Structure-activity relationship study of anti-wear additives in rapeseed oil based on machine learning and logistic regression.

Abstract

Anti-wear performance is a crucial quality of lubricants, and it is important to conduct research into the structure-activity relationship of anti-wear additives in bio-based lubricants. These lubricants are eco-friendly and energy-efficient. A literature review resulted in the construction of a dataset comprising 779 anti-wear properties of 79 anti-wear additives in rapeseed oil, at various loadings and additive levels. The anti-wear additives were classified into six groups, including phosphoric acid, formate esters, borate esters, thiazoles, triazine derivatives, and thiophene. Logistic regression analysis revealed that the quantity and kind of anti-wear agents had significant effects on the anti-wear properties of rapeseed oil, with phosphoric acid being the most effective and thiophene being the least effective. To identify the specific structural data that affect the anti-wear capabilities of additives in bio-based lubricants of rapeseed oil, a random forest classification model was developed. The results showed a 0.964 accuracy (ACC) and a 0.931 Matthews Correlation Coefficient (MCC) on the test set. The ranking of importance and characterization of MACCS descriptors in the model confirms that anti-wear additives with chemical structures containing P, O, N, S and heterocyclic groups, along with more than two methyl groups, improve the anti-wear performance of rapeseed oil. The application of data analysis and machine learning to investigate the classifications and structural characteristics of anti-wear additives in rapeseed oil provides data references and guiding principles for designing anti-wear additives in bio-based lubricants.