Non-destructive testing of steel-lined concrete structure using multiple agents with deep prior

Area of Science:

• Non-destructive testing
• Structural Health Monitoring
• Ultrasonic imaging

Background:

• Non-destructive testing (NDT) is crucial for assessing structural integrity without causing damage.
• Ultrasonic Model-Based Iterative Reconstruction (UMBIR) and its multi-frequency extension (MF-UMBIR) improve ultrasonic imaging but have limitations in complex structures.

Purpose of the Study:

• To introduce the Deep Multi-Agent Consensus Equilibrium (Deep-MACE) method for high-fidelity ultrasonic imaging.
• To reconstruct images revealing structural damage like corroded rebars and delaminations in steel-lined concrete.

Main Methods:

• Developed Deep-MACE, integrating multi-frequency forward model agents with a deep learning prior (U-Net) via consensus equilibrium.
• Combined data consistency across frequencies with a learned prior for acoustically heterogeneous environments.

Main Results:

• UMBIR and MF-UMBIR showed limitations in defect visibility and robustness on steel-lined concrete.
• Deep-MACE produced clearer reconstructions, identified all internal rebars, reduced artifacts, and improved spatial resolution.

Conclusions:

• Deep-MACE offers superior performance for non-invasive Structural Health Monitoring (SHM) in challenging environments.
• Integrating deep priors into multi-agent frameworks shows significant potential for advanced ultrasonic NDT.