Prediction of Surface Roughness in Hybrid Magnetorheological Finishing of Silicon Using Machine Learning
Authors :- Srivastava M.; Singh G.; Verma K.; Pandey P.M.; Rana P.S.; Gupta M.K.; Khanna N.
Publication :- Silicon, Springer 2024
The machining learning-based predictive model of double disc chemo-magnetorheological finishing process of silicon was proposed in the present manuscript. Six different methods such as CatBoost Regressor, XGBoost, Random Forest Regressor, Gradient Boosting Regressor, Linear Regression, and AdaBoost Regressor were used to predict the surface roughness. The models were trained by the experimental data of surface roughness of silicon wafer polished at combination of different set of parameters. The gradient boosting algorithm was introduced to train the dataset of the models for the surface roughness of the silicon wafer. The robustness of the models was verified with K-fold cross method. The models were verified with the condition monitoring data collected by experimental results. The models were also developed for ultrasonic assistance during the double disc chemo-magnetorheological finishing process. The CatBoost approach outperformed the other models. The accuracy of the CatBoost model was 99.92% and 98.35% for the experimental data without and with ultrasonic vibration assistance. The optimised values from the predicted model were 4.21 nm and 3.4 nm without and with the assistance of vibration for the chemo-magnetorheological finishing process and have good agreement with the experimental results. The acquired experimental findings indicate that the suggested data-driven predictive modelling methodology can accurately forecast the surface roughness of polished silicon wafers.