| description abstract | By 2005, approximately 60,000 bridges throughout the US were identified as having unknown foundations. The Federal Highway Administration (FHWA) required Departments of Transportation (DOTs) across the US to evaluate the safety of these bridges, in particular against scour failure, and reclassify them in the National Bridge Inventory (NBI). A probabilistic methodology is developed in this study to predict the type, embedment depth, and dimensions of unknown bridge foundations and to rigorously quantify the uncertainty of the predictions. This methodology uses artificial neural networks (ANNs) to predict the expected values of bearing capacity (BC) using available information on bridge loading, soil strength, location, and year built, among other parameters. The unknown foundation characteristics are then evaluated using the Bayesian inference method and Markov chain Monte Carlo (MCMC) simulations, based on the predicted BC by the ANN models. The proposed method was validated based on a case study and proved successful in providing reasonable estimates on minimum foundation embedment depth for scour failure risk assessments. Transportation management authorities could adopt this method to reclassify bridges with unknown foundations and to implement risk based decision making approaches for bridge management. | |
