Image Recognition of Shape Defects in Hot Steel Rolling

Abstract

A frequently occurring issue in hot rolling of steel is so-called tail pinching. Prominent features of a pinched tail are ripple-like defects and a pointed tail. In this report two algorithms are presented to detect those features accurately in 2D gray scale images of steel strips. The two ripple detectors are based on the second order Gaussian derivative and the Gabor transform, a localized Fourier transform, yielding the so-called rippleness measures. Additionally a parameter called tail length is defined which indicates to what extent the overall shape of the tail deviates from an ideal rectangular shape. These methods are tested on images from the surface inspection system at Tata Hot Strip Mill 2 in IJmuiden, it is shown that by defining a simple criterion in the feature space spanned by these two parameters a given set of strips can correctly be classified into pinched and non-pinched strips. These promising results open the way for the development of an automatic pinch detection system.