Decision Tree Analysis for Developing Weigh-in-Motion Spectra in the Californian Pavement Management System (PaveM)

Abstract

Weigh-in-motion (WIM) devices record vehicle axle loads on highways. WIM data include axle loads, inter-axle spacings, vehicle classification, gross vehicle weights, vehicle length, and travel speed. Pavement design, management, and performance studies all require WIM data, directly as a spectrum or indirectly as standard axles. In 2016, 132 WIM devices were operating on California’s highway network, one of the densest and best-maintained clusters in the United States. The University of California Pavement Research Center (UCPRC) has been working with these data since 2002, and it completed the analysis of data collected from 1998 to 2003 in 2008. This research investigated similarities in axle load distributions at the WIM sites and grouped them to generate traffic inputs for pavement design. This paper is an extension of that work, including processing the 2004–2015 data from 80 WIM sites, identifying the axle load distribution for each site, updating the original groupings, and developing a new procedure to assign WIM spectra to other locations. For the past 11 years, axle loads and gross vehicle weight showed similar patterns, but average truck speeds decreased by 11%. A combination of clustering and cut-tree analysis was used to create a decision tree to classify the stations’ data into five WIM axle load spectra. The methodology has been used to compute traffic for the entire network used in the Caltrans pavement management system, PaveM. An application tool was also developed for generating truck traffic input files for the CalME and PavementME design tools.