[lucas]

Thank you for your interest and insightful comments. Here are some explanations:

1. Page 8: When we compare A-scan amplitudes of experimental and numerical results we followed two steps:
a) For experimental results, as you mentioned, we have converted the ultrasonic response energy to percentage.
b) For numerical results, the A-scan responses are not normalized, however, calibrated. The numerical software used in this study is CIVA (CEA, Fr) is based on semi-analytical finite element model calculations. As a result, we obtain “CIVA points” for each calculation and these points are converted into percentages.
The calibration value for all work is selected as the maximum amplitude observed from the defect center of 5 mm debonding. In order to match the response for each case, results are not normalized however numerical model parameters have been adjusted.

2. Variation model with metamodels: As you described, metamodels are models of models, or response surfaces, that are mathematical approximations that are used to reduce the computation time of specific points. In this work, we have calculated 1054 real models to build, or ‘train’, our metamodel based on the number and the range of selected uncertain parameters. While in theory, it is possible to build a metamodel with a lower amount of calculations, in our case, it would have increased the error caused by mathematical approximation.
Each MAPOD curve we have obtained is based on the calculated metamodel, as a result, the confidence bound and calculated PoD values are quite similar.

I hope this explains your questions, however, in case it is not clear, please feel free to continue the discussion.