A three dimensional (3D) hybrid method combining the classical finite element (FE) method with the semi-analytical finite element (SAFE) technique is developed. This hybrid method is employed to study the interaction of guided waves with non-axisymmetric damages in cylinders. The near field surrounding the damage is analysed with the 3D FE method. The solution is expanded into sums of guided modes on both inlet and outlet cross-sections. Such eigenmode expansions enable separation into ingoing and outgoing waves, i.e., incident, reflected and transmitted waves. Using the SAFE method, elastic guided modes are then computed at the aforementioned cross-sections thus reducing the analysis to two dimensions (2D). The amplitudes of the incident modes are imposed, whereas those of the scattered modes are determined by solving the global system of the 3D hybrid FE-SAFE model. In this paper, a formula is proposed for the calculation of eigenforces and modal power flows from eigendisplacements and SAFE matrices. This has the advantage of simplifying the post-process of load eigenvectors in hybrid FE-SAFE methods. Results obtained for a vertical free-end cylinder are in good agreement with those published in the literature. Moreover, first results of the interaction of the fundamental compressional, flexural and torsional Pochhammer-Chree modes with non-axisymmetric vertical cracks are obtained and discussed. Then, the interactions of the fundamental compressional mode with oblique free-ends and cracks are briefly addressed. The power balance is shown to be satisfied with a good accuracy

Green's function can be retrieved from cross-correlation of a diffuse field generated by noise sources. Today, this important result is the fundamental of several passive imaging techniques. The aim of this paper is to establish the suitability of these methods to detect and locate a defect in a reverberant elastic plate. The relations between the noise cross-correlation function over a few number of noise sources and the imaginary part of the Green's function are derived and numerically validated. Then we show through numerical experiments that this technique is exploitable for defect detection and localization in a differential mode, despite a non-perfect estimation of the Green's functions. Finally, a filtering technique based on the singular value decomposition is shown to improve the detection. © 2014 AIP Publishing LL