Modelling the Deflection of Flexible Pavement using Artificial Intelligence Techniques

Abstract

A flexible pavement is a structural system consisting of several layers made of different materials, with stiffer layer placed at top and weaker ones at the bottom. The major function of flexible pavement is to provide a better riding quality and to distribute the traffic load uniformly, in order to protect the subgrade from excessive stresses. The riding quality and safety of the pavement are affected due to various types of distress acting over the surface during the service life of pavement. The pavements deteriorate due to combined action of traffic loads, environmental factors like climate, construction quality, material and time. To predict the rate of deterioration, various pavement performance models are evaluated which are helpful in determining the need for rehabilitation and reconstruction of the damaged pavements.
The major characteristic governing the road performance study is the pavement deflection, which is often used to evaluate a pavement’s structural condition non- destructively. These deflection measurements can be made either by static equipment or by using impact load devices. The most widely used method of determining the pavement deflection is the Benkelman Beam Deflection (BBD) test which measures pavement responses to the static load applied by a standard truck. The rebound deflection is measured using the BBD test, which indicates the elastic response of the pavement. This deflection is corrected to various grounds to obtain the characteristic deflection which is useful in designing the overlay for the flexible pavement. Though, the use of BBD is widely accepted because of the low cost but it has various drawbacks. The use of this method for evaluating pavement performance is slow, time consuming and labor intensive.
For this reason, a prediction modelling is done to estimate the characteristic deflection at the particular pavement section without conducting the Benkelman beam test. Hence, data driven modeling is done for deflection at the stretch of Durg bypass – Chhattisgarh / Maharashtra border of NH - 06 under NHDP phase IIIA (chainage 322.000km to 480.000km) using heuristic approaches for predicting the characteristic deflection using various input variables measured from the same road section. The study presents two branches of artificial intelligence (AI) techniques, namely linear genetic programming (GP) variant, multi expression programming (MEP) and multivariate adaptive regression splines (MARS) for the evaluation of deflection measured on the surface of flexible pavement using Benkelman beam. The experimental data validation is done for the model predicted by AI techniques to compute the error and propose a prediction model for characteristic deflection of flexible pavement. The input variables considered are the moisture content, plasticity index of subgrade soil and pavement surface temperature. The predicted deflection values are compared with the observed values from field study for both MEP and MARS to get the best fit model with least error and high correlation coefficient by comparing various statistical parameters and efficiency coefficients.