Prediction of surface roughness using machine learning approach for abrasive waterjet milling of alumina ceramic

The present work embodies the effect of abrasive waterjet milling process parameters on surface roughness in alumina ceramic material that is modelled with a machine learning approach. Experiments are carried out on the basis of response surface methodology (RSM) involving the Box–Behnken approach. The individual and interactive effects of the abrasive waterjet milling process parameters on surface roughness are studied through analysis of variance, and a quadratic regression model is developed. The combinations of abrasive waterjet milling input process parameters such as the pressure of 200 MPa, the step over of 0.2 mm, the abrasive flow rate of 0.42 kg/min and the traverse rate of 1000 mm/min have resulted in minimum surface roughness. In addition, the $$\varepsilon$$ -support vector regression model of machine learning is developed to predict the surface roughness. To enhance the support vector regression model, its hyperparameters are tuned using grid search with fivefold cross-validation. The tuned hyperparameters are found to have the cost function $$(C)$$ of 5, $$\varepsilon$$ -insensitive loss function of 0.0001, width of the radial basis function $$(\gamma )$$ of scale and radial basis kernel function. The support vector regression model (92.4%) has outperformed the quadratic regression model (70%) in the prediction of surface roughness.