Automated defect localization via low rank plus outlier modeling of propagating wavefield data

Area of Science:

• Materials Science
• Acoustics
• Non-Destructive Evaluation

Background:

• Laser-based nondestructive evaluation (NDE) methods analyze acoustic wavefields for material diagnostics.
• Identifying structural anomalies often requires detailed knowledge of material and mechanical properties, which can be complex or unknown.

Purpose of the Study:

• To propose an agnostic inference strategy for material diagnostics in laser-based NDE.
• To identify unknown deviations in acoustic wavefields caused by material defects without prior structural or material property knowledge.

Main Methods:

• Coupling spatiotemporal windowing with low-rank plus outlier modeling.
• Analyzing acoustic wavefields measured on a structure's surface using a scanning laser interferometer.
• Demonstration in a simulated environment with point and line defect localization.

Main Results:

• Successfully identified a priori unknown deviations in propagating wavefields.
• Localized benchmark point and line defects in a thin plate using flexural waves.
• Demonstrated the strategy's effectiveness in a simulated diagnostic scenario.

Conclusions:

• The proposed agnostic inference strategy is suitable for material diagnostics when models are complex, unknown, or unreliable.
• This approach enhances the capabilities of laser-based NDE for detecting material inhomogeneities and defects.
• The method offers a robust solution for identifying anomalies without relying on predefined structural or material information.