ABSTRACT
Abstract Cracking is one of the important components of surface distress variables used in pavement performance modeling in Pavement Management Systems (PMS). It is also a major distress mode that causes premature failure in flexible pavements. Models for predicting cracking, especially change in cracking, have been developed using age of the pavement, structural number and cumulative standard axle loads, etc. A major characteristic of the models is that they are formulated and estimated using a statistical approach. This paper uses artificial neural networks (ANNs) in predicting the area of indexed cracks in flexible pavement. The index crack is a weighted measure of area of cracking and crack width. With reference to cracking prediction, the neural networks can achieve modeling more parsimoniously than the statistical approach, and it can also identify which type of variables can be used to predict cracking. The networks are trained using past pavement data (structural number, age, ES AL, environmental factors, etc). A suitable learned network has the ability to generalize when presented with inputs no* appearing in the training data, and is capable of offering real-time solutions. Keywords: Neural networks, cracking, pavement management, flexible pavements.
