ROUGHNESS PREDICTION MODEL BASED ON THE ARTIFICIAL NEURAL NETWORK APPROACH Francesca La Torre*, Lorenzo Domenichini* and Michael I. Darter** University of Florence* Department of Civil Engineering Via S. Marta 3, I-50139 - Firenze - ITALY ERES Consultants, Inc.** 505 West University Avenue Champaign, IL 61820-3915 - USA Abstract Prediction of the future roughness of a flexible pavement section is important in programming rehabilitation needs. This study utilizes the Artificial Neural Network (ANN) approach to predict the future International Roughness Index (IRI) year by year up to a maximum pavement age of 20 years. The ANN program requires key information about a pavement section including the current IRI, pavement design, climate, traffic and other variables to predict yearly IRI values into the future. The program was developed initially using synthetic data to reproduce a well defined function and then utilized data from the Long-Term Pavement Performance (LTPP) program database to calibrate the ANN. Accuracy of the predictions was tested using LTPP sections not used in the development and was found to be reasonable for programming purposes. The ANN was developed as a Microsoft Windows95 based software tool and can be used in routine pavement management programming. INTRODUCTION Artificial Neural Networks (ANNs) have been proven to be extremely powerful tools for modeling phenomena in which many variables are involved (Simpson et al., 1995; La Torre and Domenichini, 1997). Several studies have been performed to identify the variables affecting roughness development in time. General groups include traffic, material composition, subgrade, pavement structure and climatic conditions. The analysis of the available roughness time series data with regression techniques accounting for all these variables is usually extremely difficult and requires an a priori definition of the form of the regression equation. This study has been conducted to define an ANN capable of predicting roughness progression over time based on the analysis of part of the LTPP database. FEED FORWARD ARTIFICIAL NEURAL NETWORKS WITH BACK- PROPAGATION In this study a Multi-layer Feed Forward Artificial Neural Network has been implemented. In this kind of ANNs there are connections only between nodes of 4th International Conference on Managing Pavements (1998) TRB Committee AFD10 on Pavement Management Systems is providing the information contained herein for use by individual practitioners in state and local transportation agencies, researchers in academic institutions, and other members of the transportation research community. The information in this paper was taken directly from the submission of the author(s).