UVM Theses and Dissertations
Format:
Print
Author:
Tirk, Nathan
Dept./Program:
Civil Engineering
Year:
2014
Degree:
M.S.
Abstract:
Probabilistic analysis of civil structures is primarily concerned with the quantification of uncertainty that results from random variations in demand and capacity throughout the service life of structures. In the case of bridges, one important source of uncertainty comes from demands induced by vehicles, i.e. vehicular live loads. Current bridge design guidelines (AASHTO LRFD) are operationally deterministic and define the vehicular live load model (HL-93) that is intended to provide a level of demand consistent with a predetermined level of reliability (reliability index of 3.5 or larger in 75 years) for most bridges.
The objective of this thesis is to identify suitable probabilistic models that accurately represent vehicle live load demands in bridge decks from weigh-in-motion (WIM) station data. This thesis compares two frameworks for uncertainty quantification, i.e. extreme value and full population models. We examine which of these frameworks is more consistent with the current target reliability required for current bridge designs. The WIM data corresponds to 12 different stations across the state of Vermont during a period of 12 years (2001-2012). It is shown that the statistical description of the vehicular live load demands in bridges does not follow a single standard distribution, but instead can be adequately described by a mixture of lognormal distributions.
A direct consequence of this fact is that the corresponding daily extreme values do not fit any of the traditional extreme value distributions. Bayesian model averaging methods are employed to identify both the distribution and parameters that best describe these extreme values. The identified probability distributions are used to perform reliability calculations for simply supported bridge decks of varying length. Appreciable variation in the estimated probability of failure is observed within and across WIM stations, illustrating that the random variation in vehicular live loads is not a stationary random process in time or space.
The objective of this thesis is to identify suitable probabilistic models that accurately represent vehicle live load demands in bridge decks from weigh-in-motion (WIM) station data. This thesis compares two frameworks for uncertainty quantification, i.e. extreme value and full population models. We examine which of these frameworks is more consistent with the current target reliability required for current bridge designs. The WIM data corresponds to 12 different stations across the state of Vermont during a period of 12 years (2001-2012). It is shown that the statistical description of the vehicular live load demands in bridges does not follow a single standard distribution, but instead can be adequately described by a mixture of lognormal distributions.
A direct consequence of this fact is that the corresponding daily extreme values do not fit any of the traditional extreme value distributions. Bayesian model averaging methods are employed to identify both the distribution and parameters that best describe these extreme values. The identified probability distributions are used to perform reliability calculations for simply supported bridge decks of varying length. Appreciable variation in the estimated probability of failure is observed within and across WIM stations, illustrating that the random variation in vehicular live loads is not a stationary random process in time or space.