Machine learning-aided rheological prediction models of asphalt binders based on chemical properties

Aug 5, 2026

Rapid rheological characterization of asphalt binders is enabled by linking chemical properties to prediction models using advanced machine learning. Fourier transform infrared spectroscopy (FTIR) and a dynamic shear rheometer (DSR) are used to measure chemical and rheological responses. The six raw FTIR features are compressed into two principal components, with PC1 showing greater variance contribution than PC2. Multiple linear regression accurately forecasts phase angle but struggles with modulus, while Gaussian process regression achieves both higher R2 and lower RMSE for modulus and phase angle.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages4

Reading Time10 min

AccessFree to download

Introduction

Viscoelastic rheology and performance impact

Challenges of conventional rheological testing

Role of chemical composition in rheology

Motivation for machine learning with FTIR

Materials and Methods

Materials

DSR tests

Machine learning-aided rheological prediction models of asphalt binders based on chemical properties F. Zhang 1 , D. Wang1,2 , Y. Sun 1 , A.C. Falchetto 1,3 1Department of Civil Engineering, Aalto University, Finland 2Department of Civil Engineering, University of Ottawa, Canada 3Department of Civil Environmental and Architectural Engineering, University of Padova, Italy ABSTRACT: This work aims to provide rapid rheological characterization of asphalt binders through their chemical properties based on advanced machine learning tools. With this objective, Fourier transform infrared spectroscopy (FTIR) and dynamic shear rheometer (DSR) are adopted to measure the chemical and rheological properties. Results indicate that the raw six FTIR features can be reduced to two principal components (PC 1 and PC 2), and the variance and role of PC 1 are more significant than PC 2. Multiple linear regression models can predict the phase angle accurately but not for modulus. Gaussian process regression model with higher R2 and lower RMSE values can accurately predict both modulus and phase angle. Keywords: asphalt binders; rheological properties; chemical composition; machine learning 1 INTRODUCTION Rheology of asphalt binders is the study of their flow and deformation behavior under various temperature and loading conditions (Zhang et al., 2024) . It is a crucial aspect of understanding the performance characteristics of asphalt materials used in pavements. Asphalt binders exhibit viscoelastic properties, meaning they show both viscous (liquid-like) and elastic (solid-like) responses depending on the temperature and rate of loading. At high temperatures, they behave more like viscous fluids, while at low temperatures, they act more like elastic solids. The rheological properties of asphalt binders directly influence the durability and performance of road pavements, impacting resistance to deformation (rutting), cracking, and fatigue. Accurate rheological analysis helps in selecting and modifying binders to meet specific climatic and traffic demands, ensuring better long-term performance of asphalt pavements. Measuring the rheology of asphalt binders can bea time-consuming process due to the complex and detailed analyses required to fully characterize their viscoelastic behavior. Rheological testing often involves conducting multiple assessments, such as dynamic shear rheometer (DSR) tests and bending beam rheometer (BBR) tests, across a range of temperatures and loading frequencies. These tests are designed to simulate the real-world performance of asphalt binders under different traffic and climatic conditions, which requires precise sample preparation, conditioning, and repeated measurements to ensure reliability. Additionally, each test can take significant time due to the need for temperature equilibration, application of controlled stress or strain, and data collection. The time investment is necessary to capture the binder’s behavior over short-term (high traffic speed) and long-term (slow-moving or stationary loads) performance. While these processes provide invaluable insights into the binder’s potential durability and suitability for specific applications, they require a considerable commitment of time and resources (Wang et al., 2022) . The chemical properties of asphalt binders play a fundamental role in determining their rheological behavior and overall performance (Shan et al., 2023) . Asphalt binders are composed of complex mixtures of hydrocarbons, including asphaltenes, resins, saturates, and aromatics (Salehfard et al., 2024), each contributing differently to their physical characteristics. The balance between these chemical constituents influences the binder's response to temperature changes and mechanical stress. For instance, asphaltenes contribute to the stiffness and elasticity of the binder (Ilyin and Yadykova, 2024), while lighter fractions, such as saturates and aromatics, provide fluidity and flexi

FAQs

How are FTIR and DSR used in the study?

FTIR measures chemical characteristics of asphalt binders, while DSR provides rheological data needed to relate chemistry to rheological behavior.

What happens when the six FTIR features are reduced for modeling?

The six raw FTIR features are reduced to two principal components (PC1 and PC2), and PC1 contributes more variance and interpretive importance than PC2.

Which machine learning model best predicts both modulus and phase angle?

Gaussian process regression outperforms multiple linear regression, yielding higher R2 and lower RMSE for both modulus and phase angle.

Aug 5, 2026

Upload by Evangeline