Prediction of the surface finish of hardened bearing steels was estimated in
machining with ceramic uncoated cutting tools under various process parameters
using two statistical approaches. A second-order (quadratic) regression model
(MQR, multiple quantile regression) for the surface finish was developed and
then compared with the artificial neural network (ANN) method based on the
coefficient determination (R
2), root mean square error (RMSE), and percentage error (PE). The
experimental results exhibited that cutting speed was the dominant parameter,
but feed rate and depth of cut were insignificant in terms of the Pareto chart
and analysis of variance (ANOVA). The optimum surface finish in machining
bearing steel was achieved at 100 m/min speed, 0.1 mm/revolution (rev) feed
rate, and 0.6 mm depth of cut. In addition, the ANN model revealed a better
performance than that of MQR for predicting the surface finish when machining
the hardened bearing steels because R
2 was about 0.787 and 0.903 for MQR and ANN, respectively. Besides,
these were associated with RMSE of 0.302 and 0.1071 for MQR and ANN. Further, PE
estimated from randomly selected data were about 25.56% and 10.86% for MQR and
ANN, respectively. However, MQR presented the lowest error of 2.86%, but the
highest error of 40.3%, while ANN indicated the lowest error of 0.11%, but the
highest error of 37.0%, respectively.