Modeling Crack Deterioration of Flexible Pavements

Abstract

Pavement cracking and rutting are two of the most critical distress types manifested on flexible pavements, and they often govern the overall pavement condition. Hence, many models have been developed for forecasting the deterioration of the crack condition accurately, with the traditionally preferred technique being the use of a regression relationship developed from laboratory or field statistical data, or both. However, it becomes tedious for regression techniques to predict crack performance accurately and robustly in the presence of the multitude of tributary factors, material nonlinearity, and uncertainty involved in the cracking process. With the advancement of modeling techniques, two innovative breeds of models, neural networks and recurrent Markov chains, have drawn increasing attention from researchers for their use in modeling complex phenomena such as pavement cracking. This paper compares the ability of neural networks and recurrent Markov chains to model crack performance, using the Florida Department of Transportation's pavement condition data. Comparable forecasting accuracies have been indicated for single-year forecasts using both models. When the models were applied for multiple-year prediction of crack performance, the recurrent Markov chain tends to produce more consistent forecasts as compared to the neural network, which exhibits a tendency toward overpredicting crack deterioration. Finally, further discussion gives an insight into the structure of each model.