Application of artificial neural networks (ANN) and gray relational analysis (GRA) to modeling and optimization of the material ratio curve parameters when turning hard steel

Abstract

The purpose of this study is to model the functional parameters of the surface texture (ISO 13565 standard) and to choose the optimal cutting parameters when turning hard steel (16MC5) with a hardness of 52 HRC. The artificial neural network (ANN) and the gray relational analysis (GRA) method are used to model and optimize the three parameters related to the bearing length rate curve (${R}_{pk}$, ${R}_{k}$, and ${R}_{vk}$). An experimental design of three factors (cutting speed ${V}_{C}$, feed rate $f$ and depth of cut $ap$), with five levels each, was selected according to the Taguchi L₂₅ technique. The white ceramic cutting tool was used. The models based on neural networks are compared with that obtained by the response surface methodology (RSM). The precision and the capacity of prediction of the two methods (ANN and RSM) have been investigated. The coefficient of determination of the three predictive models of ${R}_{pk}$, ${R}_{k}$, and ${R}_{vk}$ was found to be 99.99%, which shows the effectiveness of this technique. The GRA allowed the optimization of the cutting conditions for a minimum reduced peak height (${R}_{pk}$), minimum core roughness depth (${R}_{k}$), and maximum reduced valley depth (${R}_{vk}$). The combination of the optimal cutting parameters respecting the previous optimization conditions is ${V}_{C}$ = 96 m/min, $f$ = 0.106 mm/rev and $ap$ = 0.10 mm.