An improved recursive singular spectrum framework for damage identification in bridge systems under environmental excitations

Recursive singular spectrum analysis (RSSA) provides a paradigm for online detection of structural damage. RSSA iteratively analyzes variations in segmented signal data, yet its empirical window length compromises generality: oversized windows obscure transients, whereas undersized windows attenuate damage-induced features. To address this issue, an improved algorithm is proposed by integrating the autoregressive with exogenous inputs model (ARX) with the recursive principal component analysis (RPCA) derived from the core RSSA equation. The algorithmic procedure is summarized as follows: The optimal window size is fixed by ARX; Timestamps are generated per-signal via Kalman filtering; Damage timing is identified by RPCA; Damage location is determined by pre-/post-damage energy evaluation. The proposed algorithm is verified to enable more precise identification of damage timing with regional localization capability in numerical simulations of steel truss bridges, and its effectiveness is validated on a laboratory-scale truss structure. • The ARX model addresses the issue of indeterminate window size in the RSSA algorithm. • The Kalman filter generates a distinct timestamp for each signal value in real-time. • Damage location is automatically identified using a multi-sensor energy ratio index.