CorroZoom Webinar (on SCC)

Active Learning Approach Towards Discovery of New Efficient Corrosion Inhibitors

Mikhail Zheludkevich, Institute of Surface Science, Helmholtz Zentrum Hereon, Geesthacht, Germany

Abstract

Organic corrosion inhibitors, whether added to protective coatings or directly to corrosive environments, play a crucial role in various active corrosion protection strategies. However, the quest for effective corrosion inhibitors within the vast chemical space remains a difficult challenge. Over the past decades, countless research papers have documented the inhibitory effects of individual compounds on various metals across diverse corrosive conditions, creating an infinite narrative. Fortunately, recent advancements in machine learning (ML) techniques offer promising avenues for narrowing the search and identifying potential candidates more efficiently. This work highlights the potential of computer-assisted methods in rapidly screening large numbers of organic compounds as potential corrosion inhibitors for magnesium and aluminum alloys. Our approach involves developing quantitative structure-property relationship (QSPR) models using ML algorithms, specifically support vector regression and kernel ridge regression. These models learn from existing data and generalize to predict the behavior of new compounds. To assess their robustness, we conducted experimental blind testing. The ML models leverage molecular descriptors derived from geometry and density functional theory calculations of organic compounds. Notably, two systematic approaches for sparse feature selection, identifying molecular descriptors most relevant to the corrosion inhibition efficiency of chemical compounds were proposed. This framework outperforms predictions based on randomly selected descriptors. To further enhance prediction quality, an active learning approach has been implemented. Experimental results from newly predicted modulators were incorporated into extended training data sets, iteratively improving model accuracy over time. In summary, ML-driven approaches hold great promise for accelerating the discovery of corrosion inhibitors. By harnessing computational tools, researchers can efficiently explore the vast chemical space and make informed decisions on the selection of corrosion inhibitors for specific applications.

You can register (free of charge) here: https://osu.zoom.us/webinar/register/WN_AH4DD_ZaSw2RpA-xtUQR_g. After registering, you will receive a confirmation e-mail containing information about how to join the webinar.