Comparison of Neural Network Method Versus National Bridge Inventory Translator in Predicting Bridge Condition Ratings

The National Bridge Inventory (NBI) Database is a repository of data on highway bridges in the United States. It serves as the most valuable source of information for bridge engineers for evaluation and analysis of conditions of the nation’s bridges and also as a source for policy makers for prioritizing bridge maintenance activities. Some items of great interest to bridge engineers in the NBI are bridge deck, bridge superstructure and bridge substructure condition ratings. These condition ratings are assessed every two years in the field by bridge inspectors. There is however a growing interest in the collection of element level data which is a more detailed data on the conditions of the components of a bridge deck, bridge superstructure and bridge substructure. Some researchers have sought to develop models that will predict the bridge deck, bridge superstructure and the bridge substructure condition ratings from their corresponding element level condition data. One such model is the NBI Translator. Some State bridge engineers that currently use the NBI Translator have raised some concerns regarding its effectiveness. Therefore to improve upon the NBI Translator, an alternative method of predicting the bridge deck, bridge superstructure and bridge substructure condition ratings from element level condition data using Artificial Neural Networks (ANN) was investigated and the results discussed in this paper. The results indicate that the ANN is a better predictor than the NBI Translator provided the data used in training the ANN is from the same State as the data used in the predictions.