Analyses of internal structures and defects in materials using physics-informed neural networks.

Area of Science:

• Computational mechanics
• Materials science
• Artificial intelligence

Background:

• Characterizing internal structures and defects in materials presents significant challenges, often involving complex inverse problems.
• Unknown topology, geometry, material properties, and nonlinear deformation complicate accurate material assessment.

Purpose of the Study:

• To develop a general framework for identifying unknown geometric and material parameters in materials with internal structures or defects.
• To address inverse problems in material characterization, quality assurance, and structural design.

Main Methods:

• Utilized physics-informed neural networks (PINNs) for parameter identification.
• Employed a mesh-free method to parameterize material geometry using a differentiable and trainable approach.
• Validated the framework on materials with internal voids/inclusions using constitutive models spanning linear elasticity, hyperelasticity, and plasticity.

Main Results:

• Successfully predicted the size, shape, and location of internal voids/inclusions.
• Accurately determined the elastic modulus of inclusions.
• Demonstrated the framework's capability to handle diverse material behaviors, including nonlinear deformation.

Conclusions:

• The presented general framework offers a robust solution for inverse problems in material characterization.
• The physics-informed neural network approach effectively identifies geometric and material properties in complex scenarios.
• This methodology has broad applicability in quality assurance and structural design for materials with unknown characteristics.