DefinePK

Abstract

This study investigates the issue of automated defect inspection for aluminum alloy and proposes a new defect detection method based on a maneuver detector, i.e., the Generalized Likelihood Ratio GLR detector. In this method, the intensity of the defect-free aluminum alloy image is supposed to be Gaussian distributed, while the defect intensity usually follows other statistical distributions. In terms of this different statistic property between the normal and abnormal area in an aluminum alloy image, an unknown input is employed to model the change of intensity distribution. Under the assumptions, defect inspection problem is approximated as the detection of abrupt changes in stochastic dynamical system. Kalman filters are used to filter the image and the measurement residuals are estimated. Defects are located by statistical tests on measurement innovations using the Generalized Likelihood Ratio GLR test based maneuver detector. Experimental results exhibit effective defect detection for aluminum alloy with low false alarm.