Following the construction of a dataset of cross-category corrosion inhibitors at different concentrations based on 1241 data from 184 research papers, a performance prediction model incorporating 2D–3D molecular graph representation and corrosion inhibitor concentration information was established. This model was shown to effectively predict the inhibition efficiency (IE) of different categories of corrosion inhibitors for carbon steel in 1 mol/L HCl solution. The model was also able to predict IEs of corrosion inhibitors at different concentrations. The results demonstrated that 3D features of corrosion inhibitors, especially those of large molecules, had a significant impact on the prediction precision of IEs.

Current experimental verification, computational modeling, and machine learning methods for predicting corrosion inhibition efficiency (IE) are limited to specific inhibitor categories with high cost and poor generalization. In this study, a cross-category corrosion inhibitor dataset is constructed and a three-level direct message passing neural network (3 L–DMPNN) model using molecular structure information that integrates atomic-level, chemical bond-level, and molecular-level features to predict the IEs of compounds in a specific environment is established. This work demonstrates that the 3 L–DMPNN model can predict IEs of cross-category corrosion inhibitors from other independent literature and experimental dataset effectively and quickly.