Artificial Neural Network for Selecting Pavement Rehabilitation Projects

Abstract

The use of an artificial neural network simulator to develop and implement an automatic procedure for screening and recommending roadway sections for pavement preservation is described. This procedure is part of an automatic project recommendation procedure extension of the Arizona Department of Transportation's (ADOT's) pavement management system. The output of the recommendation procedure is a list of candidate projects for consideration in the 5-year pavement preservation program. The artificial neural network simulator was used to “learn” the knowledge from historical project selection. The neural network was trained with the pavement condition and characteristics and the sections selected for the ADOT's pavement preservation program for several years. The trained neural network predicted a correct output for 100 percent of the training facts and 76 percent of the testing examples. Further refinements of the artificial neural network architecture should result in better-performing networks. The artificial neural network analysis reduces the level of effort required to identify candidate sections for the pavement preservation program, reduces subjectivity, and minimizes the probability of missing sections that should be programmed.