Determination of Unknown Foundation of Bridges for Scour Evaluation Using Artificial Neural Networks

There are approximately sixty thousand bridges throughout the US identified with unknown foundations, which results in serious problems for evaluating the bridges' current state of risk, particularly due to scour. This paper presents results of a soft-computing model to determine the depth of unknown bridge foundations based on the use of Artificial Neural Networks (ANN) trained on data collected from foundation design and inspection records. This approach allows for correlating between the depth of foundation and a series of parameters associated with the bearing capacity and foundation design of bridges. A case study is presented based on a pilot project considering existing bridges from the TX-DOT Bryan District. Some of the conditioning parameters considered for the ANN training includes structure, material type of bridges, soil properties, type and amount of loading, among others. A supervised learning algorithm is applied to train multilayer neural network models using information from bridges where full records were available.