A Rapid Method for Analyzing Seismic Damage of Pile Group Foundations Based on Hilbert-Huang Transform

Weak points in bridge pile foundations with varying degrees of performance degradation may sustain corresponding damage under seismic loading. This study proposes a method integrating the Hilbert-Huang Transform (HHT) with deep learning, enabling the detection of damage severity and location in pile groups. The approach first develops an encoder-decoder deep learning model centered on a gated recurrent unit (GRU) to characterize the nonlinear relationship between seismic inputs and pile group displacement time histories. Subsequently, the pile group displacement responses are decomposed into intrinsic mode functions (IMFs) and subjected to the Hilbert transform to capture the time-frequency characteristics of the response within a two-dimensional time-frequency spectrum. By analyzing frequency variations before and after damage, the method enables rapid assessment of the extent and location of local damage within the pile group system.