Analysis of functional group index evolution and influencing factors during natural aging of asphalt binders based on interpretable machine learning

The study designed a long-term natural aging experiment involving eight types of asphalt binders, three aging conditions, and four film thicknesses. The aging conditions were all-weather natural aging (All), thermo-oxidative-ultraviolet aging (TOU), and thermo-oxidative aging (TO). The film thicknesses were 0.80 mm, 1.59 mm, 2.39 mm, and 3.18 mm. Fourier transform infrared spectroscopy (FTIR) technology was used to analyze the effects of different aging factors on the carbonyl index (I _{C O} ), sulfoxide index ( I _{S O} ), and butadiene index ( I _{C C} ) of asphalt binders. Finally, multiple machine learning models were constructed to predict the functional group indexes. The SHapley Additive exPlanations (SHAP) method was introduced to conduct interpretability analysis on the model results. The results showed that ultraviolet radiation was a key factor in accelerating aging. Under TOU condition, the increases in I _{C O} and I _{S O} were significantly higher than other conditions, and SBS modifiers degraded most severely. The generation rate of oxygen-containing functional groups in thin layers of asphalt binder (0.80 mm) greatly exceeded that in thick layers (3.18 mm). The XGBoost model demonstrated optimal predictive accuracy and generalization capability, achieving a coefficient of determination ( R ²) of 0.9283 for the prediction of sulfoxide index. SHAP analysis further revealed that aging time and aging condition were the key factors governing carbonyl formation, whereas sulfoxide generation was more significantly influenced by asphalt type. Furthermore, feature interaction analysis indicated that UV-related aging condition and thinner film thickness both contributed to the accelerated formation of oxygen-containing functional groups during the later stages of aging.