Neural network based material identification and part thickness estimation from two radiographic images

Abstract

Radiographic inspection provides extensive information about the characteristics and conditions of parts even if they are hidden behind the walls. A dual energy method is originally proposed to estimate the thickness of parts from two radiographic images by using analytical expressions. Use of neural networks is proposed when the material properties have nonlinear characteristics. Aluminum and brass test pieces were identified and their thickness was estimated from two images obtained at different energy levels. © Springer-Verlag Berlin Heidelberg 2003.