Assessing the feasibility of using a data-driven corrosion rate model for optimizing dosages of corrosion inhibitors

Abstract

Optimizing dosages of corrosion inhibitors requires experimental data gathered from time-consuming methods. The current study examines the feasibility of optimizing inhibitor dosages using a model trained for predicting corrosion rates more easily measured using linear polarization resistance in a full-scale cooling water system. A comprehensive study on variable selection showed that linearly correlated variables are necessary to predict corrosion trends. The Sobol sensitivity of inhibitors is trivialized by variables linearly correlated to the corrosion rate. The study highlights the importance of achieving high model prediction accuracy and high Sobol sensitivity of inhibitors to the corrosion rate, for using the model for inhibitor dosage optimization.