Stacking ensemble learning framework for predicting tribological properties and optimal additive ratios of amide-based greases

This study employs a stacking ensemble learning framework to establish a regression model for predicting the tribological properties of amide-based lubricating grease and determining the optimal additive ratios. Melamine cyanuric acid (MCA) was selected as the thickener, and three extreme-pressure anti-wear additives were used to prepare the lubricating grease. The tribological performance was tested using an MFT-R4000 reciprocating friction and wear machine. Based on the tribological experimental data, the synthetic minority oversampling technique (SMOTE) was utilized for data augmentation, and a stacking ensemble algorithm with Bayesian optimization of hyperparameters was used to construct a predictive model for tribological performance. Subsequently, within this model framework, single and multi-objective optimization models were developed, and the fruit fly algorithm was employed to find the optimal additive combination ratios, which were experimentally validated. The results demonstrated that the learning framework based on the stacking ensemble model could effectively predict the tribological properties of amide-based lubricating grease in small sample datasets, with the R² for the average friction coefficient prediction reaching 0.9939 and for the wear scar width prediction reaching 0.9535. In the experimental validation of the optimal additive ratios, the relative error of the friction coefficient ratio scheme was 0.51%, and the relative error of the wear scar width was 1.10%. This finding suggests that the learning framework provides a novel approach for predicting the performance of amide-based lubricating grease and studying additive combinations.