A physical reservoir involving frequency virtual nodes for structural damage detection.

Area of Science:

• Structural Health Monitoring
• Nonlinear Dynamics
• Computational Physics

Background:

• Conventional structural damage detection methods often require extensive data and computational resources.
• Physical reservoir computing (PRC) offers a promising alternative by utilizing the inherent dynamics of physical systems for computation.
• Integrating PRC with structural systems for health monitoring remains an underexplored area.

Purpose of the Study:

• To propose and validate a novel physical reservoir computing approach for structural damage detection.
• To leverage the target structure itself as a physical reservoir for damage classification.
• To demonstrate the efficiency and accuracy of this method compared to traditional techniques.

Main Methods:

• A nonlinear feedback system was designed, creating "frequency virtual nodes" by processing structural response signals in different frequency bands.
• These processed signals underwent nonlinear activation and feedback, transforming the structure into a high-dimensional physical reservoir.
• Damage classification was achieved by analyzing changes in the structure's dynamical behavior within this PRC framework.

Main Results:

• The proposed PRC method successfully classified structural damage in numerical and experimental models (thin plate, thin-walled tube).
• Damage classification accuracy was comparable to conventional neural networks.
• Significant reductions in training and inference costs were observed compared to traditional methods.

Conclusions:

• The study demonstrates the feasibility of using a structure as its own physical reservoir for efficient damage detection.
• This PRC framework presents a novel and cost-effective approach for structural health monitoring.
• The findings highlight the potential of PRC for real-world structural monitoring applications.