Machine Learning Algorithms for Rutting Modelling of Bituminous Pavements in West Midlands

This paper applies backpropagation ANN (BP-ANN) with a GDR learning algorithm to model the rutting progression in bituminous pavements. The BP-ANN can predict the rut depth with greater accuracy and define the proportion of errors contributed by explanatory variables to rut depth. The BP-ANN model predicts the maximum rut depth as a function of annual average daily traffic (AADT), speed, longitudinal variance profile (LPV), skid resistance, design traffic, temperature, road texture and equivalent single axle load (ESAL) for 638 road segments in West Midlands. Data on the selected road segments were collected from the HAPMS (Highways England Pavement Management System) database. The BP-ANN model determined that LPV at 3m is the most important variable of rut depth followed by skid resistance, AADT and ESAL. This study also finds that high temperature coupled with traffic loads accelerates the rutting progression. The accurate prediction ability of rutting progression by BP-ANN model supports the systematic maintenance of bituminous pavements increasing the drivers’ comfort and safety.