Track: Machine Learning

Abstract

Friction Drilling is a Chip less hole-making process in which a conical tool is used to form the hole on the sheet metal by frictional heating. Metal flows plastically in both upward and downward directions. The extruded metal on top is converted into the boss with the help of the shoulder of the tool while downward extruded material is well controlled to form the bush. A formed bush can be threaded and used to screw other connecting parts. The process is suitable for sheet metal having a thickness of less than 5 mm and finding applications in Automobiles, Furniture, etc. Thrust force and Torque are two Important parameters which affect the frictional heating during hole formation and thereafter tool wear and hole quality. In this work, experimentation on low carbon steel, AISI 1018 has been carried out at the Vertical Machining Center. For real-time Data Acquisition of Thrust Force and Torque, Kistler Dynamometer equipped with DynoWare software is used. The Thrust force and Torque prediction model for friction drilling of AISI 1018 steel has been developed using Random Forest (RF) and Regularization Methods (RM). Ridge and Lasso Regression which are important tools in RM have been utilized to model the measured data for Thrust force and Torque. GridSearch(GS) technique has been used for tuning hyperparameters. The Random Forest (RF) gives a reliable prediction of thrust force with R2= 0.97, Mean Absolute Error (MAE) = 71.91 N, Root Mean Square Error (RMSE) = 96 N on training data and R2 = 0.92, MAE =140 N, RMSE = 96 N on test data. Similarly, Torque Prediction Random Forest (RF) gives better predictions than RM on Test Data with R2 = 0.93, Mean Absolute Error (MAE) = 0.22 Nm, and RMSE = 0.29 Nm but on Training data Performance is almost similar both the methods. Comparisons among RF, Ridge, and Lasso regression show that RF is an effective technique to ensure high predictive accuracy of Thrust Force and Torque.