The aim of this work is to characterize the geometrical parameters of a 2D delamination: length, depth and thickness. First, we calculate analytically and numerically the surface temperature oscillation of a sample, containing a semi-infinite delamination, as the material is homogeneously illuminated by a modulated light beam. Then, we perform an inverse parametric estimation of synthetic temperature amplitude and phase data to size the geometrical parameters of the delamination. Finally, we present lock-in infrared thermography experiments performed on AISI-304 stainless steel samples containing calibrated delaminations. We fit the numerical model to the experiments to retrieve the delamination parameters successfully.

Noise in data can generate difficulty and even inaccuracy in its analysis. In some activities where diagnoses must have good technical precision, such as medical and dental examinations, non-destructive tests on industrial materials, agricultural monitoring, and others, noise present in the data can disturb this precision. In this work we present a statistical learning model, based on convolutional neural networks, capable of removing synthetic noise in infrared thermal images, obtained from a carbon fiber reinforced polymer plate.

Detecting micrometer-sized holes in the skin is a major problem in the NDT of stratospheric airships, and to the best of our knowledge, there are still very few researches so far on NDT methods for microporous defects in envelope materials. In this paper, we propose to accomplish the automatic detection of microporous defects using an improved network based on U-net. While using a laser beam as a heat source, three different laser beam extension methods were compared in order to improve the problems of uniform laser beam intensity distribution and small heating area, and a fiber laser beam shaper with excellent heating effect was chosen as the experimental method. Micron-sized holes of different sizes were produced in the epidermal material. The detection of pore defects with a minimum diameter of 75 µm can be accomplished using the improved network with an accuracy of 95.3%. The experimental results show that the proposed method is able to significantly detect debonding defects at the adhesive bond as well as pore defects at millimeter and micrometer scale.