Modelling of Deflection Basin Parameters of Asphalt Pavements Using Artificial Neural Networks and Adaptive Neuro-Fuzzy Inference Systems

Abstract

Non-destructive testing equipment, such as the Falling Weight Deflectometer, offers crucial evaluations of the structural state of the road and enhances pavement management systems. Various approaches based on pavement surface deflection measured using Falling weight deflectometers are widely used around the world for assessing structural stability. The backcalculation of pavement layer moduli has been a widely recognized approach for assessing the structural adequacy of the pavement. However, consistently performing these tests at the network level is laborious, and the subsequent interpretation of the data requires technical expertise, a great deal of time, finance, and other resources. Because of this structural component of roadways, decisions when choosing between maintenance and repair are often neglected. This study uses a variety of structural, functional, environmental, and subgrade soil properties as input parameters to develop a trusted relationship for the estimation of seven different deflection basin parameters such as surface curvature index, Base Curvature Index, Base Damage Index, Area Under Pavement Profile, Deflection Ratio, Shape factors F1 and F2. An effective model was developed using artificial intelligence-based soft computing techniques; Artificial Neural Networks (ANN) and Adaptive Neuro-fuzzy Inference Systems (ANFIS) to predict the output deflection basin parameters from the input variables. The data to train, test and validate the model were gathered through field trials. To achieve the above goal, several models based on ANN and ANFIS were trained by changing number of hidden layers, the neurons in the layer and number of membership functions. Prediction efficiency of the model is assessed based on its root mean square error and the coefficient of determination value.