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

Area of Science:

• Materials Science
• Tribology
• Machine Learning

Background:

• Material surface/interface properties significantly influence tribological behavior.
• Quantitative research is needed to understand these influences.
• Predicting tribological properties requires accurate models.

Purpose of the Study:

• To propose a hybrid machine learning model, CS-LSBoost, for evaluating and predicting tribological properties.
• To utilize surface/interface properties as input for the prediction model.
• To improve the accuracy and efficiency of tribological property prediction.

Main Methods:

• Developed a hybrid machine learning model: Cuckoo Searching-Least Square Boosting (CS-LSBoost).
• Incorporated the Cuckoo Searching (CS) algorithm to accelerate modeling and enhance accuracy.
• Used surface/interface properties as input features for the model.

Main Results:

• CS-LSBoost demonstrated superior performance compared to the conventional LSBoost algorithm.
• Reduced mean average percentage error (MAPE) for friction prediction from 15.41% to 12.10% (Cross-Validation).
• Reduced MAPE for wear prediction from 14.97% to 10.09% (Cross-Validation).
• Achieved low validation MAPEs of 6.31% for coefficient of friction and 9.54% for wear rate on a hold-out set.

Conclusions:

• The CS-LSBoost model provides accurate quantitative correlation between tribological performance and material surface/interface properties.
• The model offers physical interpretability, guiding material optimization.
• This approach advances the prediction of tribological behavior based on material characteristics.