Metamodel-based nested sampling for model selection in eddy-current testing

Caifang Cai (1) , Sandor Bilicz (2) , Thomas Rodet (3) , Marc Lambert (4) , Dominique Lesselier (1)

1 L2S - Laboratoire des signaux et systèmes
2 BME - Budapest University of Technology and Economics [Budapest]
3 SATIE - Systèmes et Applications des Technologies de l'Information et de l'Energie
4 GeePs - Laboratoire Génie électrique et électronique de Paris

Flaw characterization in eddy current testing usually involves to solve a non-linear inverse problem. Due to high computational cost, sampling algorithms are hardly employed since often requiring to evaluate the forward model many times. However, they have good potential in dealing with complicated forward models. Here, we replace the original forward model by a computationally-cheap surrogate model. Then, we apply a Markov Chain Monte Carlo (MCMC) algorithm to tackle the inversion. The expensive database training part is shifted to off-line calculation. So, we benefit from the MCMC algorithm due to its high estimation accuracy, and do not suffer from the computational burden

Keywords: metamodel; MCMC; statistical inversion; Index Terms Nested Sampling; model selection; marginal likelihood; non-destructive testing; eddy-current; surrogate model

Domains: Engineering Sciences [physics]; Signal and Image processing; Engineering Sciences [physics]; Electromagnetism; Mathematics [math]; Mathematical Physics [math-ph]

Caifang Cai, Sandor Bilicz, Thomas Rodet, Marc Lambert, Dominique Lesselier. Metamodel-based nested sampling for model selection in eddy-current testing. IEEE Transactions on Magnetics, 2017, 53 (4), pp.6200912. ⟨10.1109/TMAG.2016.2635626⟩. ⟨hal-01397025⟩