Prediction of Optimum Bitumen Content in Asphalt Mix Design Using Artificial Neural Network
Date
2024-07
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
I.O.E
Abstract
The Marshall design process, commonly employed for estimating Optimum Bitumen Con-tent (OBC), is known for its designation as the asphalt mix design and quality control of asphalt concrete is often constrained by the conventional Marshall Mix Design methodol-ogy. Which is characterized by its time-intensive nature, labor requirements, and suscepti-bility to result variations. This study explores different predictive modeling techniques to enhance the efficiency and accuracy of determining Optimum Binder Content (OBC), Me-chanical and Volumetric properties of hot mix asphalt. The study examines Multiple Linear Regression (MLR), Artificial Neural Networks (ANN), Support Vector Machine (SVM), and Random Forest (RF) models to predict mechanical (MST, MFV) and volumetric prop-erties (AV, BSG, VFA, VMA) using variables such as aggregate gradation, specific grav-ity, and proportions of fine and coarse aggregates in the mix. A comprehensive dataset of 148 Marshall mix design forms was collected, and 141 valid sets were used after outlier analysis. Descriptive statistics revealed skewness, and compliance checks against SSRBW 2016 standards highlighted areas of concern in Marshall quotient and Filler-Binder Ratio. Centrality analysis showed significant deviations between mid-point gradation and sample mean for specific gradations.
Description
The initial MLR model demonstrated moderate predictive capabilities with R² values rang-ing from 0.3017 to 0.6794 for various parameters, indicating limitations in accurately pre-dicting complex asphalt properties. To address these limitations, the integration of ANNs significantly improved predictive accuracy, achieving higher R² values for all parameters compared to MLR.
Keywords
Hot Mix Asphalt, Marshall Mix Design, Optimum Binder Content, Support Vector Machine