Properties Prediction and Design of Tire Tread Composites Using Machine Learning. Issue 3 (26th February 2020)

• Record Type: Journal Article
• Authors: Pang, Song; Luo, Jinlian; Wu, Youping
• Is Part Of: Macromolecular theory and simulations. Volume 29:Issue 3(2020)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2020-02-26
• Subjects: machine learning -- multiple linear regression -- rubber materials -- tire treads; Macromolecules -- Periodicals; Polymers -- Periodicals; Polymerization -- Periodicals; Macromolécules -- Périodiques; 547.705
• DOI: 10.1002/mats.201900063
• Languages: English
• ISSNs: 1022-1344

Abstract

Abstract: The present study focuses on exploring the relationship between various properties of tire tread composites and filler system using machine learning. Four different types of machine learning algorithms, such as multiple linear regression (MLR), artificial neural network (ANN), support vector machine regression (SVR), and classification and regression tree, are used for predicting 0 °C tanδ, 60 °C tanδ, tensile strength, and Shore A hardness of natural rubber nanocomposites from carbon nanotubes dosage, silica dosage, and total filler equivalent. The results showed that the introduction of interaction terms and square terms into the inputs evidently improved the prediction capability of MLR, ANN and SVR, and MLR possessed the smallest prediction errors (<5%). The established MLR models are further used to design tire tread composites with high 0 °C tanδ, low 60 °C tanδ, and appropriate Shore A hardness and tensile strength. The predicted values are in good agreement with the experimental results, indicating that the established MLR models can be used for properties prediction and design of tire tread composites effectively. Moreover, k‐fold cross‐validation is proved to be a reliable technique to evaluate the predictive capability of the MLR models. Abstract : The present study focuses on exploring the relationship between various properties of tire tread composites and filler systems using machine learning. Four different types of algorithms are compared and multiple linear regression (MLR) models are proved to possess the smallest prediction errors (<5%). According to the established MLR models, tire tread composites with high comprehensive properties are successfully designed.