Optimization of modified asphalt preparation design considering multiscale properties transfer: a performance-oriented approach

Variations across styrene–butadiene–styrene (SBS) types limit the generalizability of existing preparation protocols for SBS‑modified asphalt (SBSMA). Conventional design methods are often imprecise and inefficient when tuning preparation parameters to meet specified performance targets. Here, we present a performance‑oriented preparation framework that couples ensemble learning with particle swarm optimization (PSO). Three data‑driven models were developed: (i) composition and preparation parameters → micro-properties; (ii) composition and preparation parameters → macro-properties; and (iii) micro-properties → macro-properties. Among the tested algorithms, CatBoost and Extra Trees achieved the highest accuracy (R2 > 0.90, MAPE < 10%). Model interpretability indicates that appropriate shear rate and time enhance SBS dispersion, whereas excessively high shear temperature can cause polymer degradation. By integrating the predictive models with PSO, we identified preparation parameters that satisfied predefined performance targets, and SBSMA produced under these conditions met the targets with errors < 10%. This framework offers an efficient, generalizable alternative to trial‑and‑error design of SBSMA.