Abstract

The US Federal Highway administration released a study that the cost of corrosion and its preventive strategies
for 1998 in the U.S. was approximately $276 billion approximately 3.2% of the US gross domestic product.
Corrosion is an important and costly phenomenon and many models were developed to predict corrosion
behavior. This paper presents an artificial neural networks (ANN) model to simulate the complex and nonlinear
atmospheric corrosion process from observed experimental data. The input parameters to the model consists of
temperature, time of wetness (TOW), sulfur dioxide concentration, chloride concentration, exposure time and
the model output is corrosion depth. Good performance of ANN model was achieved. The interactions between
the inputs were estimated by performing sensitivity analysis based on the developed model. The results showed
good agreement with experimental knowledge. SO2 and Cl- environments are more influential than other
elements