Detection of cracks in weep holes using neural networks.

Weep holes are incorporated in the risers of wing structures in C-141 aircraft to allow fuel to be properly distributed during flight. Two types of fatigue cracks have been observed emanating from weep holes, top cracks that emanate upward and bottom cracks that propagate toward the wing surface. A new approach using creeping Rayleigh waves has been shown to be successful in detecting top cracks in weep holes. The focus of this paper is the development of a robust automated protocol for the detection of both top and bottom cracks in weep holes. A numerical technique was developed using the boundary element method (BEM) to calculate simulated waveforms for reflection and backscattering of ultrasound from weep holes with top or bottom cracks. The simulations were validated by comparison with experimental results. From parametric studies of this model, a robust protocol incorporating both the pulse echo and pitch catch creeping Rayleigh wave signals from top-cracks was defined. Protocols incorporating neural networks were developed for both top and bottom crack classification. A complete automated procedure for crack detection was successfully trained with both simulated and experimental data and was verified with experimental data for the expected variation in weep hole size and crack length. © 2000 American Institute of Physics. [ABSTRACT FROM AUTHOR]