Evaluation and modelling of permanent deformation behaviour of asphalt mixtures using dynamic creep test in uniaxial mode

Abstract

Listen

Translate article

Permanent deformation behaviour of asphalt mixtures is likely to be examined in laboratory prior to laying those mixtures in field. This study aimed at relative evaluation and modelling of permanent deformation behaviour of various wearing course mixtures using advanced statistical techniques. Thirty wearing course mixtures with variable combinations of constituent materials and gradations were tested at three temperature conditions (i.e. 25, 40 and 50 °C) and two stress conditions (i.e. 300 and 500 kPa) using uniaxial dynamic creep test (DCT). Consistent ranking has been observed at various test temperatures and stress conditions. Variation of permanent strain slope with number of loading cycles was found to be a useful parameter for asphalt mixture laboratory performance. It has been observed that permanent strain slope followed nearly similar pattern when stress was increased from 300 to 500 kPa, at the temperature condition of 25 °C. The waviness among permanent strain slope values has been increased significantly with the increase in stress from 300 to 500 kPa for the results at temperature conditions of 40 °C and higher. Non-linear regression and artificial neural networking (ANN) have been found as effective techniques to model the permanent strain in uniaxial DCT. Temperature, number of cycles, bitumen penetration, aggregate flakiness index, and stress condition are considered as significant independent variables included in the model. Normalised importance plots indicated temperature as most significant variable for the developed models. It has been found that non-linear regression model most accurately predicts permanent strain data of NHA-A graded mixtures while, ANN model most precisely estimates permanent strain data for SP-B graded mixtures as translated by larger R2 values of 0.88 and 0.99, respectively. The statistical parameters of R2, RMSE, VAF and MAPE indicated high prediction performances of artificial neural network (ANN) modelling technique as compared to nonlinear regression modelling technique.