Application of Machine Learning Tools to Enhance Corrosion Inhibitor Mixture Optimization Experimentation

To determine the optimum performance of a corrosion inhibitor blend, a mixture experiment was conducted utilizing three benzalkonium chloride surfactants with varying chain length (C₁₂, C₁₄ and C₁₆). The ternary experimental design consisted of ten mixture compositions. For each composition, linear polarization resistance (LPR) measurements were performed to assess the corrosion rate, coverage and corrosion inhibitor performance. Multiple replicates were carried out for each mixture to ensure reproducibility and to identify potential outliers.

Having established the corrosion inhibitor performance for each of the surfactant mixtures in the ternary design of experiment, a series of response surfaces were generated using three different methods. First, the traditional cubic surface was implemented with interaction parameters for each of the component combinations. Second, a response surface based on gaussian radial basis functions and leave one out cross-validation (LOOCV) was generated, and third, a predictive neural network was applied to the data.

All three response surface models provided estimates for the optimal surfactant mixture that maximizes corrosion inhibition efficiency. The predictive capabilities of optima generated by each model were compared. Finally, the corrosion inhibitor performance for each predicted optimum composition were compared and a composition identified for experimentally validation