Abstract:
The performance and longevity of asphalt pavements are fundamentally influenced by the properties of their constituent materials, particularly aggregates and asphalt binder. The conventional Marshall mix design method, while widely used, requires extensive laboratory testing to achieve optimal stability and flow, leading to increased costs and time constraints for road construction projects. This main objective of this study is to addresse these challenges by developing a predictive correlation model that links aggregate and volumetric properties directly to Marshall Stability and Marshall Flow parameters, aiming to streamline the mix design process and enhance pavement quality.
A total of 211 historical mix designs were collected from both public and private institutions, with 159 valid datasets retained after screening. Each record contained asphalt binder content (Pb), bulk specific gravity of aggregate (Gsb), maximum theoretical specific gravity (Gmm), bulk specific gravity of compacted mix (Gmb), air-voids (Va), voids in mineral aggregate (VMA), voids filled with asphalt (VFA), effective specific gravity of aggregate (Gse), and the corresponding Marshall test outputs (MS and MF).
Pearson correlation analysis showed that Gmb had the strongest positive correlation with MS (r ≈ +0.60), while VFA exhibited the highest positive relationship with MF (r ≈ +0.68). Regression models were developed using ordinary least squares with backward elimination, supported by a 70:30 training-to-testing data split and Mallows Cp/AICc criteria to avoid overfitting. The final MS model retained four predictors (Gsb, Gmb, VFA, Gse) with an adjusted R² of 53.3% and test R² of 48.8%, while the MF model included five predictors (Pb, Gsb, Gmb, VFA, Gse), yielding an adjusted R² of 56.3% and test R² of 53.9%.
MS (kN) = –44.58 – 17.5 Gsb + 125.5 Gmb – 0.162 VFA – 66.4 Gse MF (mm) = 2.58 – 0.276 Pb + 4.30 Gsb – 10.46 Gmb + 0.062 VFA + 3.86 Gse
Both models were statistically significant (F > 27, p < 0.001), met regression assumptions, and demonstrated prediction errors of ±2.42 kN (MS) and ±0.28 mm (MF). Gmb emerged as the
xv
dominant predictor, positively affecting MS and inversely influencing MF, while VFA served as a key moderating factor in both models.
It is concluded that aggregate and volumetric properties particularly Gmb, Gse, Gsb, VFA, and Pb can reliably predict Marshall Stability and Flow, offering a practical and cost-effective alternative to full-scale laboratory testing for performance-based asphalt mix design. Further research is recommended to incorporate environmental and traffic-loading variables, and to explore nonlinear or machine learning approaches for improving prediction accuracy and extending model applicability to broader pavement design contexts.
