Surface roughness prediction for the milling of Ti–6Al–4V ELI alloy with the use of statistical and soft computing techniques

Abstract

This study focuses on Ti–6Al–4V ELI titanium alloy machining by means of plain peripheral down milling process and subsequent modeling of this process, in order to predict surface quality of the workpiece and identify optimal cutting parameters, that lead to minimum surface roughness. For the purpose of accomplishing this task a set of experiments were performed on a CNC milling centre and design of experiments based on Box Behnken Design (BBD) for a three factor and three level central composite design concept was conducted. Depth of cut, cutting speed and feed rate were selected as input parameters and surface roughness was measured after each experiment performed. At first, Response Surface Methodology (RSM) was employed for establishing a quadratic relationship between input and output parameters. Analysis of variance (ANOVA) was then conducted for the evaluation of the proposed formula. RSM was also used for the optimization analysis that followed for the determination of milling cutting parameters for minimum surface roughness. The analysis indicates that the use of BBD can reduce the number of experiments needed for modeling and optimizing the milling operation of Titanium alloys. Furthermore, this method is able to provide models that can reliably be used for any cutting conditions within the limits of the input data. Finally, Artificial Neural Networks (ANN) models were developed to allow for a more robust simulation model to be built and comparison between ANN and RSM models to be performed. From the presented results, for RSM, the mean square error and the correlation coefficient were determined to be 8.633×10−3 and 0.9713, respectively; for ANN models, the corresponding values were 2×10−3 and 0.9824, for the test group of the optimum model. Simulations indicated that, although input data were too few, a considerably reliable ANN model was able to be built and despite of its complexity compared to RSM model, it was proven to be superior in terms of prediction accuracy.