Deep learning architectures for data-driven damage detection in nonlinear dynamic systems

TL;DR

The paper investigates unsupervised, data-driven damage detection in nonlinear dynamic systems using 1D CNN autoencoders and deep GANs trained on random vibrations without prior system knowledge. Through numerical studies on 1-DOF and 2-DOF cubic nonlinearities, a seismic isolator, and an experimental magneto-elastic setup, both architectures detect damage as excitation varies, with autoencoders generally offering more consistent performance and training efficiency, while GANs provide a quantitative damage index via the discriminator. The findings support the viability of deep learning for output-only SHM in nonlinear regimes and offer guidance on method selection based on robustness and interpretability, while highlighting areas for future work such as environmental robustness and damage localization. Overall, the work extends data-driven SHM to nonlinear dynamics, demonstrating practical applicability and outlining paths for further refinement and validation.

Abstract

The primary goal of structural health monitoring is to detect damage at its onset before it reaches a critical level. The in-depth investigation in the present work addresses deep learning applied to data-driven damage detection in nonlinear dynamic systems. In particular, autoencoders (AEs) and generative adversarial networks (GANs) are implemented leveraging on 1D convolutional neural networks. The onset of damage is detected in the investigated nonlinear dynamic systems by exciting random vibrations of varying intensity, without prior knowledge of the system or the excitation and in unsupervised manner. The comprehensive numerical study is conducted on dynamic systems exhibiting different types of nonlinear behavior. An experimental application related to a magneto-elastic nonlinear system is also presented to corroborate the conclusions.

Figures (17)

• Figure 1: Implemented AE architecture based on 1D CNNs for damage detection in nonlinear dynamic systems.
• Figure 2: Implemented GAN architecture based on 1D CNNs for damage detection in nonlinear dynamic systems.
• Figure 3: Frequency response curves (FRCs) of the 1-DOF system for different excitation amplitudes and damage levels.
• Figure 4: Reconstruction of the 1-DOF system response through AE for different damage levels.
• Figure 5: Time-history and corresponding time-frequency representation via wavelet transform of the undamaged 1-DOF system response: actual data and results carried out from the GAN generation block.