Quality improvement of machine vision-based non-contact inspection of surface roughness in turning through adaptive neuro-fuzzy interference system

Abstract

Today's competitive global manufacturing scenario drives a strong demand for accurate non-contact automated measurement of surface roughness in turning operations. This paper investigates the potential of various possible combinations of a number of surface image features, such as, statistical features extracted from grey-level co-occurrence matrices, discrete cosine transform coefficient and discrete Fourier transform coefficient, in machine vision-based non-contact measurement of surface roughness of turned AISI1045 steel work pieces. Adaptive neuro-fuzzy interference system (ANFIS) models, each of which utilises a particular combination of the above-mentioned image features for accomplishing non-contact prediction of surface roughness, are developed and compared in this paper. Analyses of experimental data demonstrate that the approach, which utilises statistical features for non-contact measurement of surface roughness, outperforms the other approaches in terms of surface roughness prediction accuracy and yields substantial improvement in the accuracy level of machine vision-based non-contact measurement of surface roughness in turning.