Machine learning-based prediction method for developing composite materials for tire tread compounds

Abstract

The present invention relates to a machine learning based predictive method for the development of composite materials for tire tread compounds.

Description

Technical Field

[0001] The present invention relates to a method for predicting the viscoelastic or processing characteristics of rubber compounds before, during, and after vulcanization, and this method is based on machine learning and is thus implemented by a computer for the development of composite materials for tire tread compounds.

Background Art

[0002] The present invention relates to the determination of the composition of rubber compounds used in the field of tire manufacturing, particularly in the manufacture of tire treads.

[0003] The RPA (Rubber Process Analyzer) is an advanced dynamic mechanical rheology test device that is generally available in all factories.

[0004] The RPA (Rubber Process Analyzer), as an advanced dynamic mechanical rheology test device, is generally available in all factories to monitor the manufacturing parameters of composite materials at each step of the process. In fact, the workability of composite materials is determined by specific ranges of rheometric curves and descriptors of shear modulus before and after curing defined at the development stage (e.g., ML and MH torque, T10, T50, and T90, scorch time, vulcanized and unvulcanized shear modulus G' and tand under certain load stress conditions).

[0005] These characteristics are ensured by the characteristics of the recipe used for the composite material, particularly the components, their amounts, and the specific synergistic effects established between two or more of them.

Summary of the Invention

Problems to be Solved by the Invention

[0006] Generally, determining the correct formulation for a composite material requires several validation steps in the laboratory to find a suitable technical package, followed by optimizing the formulation through stepwise fine-tuning until the objective is fully achieved.

[0007] Each of these iterative experimental campaigns increases the lead time and cost (time to market) of product development from a product perspective, and from a data perspective, it generates a database with inherent variability due to random noise in the measurements taken during various test campaigns.

[0008] Predicting product performance under the above conditions typically requires extensive laboratory testing to validate the compound, which is time-consuming and resource-intensive.

[0009] Therefore, an object of the present invention is to solve these problems that remain unresolved by the prior art by providing a process as defined in claim 1 of the patent claims.

[0010] In particular, an object of the present invention is to simulate laboratory tests in order to provide accurate estimates of some of the important viscoelastic and processability properties of composite materials for the manufacture of rubber compounds for tires, without the need to perform any physical tests.

[0011] A further object of the present invention is a composite material analyzer (RPA) as described in claim 8 of the patent.

[0012] Further features of the present invention are defined in the corresponding dependent claims.

[0013] By using software tools that can predict the behavior of composite materials, and consequently the performance of tires, the following becomes possible: • Significant reduction in operating costs (raw materials, labor costs, etc.). • Optimization of the capacity and quality of laboratory testing (allowing personnel to be allocated to other activities). • Shorter time to market for new products. • Improved prediction accuracy compared to known methodologies.

[0014] Other distinct advantages over the prior art, along with the features and applications of the present invention, will become apparent from the following detailed description of preferred embodiments, which are given as purely non-limiting examples. [Brief explanation of the drawing]

[0015] Refer to the attached diagram. [Figure 1A] This is a schematic block diagram of the training steps for the machine learning and characteristic prediction algorithm according to the present invention. [Figure 1B] This is a schematic block diagram of the training steps for the machine learning and characteristic prediction algorithm according to the present invention. [Figure 2] This graph is useful for verifying the time required for the vulcanization torque to reach a predetermined increase.

[0016] [Theoretical background] Polymer matrix composites are unique materials that exhibit characteristic elastic and viscous responses when subjected to stress.

[0017] Predicting the rheometric curve of a composite material based on one of its fundamental parameters (e.g., torque ML and MH, T10, T50 and T90, scorch time, vulcanized and unvulcanized shear modulus G', and tand under fixed shear conditions) is fundamental to determining and evaluating the workability of the composite material from mixing to extrusion and vulcanization steps, and to avoid problems such as mixer downtime, defects in extruded products, press clogging during vulcanization, or under-vulcanization / over-vulcanization.

[0018] The processability characteristics are evaluated by performing rheometric tests in multiple steps. Some of the process parameters tested in these factories have recently become associated with performance parameters by a specific evaluation, so its prediction has become even more important to estimate the variability of factory performance. Such an evaluation requires multiple laboratory tests to reach the verification of composite materials and requires time and resources.

[0019] On the other hand, using a digital prediction device makes the following possible. · Reduction of recurring costs (raw materials, labor costs, etc.). · Optimization of the workload and quality of the test laboratory (personnel can be concentrated on other activities). · Reduction of the time to market of new products.

[0020] Therefore, potential end users are all engineers and laboratory experts who may benefit from this device.

[0021] As expected in the above paragraph, the processability test device is also available in factories for monitoring composite materials to meet quality standards. The present invention can also be extended to and released for factories as end users, and the factory technical service can significantly shorten the time to evaluate the change of formulation characteristics during formulation development or simply improve the processability response of research and development to solve problems that may occur in factories with limited production downtime / loss.

[0022] The Rubber Process Analyzer (RPA) is a valuable machine designed to measure the viscoelastic or processability characteristics of polymers and composite materials before, during, and after vulcanization. The vulcanization characteristics can be determined by measuring the characteristics as a function of time and temperature. The tests can be performed under different conditions according to the required test methods, and the measurements of G’ and tand can be continuously recorded as a function of time and / or strain applied by periodic torque at different shear rates.

[0023] Some of the outputs of each test are selected within a range based on the cardinality in the dataset and the importance when the engineer evaluates the processability. 1. ML, low torque of the vulcanization curve, elastic modulus of the green (unvulcanized) compound. 2. MH, maximum torque of the vulcanization curve, or the torque when the curve rises to the plateau, elastic modulus of the vulcanized compound. 3. T10, T50, T90 are the times to reach the homologous % of ML + ΔT, where ΔT is the torque difference between MH - ML. 4. Ts, that is, the scorch time, the time to reach the vulcanization setting + 1 dNm torque increment (see Figure 2). 5. Strain G'@100% before vulcanization, which is a parameter related to the viscosity of the green (unvulcanized) composite material. 6. Strain G'@1% after vulcanization, strain tand@15%, which is related to the dynamic properties of the vulcanized compound used to predict the tread performance of the tire. 7. Strain G'@50% after vulcanization, related to the static properties at high strain.

Mode for Carrying Out the Invention

[0024] Hereinafter, the present invention will be described with reference to the above figures.

[0025] Therefore, a methodology for predicting the viscoelastic or processability characteristics (e.g., ML and MH torques, T10, T50, and T90, scorch time (Ts), vulcanized and unvulcanized strain elastic moduli G' and tand at the loading strain) of composite materials that can be used in the production of rubber compounds for tires will be described.

[0026] As a general theory, a methodology for predicting the processability characteristics such as the ML and MH torques, T10, T50, and T90 curing times, scorch time, shear elastic modulus, and tand at the loading strain of a new composite material that has never been mixed before will be described according to the following procedure. • Collect existing recipes and their corresponding viscoelastic or processability properties (i.e., ML, MH, T10, T50, T90, TS, G'@100%, G'@1%, G'@50%) to create a database and obtain a primary dataset. • The procedure for integrating the primary dataset through data augmentation to obtain the augmented dataset. • Steps to transform the augmented data in order to obtain the transformed dataset. • Training machine learning algorithms (e.g., linear mixed models) using the transformed dataset. • Prediction of viscoelastic or processability properties of new recipes to be tested using trained algorithms.

[0027] In particular, performing the dedicated "step of training the model" of a machine learning algorithm on an extended and transformed dataset makes it possible to predict processability characteristics with higher accuracy than directly applying the algorithm to the original dataset.

[0028] In fact, this method can significantly reduce the impact of random noise in the database on prediction accuracy.

[0029] In particular, after the step of training the model using the dataset contained in the expanded and transformed database as described above, it is possible to predict the viscoelastic and processability properties of composite materials with higher accuracy than applying the algorithm directly to the raw data of the database.

[0030] In fact, this approach significantly reduces the impact of database noise and inherent data variability on prediction accuracy.

[0031] This device is characterized by the implementation of the following actions and algorithms. 1. Collection of representative laboratory data for the recipe of the composite material to be tested. 2. Data Augmentation Procedure: This procedure can enhance the predictive capabilities of machine learning algorithms by utilizing various methods for introducing new characteristic parameters. 3. Data Transformation Procedure: This procedure can enhance the predictive capabilities of machine learning algorithms by utilizing various transformation methods. 4. Machine learning algorithm (e.g., linear mixed model algorithm): This algorithm aims to predict the processability characteristics of rubber compounds. The steps of the method according to the present invention will be described with reference to illustrative figures 1A and 1B.

[0032] This process includes the following steps, following the generation of the primary dataset, as already explained. 1. Data Integration from Primary Datasets (Data Augmentation): This step aims to develop new characteristic parameters that must be added to the primary dataset (e.g., a list of ingredients and their quantities for each recipe) in order to generate the augmented dataset. This means that for each recipe, a set of new characteristic parameters will be estimated to improve the predictive power of the machine learning model. The new characteristic parameters will be derived from the application of a set of heterogeneous techniques, such as clustering algorithms, nonlinear operators applied to the input set, and specially designed and customized artificial neural networks, as will be discussed later.

[0033] 2. Transformation of data included in the augmented dataset: Similar to the previous step, this aims to improve predictive performance. The difference is that this step focuses on the transformation of the data rather than adding new characteristic parameters. As detailed below, the data transformation algorithm employed is based on spline functions and Box-Cox transformations.

[0034] 3. Training Steps for Machine Learning Algorithms: The data from the augmented and transformed dataset (compound recipe and corresponding characteristic parameters) is provided as input to a predictive model implemented by a machine learning method (e.g., a mixed linear model). The intrinsic parameters of the model are then adjusted to fit the data (training step).

[0035] 4. Prediction Step: Once training is complete, the algorithm can generalize its results, making it suitable for predicting the processability properties of composite materials for new recipes of composite materials that have never been mixed before (prediction step).

[0036] The original dataset to which dataset integration (data augmentation) and transformation (dataset transformation) procedures have been applied is defined as a preprocessed dataset.

[0037] (Data integration procedure for primary dataset) Additional features were introduced to improve the prediction of the processability properties of composite materials. In fact, it has been observed that the information obtained from the composite material formulation is insufficient to achieve the goal of property prediction.

[0038] For example, mixed conditions introduced as characteristic parameters have been observed to play a significant role and be highly beneficial. Therefore, they are used to augment the original dataset.

[0039] On the other hand, models formed with too many parameters are characterized by a lack of generalization in prediction. They may be able to predict on the training dataset, but they cannot correctly predict on new data, i.e., on a new test dataset. This phenomenon is called overfitting, and when overfitting occurs, it becomes impossible to generate a model that can run in the production step.

[0040] Therefore, a technique called "data augmentation" was developed to improve prediction performance while avoiding overfitting.

[0041] The data augmentation procedure includes estimating one or more of the following characteristic parameters and integrating them into the primary dataset. • <Estimation of the Gini coefficient for each recipe>: Also known as the Gini index or Gini ratio, it is a measure of statistical variance and is applied directly to each recipe. This characteristic parameter was added to provide an explanation of the hidden structure of the formulation.

[0042] • Definition of Mixing Categories for Each Recipe: A category of the mixing method for composite material recipes has been introduced as a new characteristic parameter of the category type. This characteristic parameter was added to provide information about the mixing conditions of the components (e.g., geometric parameters of the mixer itself, temperature, time, and rotor speed) to which the predicted physical properties will be directed. A more comprehensive way to provide this information might be achieved by directly introducing all mixing parameters into the dataset, but our strategy is to use only the category type parameters to provide a lean solution and avoid overfitting. In doing so, all mixing parameters are summarized by a single parameter of the category type that indicates the type, and thus indicate the applied mixing category type, such as tangential or interpenetration screw mixer.

[0043] • Definition of Composite Material Application Types for Each Recipe: The application of composite materials (e.g., tire components or the category of vehicle on which the tire is mounted) has been introduced as a new category-specific characteristic parameter. This characteristic parameter was added to provide information about the final application of the composite material and to provide a category-specific preview of the macro requirements expected when the composite material is developed. In this way, all expected macro requirements are summarized by the category type parameter that indicates the type of application in which the composite material will ultimately be used. In reality, various applications require different macro requirements. The types of composite material applications are, for example, for automobiles or commercial vehicles, where the pressure and temperature conditions during vulcanization differ from each other.

[0044] • <Interaction between mixing method and composite material application type for each recipe>: All possible combinations between the mixing category and the composite material application are estimated and used to define new characteristic parameters for each recipe's category type. These characteristic parameters for the category type are introduced to explicitly describe the potential effects of mixing conditions on the composite material application, i.e., the macro-requirements assumed for the composite material. In fact, the two variables of the category type are connected, since the mixing category can influence the physical properties of the composite material, and the type of composite material application provides a description of the macro-requirements for the composite material. This connection is described through this new category variable that represents all possible combinations between the mixing category and the composite material application.

[0045] • <Estimated total ingredient amounts for each recipe>: This represents the total amount of each ingredient and is estimated for each recipe. Thus, the amount of each ingredient is always related to the total amount and differs from recipe to recipe.

[0046] • <Estimation of component ratios for each recipe>: An exhaustive search was conducted to determine how the ratios of different components correlate with the predicted processability properties. Therefore, the estimated component ratios that were shown to correlate with the target properties were added as new characteristic parameters. These characteristic parameters were added to provide an explicit explanation for the nonlinear component terms that are useful in predicting the target properties.

[0047] · <Louvain grouping by the Louvain method>: The method by the Louvain method was used to group recipes based on the co-occurrence of components. In fact, the Louvain unsupervised algorithm was attached to the recipes of the compounds constituting the available dataset. As an example, the recipes can be grouped according to the presence (co-occurrence) of synthetic rubber and silica rather than the co-occurrence of natural rubber and carbon black. Therefore, each recipe was grouped using the Louvain method, and as a result, new categorical characteristic parameters were estimated for the assigned clusters. This categorical parameter corresponds to the identifier of the grouping / cluster itself. This characterization parameter was introduced to provide an explanation for the non-uniformity of the primary dataset. In fact, the clustering algorithm can group different recipes based on the co-occurrence of components. This means that by integrating such grouping information into the dataset, it may be possible to explicitly provide insights into how components are used in different recipes.

[0048] · <Grouping by the K-means method>: Each recipe was grouped using the method by the K-means method, and as a result, categorical type characteristic parameters were estimated for the assigned clusters. The categorical parameter corresponds to the identifier of the grouping itself. The grouping algorithm by the K-means method was attached to the recipes of the composite materials constituting the available dataset. Optimization by grid search was implemented to find the optimal hyperparameter optimization settings, that is, the number of clusters to identify or the optimization algorithm to implement. This characteristic parameter was introduced to provide an explanation for the non-uniformity of the primary dataset. In fact, the clustering algorithm can group different recipes according to the criterion of the distance in the space of components. That is, by introducing such grouping information into the primary dataset, it is possible to explicitly provide insights into how components are used in different recipes.

[0049] • Dimensionality Reduction via Autoencoders: An autoencoder (AE) is a special type of unsupervised artificial neural network trained to copy inputs to outputs. To achieve this, the AE can first map the input to a small latent space, and then encode the latent representation to the output. As a result of this operation, the AE is trained to compress the data by reducing reconstruction errors. Grid search optimization was used to find the optimal number of dimensions for the compressed data, i.e., the number of dimensions in the latent space. A so-called dimensionality reduction algorithm using autoencoders was developed to process a set of data formed by composite material recipes to create a dimensionally reduced, i.e., compressed representation. This compressed data, while having reduced the dimensionality of the original recipe, was introduced as a new characteristic parameter because it is a very useful representation.

[0050] According to a preferred embodiment, all of the above characteristic parameters are integrated into a primary dataset.

[0051] (Conversion of extended data) Performing appropriate transformation operations before using data to train machine learning algorithms significantly improves prediction accuracy.

[0052] The procedure for transforming data present in the augmented dataset involves using one or more of the following transformation functions on the feature components and / or parameters. • B-spline smoothing: The numerical data of the augmented dataset was smoothed using the B-spline function before the next training step. An optimized grid search algorithm was used to select the components and / or characteristic parameters to transform.

[0053] • Box-Cox transformation: The Box-Cox transformation was applied to the numerical data in the augmented dataset to approximate the distribution of a normal distribution.

[0054] • Scaling transformation: To improve subsequent training steps of the machine learning model, all numerical data in the augmented dataset was scaled to the same numerical range.

[0055] Applying one or more of the above transformation functions generates a transformed dataset.

[0056] According to a preferred embodiment of the present invention, all of the above transformation functions are used to process the recipes of the primary dataset.

[0057] Furthermore, according to a further embodiment, the aforementioned conversion functions constituting the data conversion procedure must preferably be performed in the proposed order.

[0058] In these embodiments, all the steps described are incorporated into the preprocessing pipeline because their synergistic interaction can maximize predictive performance.

[0059] The transformed (preprocessed) dataset thus obtained can be used to initiate a more general training step of a machine learning model after the data augmentation and transformation procedures have been performed. According to the embodiments described herein, a mixed linear model is implemented and trained to provide a predictor of processability properties. Nevertheless, any other machine learning model may be used to make final predictions as soon as the primary dataset has been processed as described.

[0060] Finally, the same preprocessing steps (data augmentation and data transformation) are applied to data related to or representative of the recipe of the composite material to be tested (Figure 1B) before it is fed as input to a machine learning algorithm that has already been trained in order to predict the viscoelastic or processability properties of the composite material to be tested.

[0061] The present invention has been described herein with reference to its preferred embodiments. Purely as examples, each of the technical features implemented in the preferred embodiments described herein may be advantageously combined with other features in ways not described herein to form other embodiments belonging to the same inventive concept, all of which are intended to fall within the scope of protection provided by the claims set forth below.

Claims

1. A method performed using a computer to predict the viscoelastic or processability properties of a composite material to be tested for the manufacture of a tire tread compound, a) A step of preparing a database of raw data to be used as a reference, i.e., a primary dataset including recipes for existing composite materials and corresponding known viscoelastic or processability properties, b) A step of preprocessing the primary dataset, i. A step of integrating one or more characteristic parameters in the primary dataset, wherein the one or more characteristic parameters are as follows: • Gini coefficient for each recipe Mixed categories for each recipe • Types of application for each recipe • Total amount of ingredients for each recipe • Ratio of ingredients in each recipe - Grouping of the composite material recipe using the Reuven method - Grouping of the composite material recipe using the K-mean method, and - Data whose dimensions have been reduced through an autoencoder applied to the primary dataset formed by the composite material recipe. The integration step is to select from the following, thereby obtaining an extended dataset. ii. A transformation step of the extended dataset, wherein one or more transformation functions are applied to the components and / or numerical characteristic parameters of the extended dataset, the one or more transformation functions being:

1. B-spline smoothing 2. Box-Cox conversion, and 3. Scaling Transformation The transformation step is selected from the following, thereby obtaining the transformed dataset. The preprocessing step, which is performed by integrating one or more characteristic parameters in the primary dataset, includes the integrating step of all the characteristic parameters shown in step i. c) A step of training a machine learning-based algorithm using the data from the transformed dataset, d) Applying the algorithm trained according to step c) to a set of data that has been preprocessed according to step b) and represents the recipe of the composite material to be tested, in order to predict the viscoelastic or processability properties of the composite material to be tested, A method that includes [a certain feature].

2. A method according to claim 1, wherein the viscoelastic or processability characteristics include torque ML and MH, T10, T50 and T90, scorch time (Ts), vulcanized and unvulcanized shear modulus G' and tan under imposed conditions.

3. A method according to claim 1, wherein the step of transforming the extended dataset includes the step of applying all of the transformation functions shown in ii.

4. The method according to claim 3, wherein the conversion function shown in ii. is executed sequentially.

5. The method according to claim 4, wherein the conversion function shown in ii. is performed in the order shown (1., 2., 3.).

6. A method according to claim 1, wherein the machine learning-based algorithm is configured to apply a mixed linear model.

7. A composite material analyzer (RPA) configured to carry out the method described in any one of claims 1 to 6.

Images