The Study on Establishing the Pavement Performance Prediction by Artificial Neural Networks

Pavement managment system

Abstract

Measuring the condition of current pavement is accomplished by collecting field distress data and synthesizing data to identify appropriate alternatives for rehabilitation or reconstruction. Many agencies have pavement management systems (PMS) to assist with data collection, evaluation, and decision-making during this process. The present serviceability index (PSI) is a common tool for quantifying information concerning the serviceability of the pavement. A primary factor used in establishing the PSI is the roughness of the surface profile. The PSI can also include standard distress criteria such as rutting, fatigue cracking, and thermal cracking. However, the actual causes and conditions of pavement distress are very complex. The statistical modeling can only consider no more than a few of the parameters, in a simplified manner, and in some cases various transformations of the original data. Because of the statistical nature of models this does not mean that cracking and rutting are not important, since they will react on the roughness of the surface profile. The artificial neural networks (ANNs) offer a number of advantages over the traditional statistical methods, caused by their generalization, massive parallelism and ability to offer real time solutions. In this paper, real pavement condition and the subjective present serviceability rating (PSR) in Taiwan are used to develop a generic intelligent pavement performance prediction using ANNs. In contrast to statistical analysis, it is concluded that the good predictive results can be obtained from the pavement performance model established by neural network.