Dynamics-Embedded Sparse Bayesian Inversion for Robust Transient Impact Identification in Marine Structures

Abstract

To address load reconstruction misidentification caused by low-damping free-decay oscillations in thin-walled engineering structures under transient impact, a dynamics-embedded sparse Bayesian inference (DE-SBI) framework is developed for impact load reconstruction. The proposed method embeds a representative Duhamel temporal kernel into Gaussian temporal atoms and combines it with spatial Gaussian load atoms, finite-element modal strain mapping, and automatic relevance determination (ARD)-based sparse Bayesian inference. This formulation yields a dynamics-embedded spatiotemporal dictionary, which reduces the risk that conventional quasi-static dictionaries misinterpret structural ringing responses as sustained external loads. Controlled numerical validation on a representative stiffened plate structure shows that DE-SBI can effectively reconstruct the impact load histories and spatial distributions under single-impact, off-grid impact, and spatiotemporally overlapping dual-impact cases. Compared with Tikhonov regularization, Lasso regularization, and standard SBI, DE-SBI exhibits more stable identification performance in terms of correlation coefficient, peak error, and spatial relative error. Further analyses of parameter sensitivity, sensor layout, noise perturbation, and Duhamel temporal-kernel mismatch indicate that the method maintains good robustness under the considered controlled numerical conditions. These results provide an interpretable dynamics-embedded Bayesian inversion strategy for transient impact load identification under sparse-observation conditions.

Bibliographic Information