Developing Deterioration Models for Wyoming Bridges

Deterioration models for the Wyoming Bridge Inventory were developed using both stochastic and deterministic models. The selection of explanatory variables is investigated and a new method using Least Absolute Shrinkage and Selection Operator (LASSO) regression to eliminate human bias in explanatory variable selection.The cross validation technique is used to determine the minimum number of explanatory variables. The relative significance of candidate variables is used to rank the explanatory variables in hierarchical order. The deterministic deterioration models are developed by using curve-fitting methods for the mean of bridge ages for each condition rating. In order to improve the accuracy in the model, bridges are split into the multiple subsets using first two explanatory variables for deck, superstructure, and substructure. Although the deterministic deterioration model is insufficient to predict condition ratings for a specific bridge, it is worthy to observe a general feature of how the functionality of bridges becomes worse over time. The stochastic models are developed to capture the uncertainty in the deterioration process using the Markov chain. The transition probability matrix is estimated using percentage prediction method,which counts the numbers corresponding to the element of transition probability matrix. The same subsets used in the deterministic deterioration models are considered. For each subset, zoning technique is used such that the bridge data is grouped for every 30 years to estimate transition probability matrix separately. The source codes are provided for the future update of bridge inventory and stochastic deterioration models. A computer program is used develop and plot deterioration models. A simple guideline is also included so that the user can access the source codes conveniently.