Looseness detection system of bolted joints using a VMD-based nonlinear transformation approach with deep residual network

Abstract Bolted structures are subject to various vibrations, external forces and environmental factors, all of which can reduce their structural stability and compromise the integrity of bolted connections. Detecting bolt loosening in advance is crucial, as these effects often cause bolts to become loose, potentially leading to structural failure or collapse. However, identifying looseness in complex or large structures poses significant challenges, particularly when there is insufficient prior information about the loose-fit condition. To address this issue, the present study proposes a novel detection system for bolted joint looseness, employing a variational mode decomposition (VMD)-based nonlinear transformation (NT) approach integrated with a deep residual neural network, under several underlying assumptions. The proposed method utilizes VMD to decompose transverse vibrational modes into intrinsic mode functions (IMFs), selectively extracting signals with desired modes. The NT method is then applied to scale and shift the extracted signals, transforming them into a form that facilitates approximate classification. Image-based spectrograms are generated from the differences between transformed and reference signals, which are subsequently analyzed by the deep residual network. To validate the proposed method, several plates with bolted joints are considered.