Document Type : Research Paper

Authors

Abstract

This paper uses neural network to predict corrosion rate. Corrosion can not modeled easily, because of wide range of causes either known or  unknown. In  mechanistic  approach,  physical,  chemical and electrochemical reactions and processes are considered to model and predict. But as stated before this models are not practically successful in prediction because of unknown parameters.
This paper uses genetic optimized neural network to predict corrosion rate. Among different neural networks, multi  layer neural network with  gradient  descent  learning algorithm has  been  selected. After developing the network, learning process has been done, using an oil refinery'S data. Then evaluation and test have been performed. After preparing  the  network  Garson's  algorithm and  sensitivity  analysis have been used for knowledge extraction.
According to results, neural network approach can predict corrosion rate  with  acceptable  correlation coefficient  (R)  and  mean  squared error (MSE). Sensitivity analysis depicts the strength of each oil parameter influence on corrosion rate. Among these results, salt and sulphur are the most affecting parameters in corrosion rate.

Keywords