Adaptive TFM imaging in anisotropic steels using optimization algorithms coupled to a surrogate model
Résumé
An optimization method is studied to enhance the reliability of TFM (Total Focusing Method) images in anisotropic nuclear materials. The method is able to adapt to a given anisotropic structure (weld, cladded steel) when the parameters governing the wave propagation are uncertain. The optimization scheme combines a surrogate model to bypass the extensive computation times of the propagation forward model, and a gradient descent algorithm to minimize a multivariate cost function. The gradient-based local optimization is compared with a global optimization tool, the Particle Swarm algorithm. Finally, the parameters (stiffness constant, grain orientation, cladding thickness…) corresponding to the optimal TFM image are compared with those measured by other characterization techniques.