by unsupervised machine learning using the arithmetic device based on the feature amounts of the multiple types of features selected from the dataset as target features, thereby creating a classification dataset in which the dataset is classified into a specified number of multiple groups, a prediction model is constructed by machine learning using the classification data set as training data by using the computing device, with the feature amounts of the target features as explanatory variables and a predetermined number of the classified groups as objective variables; A standard processing condition for processing the unvulcanized kneaded rubber into a non-defective product by the processing device is preset for each of the groups and input to the arithmetic device, When the unvulcanized kneaded rubber is processed by the processing device, feature amounts indicating the target features of each of the unvulcanized kneaded rubber fed into the processing device are acquired and input as input data to the calculation device, and a group into which the unvulcanized kneaded rubber fed into the processing device is classified is predicted from among the multiple groups using the input data and the predictive model, and the unvulcanized kneaded rubber fed into the processing device is processed by the processing device under the standard processing conditions set for the predicted group.This is an unvulcanized kneaded rubber processing system. [Explanation of symbols]

[0078] 1 Processing system 2. Mixing machine 3 Processing equipment 4 Arithmetic unit 10 Dataset 10a Classification Dataset 20 Predictive Models 30 Preset Data

Claims

1. A method for processing unvulcanized kneaded rubber, which comprises feeding unvulcanized kneaded rubber that has been subjected to a kneading process by a kneader into a processing device, controlling the processing device by a computing device, and processing the unvulcanized kneaded rubber by the processing device under processing conditions inputted to the computing device, comprising: In a preparatory step, feature amounts indicating the characteristics of many types of unvulcanized mixed rubber having different formulations are obtained for many types of features before the unvulcanized mixed rubber is fed into the processing device, and the feature amounts are input to the arithmetic device as a data set. A plurality of types of the features are selected from the data set as target features, and classification is performed by unsupervised machine learning using the arithmetic device based on the feature amounts of the respective target features, thereby creating a classification data set in which the data set is classified into a specified number of groups. constructing a prediction model by machine learning using the arithmetic device with the feature amount of each of the target features as explanatory variables and a predetermined number of the classified groups as objective variables, and using the classification data set as training data; Preset data is input to the arithmetic unit, which is a standard processing condition for processing the unvulcanized kneaded rubber into a good product by the processing device for each of the groups; A method for processing unvulcanized kneaded rubber, in which, when processing the unvulcanized kneaded rubber using the processing device after completion of the preparatory stage, feature quantities indicating the target features of each of the unvulcanized kneaded rubber fed into the processing device are obtained and input as input data to the arithmetic device, and the arithmetic device uses the input data and the predictive model to predict a group from among a plurality of groups into which the unvulcanized kneaded rubber fed into the processing device will be classified, and the unvulcanized kneaded rubber fed into the processing device is processed by the processing device under the standard processing conditions set for the predicted group.

2. The data set is added with the time when each of the features was acquired, A method for processing unvulcanized kneaded rubber as described in claim 1, wherein the computing device performs a time series analysis of the feature amounts of a specified target feature based on the acquisition time of each of the unvulcanized kneaded rubbers classified into the same group, and when predicting the group, the input data corrected based on the results of the time series analysis is used.

3. The processing method for unvulcanized kneaded rubber according to claim 1 or 2, wherein the classification by unsupervised machine learning is performed based on principal component scores of a large number of types of the unvulcanized kneaded rubber obtained by performing dimensionality reduction by principal component analysis on each of the target features.

4. 3. The method for processing unvulcanized kneaded rubber as described in claim 1 or 2, wherein the data set includes features having the feature amount in each of a plurality of deformation regions having the same feature type but different degrees of deformation, and when the feature type having the feature amount in each of the plurality of deformation regions is selected as the target feature, each of the deformation regions is treated as an individual feature and selected as the target feature.

5. The method for processing unvulcanized kneaded rubber described in claim 1 or 2, wherein in selecting the target feature, at least two feature groups are selected from a feature group having a plurality of features indicating the viscoelastic characteristics of the unvulcanized kneaded rubber, a feature group having a plurality of features indicating the viscoelastic characteristics, and a feature group having a plurality of features indicating the torque characteristics obtained by rotational vibration, and at least one feature is selected from each of the selected feature groups as the target feature.

6. The method for processing unvulcanized kneaded rubber according to claim 1 or 2, wherein the data set includes a compounding label that identifies the compounding of the unvulcanized kneaded rubber as the feature, and the compounding label is selected as the target feature.

7. 3. The method for processing unvulcanized kneaded rubber according to claim 1 or 2, wherein, when the processing device is an extruder, the processing conditions are a screw temperature, a conveyor speed, an extrusion die shape, a screw temperature, and a cylinder temperature; when the processing device is a rolling machine, the processing conditions are a conveyor speed, a cooling drum temperature, a gap between rolls, a roll speed, and a cut-out width; and when the processing device is a vulcanizing device, the processing conditions are a vulcanizing temperature, a vulcanizing pressure, and a vulcanizing time.

8. A processing system for unvulcanized kneaded rubber, comprising: a processing device into which unvulcanized kneaded rubber that has been kneaded by a kneader is input; and a calculation device that controls the processing device, the unvulcanized kneaded rubber being processed by the processing device under processing conditions input to the calculation device, In the arithmetic device, feature amounts indicating the characteristics of many types of unvulcanized mixed rubber having different formulations are acquired for many types of features before the unvulcanized mixed rubber is fed into the processing device, and the feature amounts are input as a dataset, and classification is performed by unsupervised machine learning using the arithmetic device based on the feature amounts of the multiple types of features selected from the dataset as target features, thereby creating a classification dataset in which the dataset is classified into a specified number of multiple groups, a prediction model is constructed by machine learning using the classification data set as training data by using the computing device, with the feature amounts of the target features as explanatory variables and a predetermined number of the classified groups as objective variables; A standard processing condition for processing the unvulcanized kneaded rubber into a non-defective product by the processing device is preset for each of the groups and input to the arithmetic device, When the unvulcanized kneaded rubber is processed by the processing device, feature amounts indicating the target features of each of the unvulcanized kneaded rubber fed into the processing device are acquired and input as input data to the calculation device, and a group into which the unvulcanized kneaded rubber fed into the processing device is classified is predicted from among the multiple groups using the input data and the predictive model, and the unvulcanized kneaded rubber fed into the processing device is processed by the processing device under the standard processing conditions set for the predicted group.This is an unvulcanized kneaded rubber processing system.

Citation Information

Patent Citations

  • Extruding method for unvulcanized rubber

    JP1995060821A