Probabilistic approaches to compute uncertainty intervals and sensitivity factors of ultrasonic simulations of a weld inspection.

Abstract For comprehension purpose, numerical computations are more and more used to simulate the propagation phenomena observed during experimental inspections. However, the good agreement between experimental and simulated data necessitates the use of accurate input data and thus a good characterization of the inspected material. Generally the input data are provided by experimental measurements and are consequently tainted with uncertainties. Thus, it becomes necessary to evaluate the impact of these uncertainties on the outputs of the numerical model. The aim of this study is to perform a probabilistic analysis of an ultrasonic inspection of an austenitic weld containing a manufactured defect based on advanced techniques such as polynomial chaos expansions and computation of sensitivity factors (Sobol, DGSM). The simulation of this configuration with the finite element code ATHENA2D was performed 6000 times with variations of the input parameters (the columnar grain orientation and the elastic constants of the material). The 6000 sets of input parameters were obtained from adapted statistical laws. The output parameters (the amplitude and the position of the defect echo) distributions were then analyzed and the 95% confidence intervals were determined.