Forecasting Overall Pavement Condition with Neural Networks: Application on Florida Highway Network

Abstract

Timely identification of undesirable crack, ride, and rut conditions is a critical issue in pavement management systems at the network level. The overall pavement surface condition is determined by these individual pavement surface conditions. A research project was carried out to implement an overall methodology for pavement condition prediction that uses artificial neural networks (ANNs). In the research, three ANN models were developed to predict the three key indices—crack rating, ride rating, and rut rating—used by the Florida Department of Transportation (FDOT) for pavement evaluation. The ANN models for each index were trained and tested by using the FDOT pavement condition database. In addition to the three key indices, FDOT uses a composite index called pavement condition rating (PCR), which is the minimum of the three key indices, to summarize overall pavement surface condition for pavement management. PCR is forecast with a combination of the three ANN models. Results of the research suggest that the ANN models are more accurate than the traditional regression models. These ANN models can be expected to have a significant effect on FDOT's pavement management system.