Mechanics-informed autoencoder enables automated detection and localization of unforeseen structural damage

Area of Science:

• Civil Engineering
• Mechanical Engineering
• Artificial Intelligence

Background:

• Growing infrastructure scale necessitates advanced Structural Health Monitoring (SHM).
• Existing SHM methods often lack scalability, affordability, and adaptability for diverse structures.
• There is a need for passive, automated SHM without complex baseline models.

Purpose of the Study:

• To introduce a novel, automated, and passive SHM approach for damage detection and localization.
• To develop a system that learns a bespoke baseline model for each structure.
• To reduce human intervention and inspection costs in structural maintenance.

Main Methods:

• Synergistic integration of passive sensor measurements, data compression, and a mechanics-informed autoencoder.
• Continuous learning and adaptation of a structure-specific baseline model from undamaged state data.
• Autonomous detection and localization of unforeseen damage after a short learning period (3 hours).

Main Results:

• The proposed autoencoder achieved up to a 35% improvement in detecting and localizing minor damage compared to standard autoencoders.
• Numerical simulations and experimental results validated the system's effectiveness.
• The system demonstrated autonomous damage detection and localization capabilities.

Conclusions:

• The "deploy-and-forget" SHM approach offers a scalable, inexpensive, and automated solution for structural monitoring.
• Incorporating mechanical characteristics enhances damage detection accuracy.
• This method promises to improve infrastructure lifespan, reliability, and sustainability through proactive maintenance.