Two-Step Algorithmic Adaptive Particle Filter for System Identification of Sudden Structural Damage

This paper proposes a new algorithmic adaptive tracking rule to advance the applicability of the particle filter (PF) algorithm for dynamic system identification of shear building structures undergoing abrupt property changes during an earthquake event. In particular, to rapidly identify the system states after sudden major structural damage using real-time observations of response time series, an adaptation rule is proposed, which can detect the occurrence of damage by monitoring innovation patterns and estimate the postdamage system states by artificially expanding the parameter search space. Leveraging the conventional adaptive tracking rules and combining them with a modified innovation definition, the structural damage is detected in real time in terms of the outliers in the time history of jumping parameters. Once the damage is detected, the parameter search space is enlarged to cover a broad range of possible damage scenarios, and, accordingly, time/measurement steps are repeated to recalibrate the system parameters and adjust to the postdamage states. Once the parameters are updated, the algorithm continues to monitor additional occurrences of damage by monitoring the jumping parameters. The proposed adaptation rule is designed to restrain the addition of new hyperparameters, deliberately avoiding the introduction of regularization techniques that require manual tuning. The proposed approach is applied to a 6-story shear building model, of which the nonlinearity is characterized by the Bouc–Wen hysteresis model. The numerical case study results using different earthquake records and various damage scenarios demonstrate the effectiveness of the proposed adaptive approach compared to the plain PF and existing adaptive tracking algorithms.