ABSTRACT
Highway administrative department sets the charging standard according to vehicle type and its loading capacity, of which the vehicles are classified as two groups, i.e., passenger car and truck, and nine sub-groups. Wang divided vehicles into five categories according to the results of field vehicle investigation. Therefore, a weighted stack of five to nine probability models is needed to fit the vehicle load effects while all vehicles data is considered, that is, about twenty parameters are needed to be determined, which means computation intense and many uncertainties are generated. According to the EV theory, EV estimation of vehicle load effects only relate to those heavy vehicles. Therefore, the load effect distribution of vehicles of three heaviest types is studied to improve the accuracy of estimated EV and reduce computation amount in this paper. To conform the probabilistic model of strain peaks caused by heavy vehicles, the strain data of 324 days of Taiping Lake Bridge located in Anhui Province of China is taken as the real example. The strain time history curve caused by vehicle loads of 24 hours of 27th December, 2015 of Taiping Lake Bridge is shown in Figure 1, in which each ‘jump’ indicates the strain time history due to a vehicle
1 INTRODUCTION
The distribution of the vehicle loads effect on bridge, which shows multiple peaks, is usually complex and unknown (SUN, 2012). A single probability model or a weighted form of multiple probability models is always used to fit the whole measured data directly before, which is easy to bring errors in high quantiles. These errors would be accumulated and exaggerated during the process of Extreme Value (EV) estimation, which greatly reduce the accuracy of estimated EV. EV theory (Shi, 2006) shows that EV estimation only relate to the right tail of the base distribution. GPD (Generalized Pareto Distribution), as a fitting model for right tail distribution, which is widely used in EV estimation of flood height, wind speed, temperature and so on, has been well applied in the EV estimation of vehicle weight recently (Li, 2012). EV Estimation based on a GPD is closely related to its threshold, therefore, how to choose an appropriate threshold for a given set of data has become important. If the threshold is too large, few data will lead to the increase of the variability of estimated parameters of the GPD. On the contrary, if the threshold is too small, too many data will lead to the increase of the deviation of estimated parameters of the GPD.
