A Machine Learning-Enabled Method for Predicting the Tribological Performance of Materials Considering...

The surface/interface properties of materials affect their tribological behavior significantly, necessitating quantitative research. Here, a hybrid machine learning model named CS-LSBoost (Cuckoo Searching-Least Square Boosting) is proposed to evaluate and predict the tribological properties, considering surface/interface properties as input. The incorporation of the CS algorithm in the proposed model accelerates the modeling procedure and improves accuracy. In contrast to the conventional LSBoost algorithm, the CS-LSBoost algorithm exhibits superior performance that reduces the MAPE (mean average percentage error) on the CV (Cross-Validation) set in the friction task from 15.41% to 12.10% and in the wear task from 14.97% to 10.09%. When predicting the coef. friction and wear rate, the validation MAPEs on the hold-out set were only 6.31% and 9.54%, respectively. The proposed prediction model, with accurate quantitative correlation and appropriate physical interpretability between the tribological performance and the surface and interface properties of materials, can provide guidance for material optimization.