Surface roughness prediction as a classification problem using support vector machine

Abstract

In order to achieve a high level of product quality, it is imperative to gain a high degree of predictability especially in automated manufacturing setup. Surface finish is one of the most important measures for determining the quality of products in machining. Therefore, accurate predictive models for surface finish are needed. This paper utilizes vibration signals that are experimentally obtained during the end milling of aluminum plates at different cutting conditions. Several features are extracted by processing the acquired signals in both the time and frequency domains. The feature sets include statistical parameters, fast Fourier transforms (FFT) spectra, and the wavelet packets. This work introduces a classifier based on a support vector machine to analyze the set of features in order to predict the type of surface finish. Experiments are conducted for three different types of kernels and parameter configurations. One objective is to examine the effect of feature reduction on the performance of the proposed classifier using three different feature selection algorithms. Another objective is to compare the results with k-nearest neighbor, decision tree, and random forest classifiers. The results show the effectiveness of feature reduction and support vector machine in the success of the proposed classifier.