The Use of Autoregression to Generate Patterns of Extreme Loading on Long-Span Bridges

Abstract

For long-span bridges, there is little known about the gaps between vehicles in the congested traffic loading events that govern the characteristic maximum load effects. In many studies for the derivation of national standards, the gaps between trucks in Weigh-in-Motion (WIM) data, taken in free-flowing traffic, are ﾑcollapsedﾒ to minimum values to generate simulated congested events. Unfortunately, while this does generate extremes, the trend of increasing load effect with decreasing probability (on a Gumbel scale), is very different from reality. In this paper, an ARMA auto-regression algorithm is used to generate future predictions of load effect from measured past effects. This is shown to capture the trend in the extreme value extreme value data much more effectively. The concept is illustrated using traffic data collected using images taken from an Unmanned Aerial Vehicle (UAV) hovering over a highway. At the time of measurement, the highway was undergoing maintenance and there was regular congestion. Vehicle positions and spacing data taken from the UAV are complemented by truck weight data from a nearby WIM station. The ARMA model is shown to be an effective method of simulating extreme value congested event data.