Machine learning approach for the prediction of mixed lubrication parameters for different surface topographies of non-conformal rough contacts

Purpose: This study aims to use a machine learning (ML) model for the prediction of traction coefficient and asperity load ratio for different surface topographies of non-conformal rough contacts. Design/methodology/approach: The input data set for the ML model is generated using a mixed-lubrication model. Surface topography parameters (skewness, kurtosis and pattern ratio), rolling speed and hardness are used as input features in the multi-layer perceptron (MLP) model. The hyperparameter tuning and fivefold cross-validation are also performed to minimize the overfitting. Findings: From the results, it is shown that the MLP model shows excellent accuracy (R² > 90%) on the test data set for making the prediction of mixed lubrication parameters. It is also observed that engineered rough surfaces with high negative skewness, low kurtosis and isotropic surface patterns exhibit a significant low traction coefficient. It is also concluded that the MLP model gives better accuracy in comparison to the random forest regression model based on the training and testing data sets. Originality/value: Mixed lubrication parameters are predicted by developing a regression-based MLP model. The machine learning model is trained using several topography parameters, which are vital in the mixed-EHL regime because of the lack of regression-fit expressions in previous works. The accuracy of MLP with random forest models is also compared.