Abstract

This paper presents the threshold based image denoising methods for turned images using hard threshold, soft
threshold, and an improved threshold method. The turning experiments were carried out on a CNC turning
center, on Ti-6Al-4V bars under different cutting conditions namely speed, feed, and depth of cut. A computer
vision system has been used to capture different turned surface images. The denoising results of the new improved
threshold method is superior than the soft- and hard-threshold methods with minimum mean square error and
maximum peak signal to noise ratio. Wavelet packet transform have been used for extracting the statistical
features namely mean, variance, standard deviation, skewness, and kurtosis from the denoised turned surface
images. These statistical features along with cutting conditions namely speed, feed rate, and depth of cut along
with tool flank wear have been used as inputs to radial basis function neural network for modeling and prediction
of surface roughness parameter (Ra) from the turned images.