Predicting Performance of Acidizing Corrosion Inhibitors through Machine Learning Techniques

Abstract Tests for Acid Corrosion Inhibitors (ACI’s) are time consuming, require significant resources and present hazards due to high acid concentrations, pressures, and temperatures. A machine learning model was developed to help drive efficiency and reduce the risk of traditional ACI testing. Intelligent models (based on regression and classification) were built to predict corrosion rate (CR) and pitting for ACI’s. These models, used as a pre-screening tool, improve efficiency by reducing the number of tests conducted and achieve a faster speed to solutions. The models were developed using Python and trained from a historical database with over 12,000 tests dating back to 2010. Historical tests were run using high pressure high temperature NACE cells, an autoclave, and ovens. The database contains commercial and experimental formulas. The models were trained on the commercialized portion, approximately 8,000 tests. Conditions such as metallurgy, temperature, ACI, and duration were inputs to the models to give a better picture of the tests to be conducted or eliminated. Twenty predicted results showed an average CR of 0.0190 lb/sq. Ft versus the actual average of 0.0221 lb/sq. Ft. A formulator and product selector are also being developed using the database and models.