Search

International audience; Cellular materials, in particular metal or polymer foams and beam lattices, are characterized by a complex architecture where the material is concentrated in slender struts. This study is concerned with the construction of numerically efficient samplespecific digital twin for the analysis and characterization of the mechanical behavior of such meta-materials. A method to build an explicit, light, and analysis-suitable isogeometric beam model directly from digital images is proposed. More precisely, as a main contribution, the Virtual Image Correlation method is extended to the case of multiple interconnected beams with varying cross-sections. Special care is also given to the initialization for better efficiency and robustness. Synthetic and real examples of increasing complexity, i.e. curved beams, lattices, and foams, are presented to account for the performance of the developed workflow from digital images to an isogeometric twin for mechanical modelling and simulation.

International audience; This paper is concerned with analytical developments of results firstly introduced by the authors in [1]. These developments are devoted to the optimization of upper bounds of the interval covariance matrices appearing in the Interval Kalman Filter [2]. The proposed study is mainly highlighted through two aspects. Firstly, the optimization is further performed by considering a class of upper bounds and minimizing the traces of these bounds in two stages (in terms of a gain matrix and then with respect to a scalar parameter). Secondly, the paper provides conditions under which the optimal trace value is controlled and hence the proposed Algorithm in [1], namely Optimal Upper Bound Interval Kalman Filter (OUBIKF), is ensured to perform with stability (i.e. without width explosion of the resulting interval estimators). Also under these conditions, the OUBIKF Algorithm, having a similar structure of the Standard Kalman Filter (SKF), is ensured to get a smaller trace upper bound of the covariance matrices in the correction step than the one in the prediction step. Numerical simulations based on anautomotive model is performed to illustrate the developed results.