Physics-Informed Neural Network Modeling of Inflating Dielectric Elastomer Tubes for Energy Harvesting Applications

Area of Science:

• Soft Robotics
• Materials Science
• Energy Harvesting

Background:

• Dielectric elastomer tubes are promising for soft energy harvesting.
• Modeling their large deformation and electromechanical coupling is complex.
• Analytical and conventional numerical methods face challenges with nonlinearities and boundary conditions.

Purpose of the Study:

• To develop a physics-informed neural network (PINN) framework for modeling dielectric elastomer tubes.
• To accurately predict large deformations and coupled electromechanical responses for energy harvesting applications.
• To provide a mesh-free, data-efficient modeling approach.

Main Methods:

• Developed a PINN framework integrating incompressible neo-Hookean elasticity, radial electric loading, and gas inflation.
• Embedded governing nonlinear equations and deformation-dependent boundary conditions into the PINN's loss function.
• Utilized the Optuna algorithm for stable network training across coupled stages.

Main Results:

• Achieved accurate, mesh-free solutions without labeled data.
• Captured complex pressure-volume interactions, with internal volume increasing over 290% during inflation.
• Observed residual stretch up to 9.6 times the undeformed volume and pressure-dependent axial force behavior.
• Determined that harvested energy strongly depends on pressure, with voltage becoming significant above a critical threshold.

Conclusions:

• The PINN framework provides a robust and flexible tool for predictive modeling of soft energy harvesters.
• This approach overcomes limitations of analytical and conventional numerical methods for complex electromechanical systems.
• Enables efficient design and optimization of dielectric elastomer-based energy harvesting devices.