Modelling of traffic loading on long span bridges

Existing long span load models have typically been developed using a number of conservative assumptions, and as such are more applicable to the design of new bridges rather than the assessment of existing structures. Excessive conservatism in such assumptions can lead to expensive and unnecessary interventions in existing bridges. Furthermore, existing load models do not always allow for correlations in traffic weights and vehicle positions on the bridge. This thesis proposes a method of simulating load effects on long span bridges termed ‘Long Span Scenario Modelling’ (LSSM). The ‘scenarios’ are blocks of vehicles extracted from a stream which contain the inherent correlations between vehicle weights and positions. The correlation in load intensity between successive scenarios is explicitly modelled. The scenarios can be used to simulate congested conditions for the required number of congestion events. A large Weigh-in-Motion (WIM) dataset from a site in the USA is used to demonstrate the process. Free-flowing WIM data is converted into a congested traffic stream using lane changing and gap distribution models. Recurring rush hour type congestion is simulated. The load intensities for 500, 1000 and 1500 m loaded lengths are determined for 1000-year return periods. An efficient computer algorithm to allow simulation of long span bridge load events for long return periods is also developed. Such long run simulations can avoid the uncertainties with extrapolation techniques where data recorded over relatively short periods of time is extrapolated to large return periods. To ensure a diversity of load events, the measured scenario library is extended through the generation of new scenarios. The LSSM is shown to better represent the long span load intensities when compared to measured traffic, particularly when the correlation between successive scenarios is accounted for. For the studied cases it is also shown that a Gumbel (linear) extrapolation from one year of WIM data overestimates the long run simulation value by approximately 12%. The developed algorithm therefore allows the long run simulations for the LSSM method to be carried out on a desktop computer and therefore greatly reduce the variability of results and limit potential issues regarding extrapolation techniques and choice of suitable statistical distributions.