ABSTRACT
Over the past two decades, extremely heavy vehicles, or superloads, have been increasingly utilized to transport heavy loads, such as prestressed concrete girders, automotive presses, transformers, wind turbine components, or other heavy loads. Since these superload have a significant effect on the infrastructure system in comparison to the regularly permitted vehicle, they should be subject to special consideration in the permitting and operation process. Despite the great research effort that has been made to improve the superload permitting process, few studies have been performed on characterization and prediction of superload. The superload has its own distinct characteristics that differ from other vehicle loads. Thus, there is a need to better understand the characteristics of superload and to develop an analytical procedure for future use. In this paper, the major focus is to develop an analytical procedure for the characterization and prediction of superload using advanced gradient boosting machine (GBM) learning algorithms. Weigh-in-Motion (WIM) data collected from 31 sites in Florida over 10 years were used as the database for this study. The raw data has been processed with a newly developed procedure and the superload data was extracted. A comprehensive analytical technique was developed using GBM with regression, classification trees, and time series modeling. This analytical procedure was specifically altered to accommodate the unique features of superload data. By applying the new analytical procedure, the characterization of superload was performed for Florida WIM sites and the prediction of various key parameters, such as maximum axle weight and gross vehicle weight, was conducted.
