Recalibration and Development of Prediction Models For Dynamic Modulus Of Bituminous Mixtures

Abstract This study was conducted to predict the dynamic modulus of Pakistan’s regionally prevalent bituminous mixtures using a wide range of material properties, mix design parameters, and testing conditions. A total of twenty-four bituminous mixes, characterized by sixteen distinct gradations and four different types of bitumen, were tested for dynamic modulus at various temperatures and frequencies. Three different types of models were used on the experimental data, namely, the traditional Witczak 1-37 model, recalibrated Witczak 1-37 model, and the recalibrated Witczak 1-37 model with dummy variables. Recalibration was done using 75% of the data, while the remaining 25% was used for testing. In each case, the model’s accuracy was reported using R-squared values and two error parameters. The comparison of the models clearly demonstrated that the traditional Witczak model had the lowest accuracy in terms of the error parameters with error terms exceeding 3. Recalibrated models with and without the dummy variables, on the other hand, produced comparable results, with errors less than 0.5. The recalibration revealed an insignificant impact of fine material on stiffness. It also showed that 60/70 asphalt from a specific source has a significant impact on the model, while other types had an insignificant effect. The study findings are expected to be equally beneficial to both researchers and practitioners working on pavements mechanistic-empirical design. Future study might include incorporating other parameters and the most recent machine learning techniques to build models for the prediction of locally utilized bituminous mixes in Pakistan.