Modeling the AISI 1035 Steel Turning Process Using RSM, DNN-GA and DT Models

الملخص

The current study deals with the development of the predictive models of surface roughness parameters, such as arithmetic mean deviation of the roughness profile (Ra) and total height of the roughness profile (Rt) in the process of turning of AISI 1035 medium carbon steel. Three modelling techniques, which have been comparatively studied, include the Response Surface Methodology (RSM), the Deep Neural Network in combination with Genetic Algorithm optimisation method (DNN-GA) and the Decision Tree model (DT). For the purpose of the present experimental work, 17 run design of experiments, which considered variations of three factors, i.e., cutting speed (Vc) within 170 – 345 m/min; feed rate (f) within 0.08 – 0.20 mm/rev; and depth of cut (ap) within 0.3 – 1.2 mm was carried out. Quadratic RSM model with 7 degrees of freedom showed excellent prediction ability, giving R² equal to 0.988 for Ra and 0.996 for Rt along with F-values of 65.40 and 192.86, correspondingly. Variance analysis demonstrated that f was the major factor influencing Ra (with 77.86% contribution) and Vc had the largest effect on Rt (71.75%). DT produced similar results in terms of accuracy, reaching an R² of 0.987 with a tree depth of 3 and producing clear decision rules that can be implemented on the shop floor. Interactions between parameters were presented graphically using response surfaces in 3D and radar charts.