Optimizing asphalt compaction: Vibratory roller amplitude and predictive modeling

Abstract

Effective compaction quality significantly impacts pavement durability and performance, with uneven compaction often resulting from traditional empirical approaches that adjust vibration modes and rolling periods. This paper investigates the effects of vibratory roller amplitude optimization on asphalt pavement compaction and develops predictive models for intelligent compaction (IC) parameters and in-place density using machine learning (ML) methods. Results indicate that higher-amplitude vibration passes yielded higher average intelligent compaction measurement values (ICMVs) and in-place density. Predictive models were developed using XGBoost and CatBoost, optimized through the Bayesian Optimization Algorithm (BOA). Among them, the XGBoost models achieved the best performance in predicting ICMVs and non-nuclear density (NND) values, demonstrating high accuracy (R² = 0.9362, 0.9916) and low error (RMSE = 4.5678, 1.1934) during validation. These findings have significant implications for compaction quality control and provide a foundation for future research to enhance predictive models for ICMVs and NND under diverse construction conditions.