Efficient Propagation of Uncertainty in Simulations via the Probabilistic Collocation Method (postprint)

Eddy current models have matured to such a degree that it is now possible to simulate realistic nondestructive inspection (NDI) scenarios. Models have been used in the design and analysis of NDI systems and to a limited extent, model-based inverse methods for Nondestructive Evaluation (NDE). The science base is also being established to quantify the reliability of systems via ModelAssisted Probability of Detection (MAPOD). In realistic situations, it is more accurate to treat the input model variables as random variables rather than deterministic quantities. Typically a Monte-Carlo simulation is conducted to predict the output of a model when the inputs are random variables. This is a reasonable approach as long as computational time is not too long; however, in most applications, introducing a flaw into the model results in extensive computational time ranging from hours to days, prohibiting Monte-Carlo simulations. Even methods such as Latin-Hypercube sampling do not reduce the number of simulations enough for reasonable use. This paper presents the Probabilistic Collocation Method as a non-intrusive alternative to other uncertainty propagation techniques.