LSWNet: A physics-informed neural network for ultrasonic wavefield prediction and elastic constant inversion in...

Area of Science:

• Materials Science
• Non-destructive Testing
• Computational Mechanics

Background:

• Accurate elastic constants are vital for ultrasonic defect detection in carbon fiber reinforced plastic (CFRP).
• Full-waveform inversion is computationally expensive due to wavefield simulation costs, hindering non-destructive in-situ characterization.
• Existing methods face challenges in computational efficiency and real-time application for CFRP analysis.

Purpose of the Study:

• To propose a novel physics-informed neural network (PINN) method, LSWNet, for efficient forward wavefield prediction and elastic constant inversion in unidirectional CFRP.
• To overcome the computational limitations of traditional wavefield simulations for non-destructive ultrasonic testing.
• To enable accurate in-situ characterization of elastic constants and high-resolution damage imaging in CFRP.

Main Methods:

• Developed a Longitudinal and Shear Wavefield Net (LSWNet) based on PINNs.
• Embedded elastic wave equations and ultrasonic measurement data as physical constraints for wavefield and elastic constant prediction.
• Utilized transfer learning from small-scale to large-scale models to accelerate convergence.
• Validated the method using finite element simulations and experimental data.

Main Results:

• LSWNet accurately predicted wavefields with low mean squared errors (≤ 3.2 × 10⁻³ compared to finite element simulations).
• The method successfully inverted elastic constants (C₆₆, C₁₃, C₄₄) close to actual values.
• Elastic constants derived from LSWNet improved delamination detection resolution when applied to the total focusing method.

Conclusions:

• The proposed LSWNet method effectively addresses forward and inverse problems in CFRP ultrasonic wavefield analysis.
• It enables efficient, in-situ characterization of elastic constants and high-resolution damage imaging.
• This approach offers a significant advancement for non-destructive testing and structural health monitoring of CFRP materials.