Automatic Detection of Defects in Riveted Lap-joints using Eddy Current

A method for detection and classiication of defects originating from rivet holes in a riveted lap-joint using eddy current (EC) inspection is presented in this paper. The EC-signals were acquired from a number of lap-joints using a tailor-made deep penetrating probe, and a digital single frequency EC-instrument. A number of rivets in the lap-joints had defects located in the second layer of the lap-joint. The problem consisted in detecting and classifying defect signals in presence of strong interference due to rivet responses. The preprocessing before detection consisted of median ltering, rotation and de-biasing of the eddy current pattern. The rotation was performed so that the signal energy from the rivet responses was maximized along the quadrature direction, and the defect response energy was maximized in the in-phase direction of the EC-signal. Feature extraction was then performed using wavelets, PCA and block-mean values of the defect signal. The classiication was performed using a standard multi-layer perceptron neural network.