Inverse problem filtering for noise reduction in QNDE

Model-based nondestructive evaluation proceeds measuring the response after an excitation on an accessible area of the structure. The basis for processing this information has been established in recent years as an iterative scheme that minimizes the discrepancy between this experimental measurement and a sequence of measurement trials predicted by a numerical model. The unknown damage that minimizes this discrepancy by means of a cost functional is to be found. The damage location and size is quantified and sought by means of a well-conditioned parametrization. The design of the magnitude to measure, it's filtering for reducing noise effects and calibration, as well as the design of the cost functionl and parametrization, determines the robustness of the search to combat noise and other uncertainty factors. These are key open issues to improve the sensitivity and identifiability during the information processing. Among them, a filter for the cost functional is proposed in this study for maximal sensitivity to the damage detection of steel plate under the impact loading. This filter is designed by means of wavelet decomposition together with a selection of the measuring points, and the optimization criterion is built on an estimate of the probability of detection, using genetic algorithms. Numerical examples show that the use of the optimal filter allows finding damage of a magnitude several times smaller.