Vision-based inspection for periodic defects in steel wire rod production

Abstract

Defective rollers in rolling plants often create spatially periodic defects on the product surface. It is essential for operators to receive early warnings of such periodic defects and take appropriate action to avoid a large quantity of defective products. We present a robust and efficient real-time method for inspecting periodic defects in continuous steel wire rods. The proposed inspection method consists of two parts: (1) an algorithm for detecting all defect candidates and (2) an algorithm to detect periodicities in the detected defect candidates and to determine their periods. The first algorithm exploits the translation-invariant property of the Haar undecimated discrete wavelet transform to improve the signal-to-noise ratio of the surface image. The second algorithm is based on analysis of the frequency spectrum of defect candidate positions. It identifies periodicities of the defect candidates by finding repeating impulse-like spectra. The inspection system using the proposed method is applied to a real production line and detects periodic defects in real time and identifies their periodicities with high accuracy.