Vol. 54 No. 03 (2005): Volume 54 Number 03, June 2005

Comparison of multilayer perceptron and adaptive neuro-fuzzy system on backcalculating the mechanical properties of flexible pavements

Ege University, Faculty of Civil Engineering, Istanbul, Turkey
Emine AĞAR
Istanbul Technical University, Faculty of Civil Engineering, Istanbul, Turkey
A. Hilmi LAV
Istanbul Technical University, Faculty of Civil Engineering, Istanbul, Turkey

Published 2005-02-01


  • Nondestructive testing

How to Cite

GÖKTEPE, A. Burak, Emine AĞAR, and A. Hilmi LAV. 2005. “Comparison of Multilayer Perceptron and Adaptive Neuro-Fuzzy System on Backcalculating the Mechanical Properties of Flexible Pavements”. ITU ARI Bulletin of Istanbul Technical University 54 (03):65-77. https://ari.itu.edu.tr/index.php/ituari/article/view/64.


Nondestructive testing (NDT) is the integral part of the performance evaluation of exible pavements. In all NDT methods, Falling Weight Deectometer (FWD) is probably the most popular technique. Basically, it measures time-domain de ections from numerous road sections emerging by the applied impulse load. In order to characterize the structural integrity of considered pavement system, it is required to make an inversion for the calculation of mechanical pavement properties using a backcalculation tool covering both a forward pavement response model and an optimization algorithm. On the other hand, backcalculation problem can also be solved by an adaptive system using a supervised learning algorithm. In this manner, multilayer perceptron (MLP) and adaptive neuro-fuzzy system (ANFIS) methodologies, popular universal functional approximating techniques of Artificial Intelligence (AI), are appropriate for pavement backcalculation problem. Therefore, two-phased (forward and backward) structure of traditional backcalculation approaches is reduced into one step with the help of the supervised learning mechanisms of MLP and ANFIS. In this study, these methodologies are both employed to backcalculate mechanical properties of exible pavements and compared in terms of modeling precision, uncertainty handling, computational expense, and data requirements. Results indicated that, both techniques are valid and have certain advantages over each other and should be preferred with respect to quantity and quality of the data at hand. In addition, AI-based supervised nonlinear mapping techniques not only exhibit precise backcalculation results, but also enable real-time pavement analyzing abilities.