Adaptive Neuro-Fuzzy Inference System to Predict the Dynamic Modulus of Hot Mix Asphalt

Abstract

The dynamic modulus |E*| of hot mix asphalt (HMA) plays a fundamental role in the mechanistic–empirical pavement design. The Witczak regression-based predictive model can be considered as the most fundamental and widely used model to estimate the |E*| of HMA. However, the effect of confining stress has not been considered in this model. In this paper, the feasibility of applying the robust machine learning technique called “adaptive neuro-fuzzy inference system” (ANFIS) was investigated to predict the |E*| of HMA using 1,320 test results performed at the University of Maryland. Asphalt mix parameters, testing frequency, temperature, and the level of confining stress were considered as the model inputs. Also, intercept of temperature susceptibility relationship (A) and slope of temperature susceptibility relationship (VTS). Two new ANFIS models were developed using two different structures, including fuzzy C-mean clustering (FCM) and subtractive clustering (SC) algorithms. In addition, two computer programs were developed for optimizing the structure of the FCM ANFIS and SC ANFIS models to achieve the highest predicting accuracy. The obtained results indicate that the SC and FCM ANFIS models predict the |E*| of HMA with high coefficients of determination of R2=0.948 and 0.945, respectively. Moreover, a sensitivity analysis was conducted to evaluate the level of influence of model inputs, and the results show that the confining stress has a significant impact on the |E*|. Also, the results demonstrate that the standard results for the ANFIS model and real data are in complete agreement.