ABSTRACT
Corrosion inhibitors are useful to mitigate corrosion of metal/alloy components. However, traditional corrosion inhibitors are toxic and need to be replaced by greener alternatives. Efficient screening models are required to find molecules with desired properties from millions of molecules available in public domain. To make these models, database of experimental inhibition efficiency of molecules is essential. In this work, we have developed a computational framework to accelerate the discovery of new corrosion inhibitors. We have used machine learning based algorithms to predict corrosion inhibition efficiency of organic molecules for steel in hydrochloric acid by using structural information of the molecules along with experimental conditions. Our multitask learning based neural network architecture was able to outperform traditional machine learning algorithms such as random forest, lasso and ridge regression. We have also created the largest dataset for predictive modelling of corrosion inhibitors for steel. Besides, we have also used the model to screen molecules from the ZINC15 dataset and found potential inhibitors with high inhibition efficiency.
