Application of quantitative structure tribo-ability relationship model developed by random forest algorithm in polyimide-induced alignment superlubricity system

In this study, 38 organic compounds were used as lubricants in the GCr15 steel/polyimide (PI) friction system to systematically compare the influence mechanism of lubricant molecular structures on the friction performance of the system. Quantitative structure tribo-ability relationship (QSTR) models linking structural descriptors with coefficients of friction were established using random forest (RF) and extreme gradient boosting (XGBoost) algorithms, respectively. Based on comparisons of performance metrics such as fitting degree and generalization ability, the QSTR model developed by the RF algorithm was selected to quantitatively analyze the nonlinear influence mechanism of lubricant molecular structures on coefficients of friction. The study preliminarily reveals that the ability of lubricant molecules to form hydrogen bonds with PI macromolecules, the strength of van der Waals forces, and the influence of molecular geometric conformation on the effective interfacial interaction distance collectively constitute the key factors affecting the friction performance of the system.