Theoretical background corrosion

Neural nets are generally used where it is difficult to develop an analytical model such as in prediction or pattern recognition problems. Neural net models for corrosion prediction are an extension of empirical models. They too are not based on any theoretical background, with constants used in them representing best-fit parameters based on their training data set [1]. 

Several deterministic corrosion models have been developed, some purely theoretical while others incorporate empirical results. For a review, see for example Nyborg (2002) or Nesic (2007). Many researchers have studied the mechanisms of corrosion on carbon steel in hydrocarbon environments. The purpose of the experiments has mainly been to study the effects of different influencing factors on the corrosion rate. With a theoretical background and mathematics based on physical understanding of the corrosion principles, these models often perform well under controlled experiments. However, their vahdation is limited outside the laboratory. This limited validation for field applications is mainly due to two reasons ... 

---