Data-driven modeling of structural dynamics: improving measurement noise robustness with Augmented Dynamic Mode Decomposition

Accurate modeling of structural dynamics is essential for designing effective vibration control strategies in earthquake engineering. Traditional system identification techniques often struggle to distinguish intrinsic structural behavior from external forces, particularly in the presence of measurement noise. Dynamic Mode Decomposition with Control (DMDc) offers a data-driven approach to extracting system dynamics while accounting for external inputs. However, its performance degrades significantly in noisy environments, leading to inaccurate mode identification and misrepresented damping characteristics. To address this challenge, we introduce Augmented Dynamic Mode Decomposition with Control (AugDMDc), a novel extension that incorporates time-delay coordinates into DMDc to enhance noise robustness. By
embedding past measurements into the state-space representation, AugDMDc effectively reduces the impact of sensor noise while preserving the ability to separate a structure’s natural response from external excitations. We validate this approach using a structural dynamic system equipped with an Active Mass Damper (AMD), comparing the performance of DMDc and AugDMDc under various conditions.

Subjects

Dynamic Mode Decomposition (DMD)

DMD with Control (DMDc)

Augmented DMDc (AugDMDc)

Measurement Noise Robustness

Time-Delay Coordinates

System Identification

DDC Class

690: Building, Construction

Options

Data-driven modeling of structural dynamics: improving measurement noise robustness with Augmented Dynamic Mode Decomposition