Neural Networks and AASHO Road Test

Abstract

The American Association of State Highway Officials (AASHO) road test, conducted during the period of 1958 through 1960, was a factorial test of pavement durability that considered layer depths, axle load, and number of load applications as the primary variables. These data were processed using traditional statistical techniques. The AASHO formula is the resulting databased model of the road-test data. In the present paper, we reexamine the AASHO road-test data, using the Monte Carlo Hierarchical Adaptive Random Partitioning (MC-HARP) neural-network model developed by Banan and Hjelmstad (1995), and show that an MC-HARP model can represent the data far better than the AASHO formula can. We conclude that the MC-HARP neural network may be an appropriate tool for the development of databased models of pavement performance in the future.