Collaborative feature screen with large language model and machine learning model to enhance corrosion inhibitor prediction

Corrosion affects all areas of the national economy, from industrial and agricultural production to national defense technology. It seriously threatens the safety of equipment in service, causes huge economic losses, and poses great risks to human life and health. Metal corrosion inhibitors can change the surface state of metals, increase the activation energy barrier of reactions, affect surface electrochemical behavior, and slow down the corrosion rate. These inhibitors have advantages such as low dosage, low cost, and high efficiency, making them one of the most widely used methods for corrosion control. However, there are many types of inhibitors, their mechanisms are complex, and they are closely related to environmental factors. Traditional corrosion research methods, such as weight loss testing and electrochemical testing, usually require a lot of manpower, resources, time, and cost, which greatly hinders the design and application of high-performance inhibitors. There is an urgent need for a more efficient approach to advance inhibitor research. In recent years, the development of materials genome engineering has driven corrosion research from trial-and-error methods toward digital and intelligent approaches. Machine learning can be used to analyze existing data to predict a vast unknown space and explore potential relationships between material composition/structure and performance. This study used a large language model (LLM) assisted feature screening method. Based on data and knowledge about corrosion inhibitors reported in the literature, 13 descriptors most related to corrosion inhibition performance in a saturated carbon dioxide environment were selected from thousands of possible descriptors. These descriptors involve molecular physicochemical properties, molecular structural properties, and environmental parameters. After screening, the mean square error of model predictions dropped from 121 to 11. Follow-up corrosion experiments confirmed the prediction accuracy and generalization ability of the model. The feature screening process and machine learning model developed in this study significantly improved the efficiency of developing high-performance corrosion inhibitors for the target corrosion environment.