Development of experimental error-Driven model for prediction of corrosion rates of amines based on their chemical structures

This work investigated the relationships between amine corrosion rates and their chemical structural properties for application in the development of a Gaussian Process Regression (GPR) model for chemical structure-based prediction of corrosion rate of any amine. The GPR model accounted for experimental errors, which widened its scope to accurately predict the true corrosion rates, being restricted only to error associated with the trained model. The Average Absolute Deviation (AAD) between experimental corrosion rates and model predicted rates was 4.26 % for the test data, and 5.32 % for two test data unknown to the model. This showed that the model is generalizable and its predictions are accurate. This work also developed a user-friendly Graphical-User Interface, which allows a user to define any amine's structure to provide needed information to calculate its surface tension and steric effects for use as input variables to the model in predicting the corrosion rate of the amine.