AUT Journal of Mechanical Engineering

AUT Journal of Mechanical Engineering

Performance Assessment of Dynamic Mode Decomposition for Identifying Closely Spaced Modes in Structures

Document Type : Research Article

Authors
Department of Aerospace Engineering, Amirkabir University of Technology, Hafez Avenue, Tehran 159163-4413, Iran
10.22060/ajme.2026.26172.6310
Abstract
This study evaluates the capability of standard Dynamic Mode Decomposition (DMD) to identify closely spaced and nearly repeated structural modes. Two numerical benchmarks are considered: a four-degree-of-freedom (4-DOF) mass–spring system with controlled modal spacing and a proportionally damped simply supported plate with varying aspect ratio. Natural-frequency and damping-ratio errors are quantified using the Frequency Spacing Index (FSI) and Modal Overlap Factor (MOF). Robustness is further assessed by adding Gaussian measurement noise at SNR levels of 90, 80, and 70 dB using 100 realizations, while the Eigensystem Realization Algorithm (ERA) is employed as an independent benchmark. The results show that DMD can resolve very small frequency separations once an exact repeated condition is slightly perturbed. Natural-frequency estimates remain comparatively stable as modal overlap and noise increase, whereas damping-ratio estimates are more sensitive, particularly for higher-order modes. Measurement noise also reduces the maximum MOF for which two closely spaced modes can be reliably separated. Under noise-free conditions, DMD and ERA exhibit comparable resolution limits, while ERA generally maintains reliable modal separation over a wider MOF range as noise increases. These findings clarify the capability, sensitivity, and practical limitations of standard DMD for closely spaced structural modal identification.
Keywords
Subjects


Articles in Press, Accepted Manuscript
Available Online from 22 November 2026