Surrogate recycling for structures with spatially uncertain stiffness

This study expands the existing methods for non-destructively identifying the spatially varying material properties of a structure using modal data. It continues a recently published approach to this inverse problem that employed Bayesian inference in conjunction with the KarhunenLoeve expansion to solve it. Here, we present two developments. Firstly, eigenvectors are used instead of eigenvalues, improving the results significantly. Secondly, a generalized polynomial chaos surrogate accelerates the inversion procedure. Finally, we develop a methodology for reusing the surrogate model across inversion tasks. We demonstrate the efficacy and efficiency of this methodology via the field of additive manufacturing and the fused deposition modeling process. The good results promise profound computational cost saving potential for large-scale applications.

Automated summary

This study expands the methods for non-destructively identifying material properties of a structure using modal data. It improves the results significantly by using eigenvectors instead of eigenvalues and accelerates the inversion process with a generalized polynomial chaos surrogate. A methodology for reusing surrogate models across inversion tasks is also developed.