A data-driven approach to characterizing nonlinear elastic behavior of soft materials.

Area of Science:

• Biomechanics
• Materials Science
• Computational Mechanics

Background:

• The Autoprogressive (AutoP) method, combining finite element analysis (FEA) and machine learning (ML), previously focused on linear elastic properties.
• Characterizing nonlinear elastic behavior in soft tissues and biomaterials under finite deformations remains a challenge.

Purpose of the Study:

• To extend the Autoprogressive (AutoP) method for characterizing nonlinear elastic mechanical behavior under finite compressive deformations.
• To enhance the learning speed and accuracy of AutoP by modifying training data based on nonlinear media priors.

Main Methods:

• Modified AutoP training data generation to incorporate nonlinear material priors.
• Validated AutoP using synthetic and experimental force-displacement data from 3D objects.
• Utilized ultrasonic imaging for experimental data acquisition from heterogeneous agar-gelatin phantoms.
• Independently measured material properties of phantom components for comparison.

Main Results:

• Successfully characterized nonlinear elastic mechanical behavior using the enhanced AutoP method.
• Neural network constitutive models (NNCMs) trained with AutoP accurately predicted material properties.
• Demonstrated robustness of AutoP results against measurement errors and spatial material property variations.

Conclusions:

• The extended AutoP method is effective for characterizing nonlinear elastic behavior in complex materials.
• AutoP provides accurate and robust material property identification, suitable for heterogeneous and error-prone data.
• This advancement enables better modeling of soft tissues and biomaterials under significant deformation.