Prediction of surface roughness when maching mild steel using statistical methods

Abstract

ABSTRACT Surface roughness as a machinability index is very important parameter for getting a quality product in manufacturing industry because of affecting the fatique and fracture behaviour of any type of materials. Aim of the study is to generate statistical models in surface finish when machining AISI 1040 hardened mild steels with coated carbides including two different geometries of triangle and square under different conditions. The machining test was carried out at different cutting parameters of cutting speeds (65-76.5-90-104-119) m/min, feed rates (0.11-0.132-0.15-0.173-0.198) mm/rev. and depth of cuts (0.36-0.425-0.50-0.575-0.66) mm. The tools for experiment were coated carbide cutting tools that are formed by chemical vapour deposition while material used for the study was a mild steel, which is the generally used material for manufacturing industry. Furthermore, analysis of variance was used for determining the main factors/interactions. These results are compared and predicted through response surface methodology (RSM) and artificial neural network (ANN). Both approaches indicated that feed rate was dominant while others were not significant. Further, ANN’s predicion ability was higher than RSM because coefficients of determination were equalled to 0.995 and 0.925 for ANN and RSM, respectively, which were associated with mean squared errors of 0.00056 and 0.0088 for ANN and RSM, respectively. Moreover, percentage errors for square-shaped inserts in randomly selected test was 6.66% and 1.50% for RSM and ANN, respectively. Hence, two models are good at predicting surface roughness, which in turn can provide a very valuable tool for design and manufacturing of engineering applications, but ANN spends less time and gives more accuracy.