Improved Asphalt Binder Performance Grading Using Surrogate FTIR Data and Gradient Boosting

This study develops a novel framework integrating surrogate spectral data enrichment with gradient boosting for high-throughput performance grading (PG) of asphalt binders. Addressing the limitations of Fourier-transform infrared (FTIR) spectroscopy in quantifying the complex chemo-mechanical correlation of asphalts and the scarcity of diverse laboratory samples, we generate statistically representative surrogate datasets by fitting input variables derived from FTIR spectra to the observed distributions and applying the Kriging-based surrogate (KS) approach to the original samples. The enriched dataset, combined with the original data, trains a robust KS-XGBoost model to predict high-temperature PG (HTPG) and low-temperature PG (LTPG). Results demonstrate that KS-XGBoost significantly outperforms standard XGBoost, achieving testing R2 values of 0.976 for HTPG and 0.983 for LTPG, with sharply reduced errors. SHAP analysis further reveals nonlinear, threshold-dependent roles of sulfoxide (dominant for both HTPG and LTPG) and butadiene (critical for LTPG), aligning with polymer modification mechanisms. This approach enables accurate performance grading of asphalt binders, accelerates binder research while reducing experimental burden, and serves as a complement to confirmatory laboratory tests.