A posterior hybrid ML optimization to analyze the relationship between high temperature rheological factors of the Superpave PG and improved PG + asphalt specifications

Abstract

This study investigated the relationship between the parameters of the Superpave PG asphalt classification method and the improved MCSR PG-Plus asphalt classification method with two hybrid fuzzy models based on deep learning artificial neural network (ANN) and Support Vector Machine (SVM). For this purpose, virgin asphalt was modified with four well-known polymers namely, styrene butadiene styrene (SBS), styrene butadiene rubber (SBR), ethylene vinyl acetate (EVA) and high-density polyethylene (HDPE) at various modification percentages. In order to collect data, a total of 294 high temperature rheological experiments were carried out on neat and modified samples. Complex modulus (), phase angle (), percent recovery () and the non-recoverable creep compliance () were determined at five different temperatures to construct dataset. The relationship between, and factors two models were utilized namely ANN_FL and SVM_FL. Results showed that the ANN based model outperforms the SVM based hybrid model with a value of 0.96 R2 and 0.83 RMSE. The uncertain situation behind road traffic classification were interpreted with the help of FL. Regarding the experimental procedure and machine learning (ML) model results, it has been concluded that the all ML methods were represented the meaningful relation between Superpave PG and MSCR PG-Plus asphalt classification methods.