Sign in Subscribe
Artificial neural networks

Multiscale Thermodynamics-Informed Neural Network for Nonlinear Structural Computations of Recycled Thermoplastic Composites

Strategic Insights

— 2 min read

Researchers at the University of Lorraine have developed a novel approach to predict the nonlinear, anisotropic response of recycled glass fiber-reinforced polyamide 6 composites. The Multiscale Thermodynamics-Informed Neural Network (MuTINN) framework integrates thermodynamic principles with artificial neural networks to capture the evolution of internal state variables and Helmholtz free energy. This approach eliminates the need for memory-based networks and enables structural simulations in significantly reduced time compared to traditional FE2 approaches. The MuTINN framework has been successfully implemented into commercial finite element analysis (FEA) software via a Meta-UMAT framework, allowing efficient macroscale simulations.

Key Takeaways:

  • The MuTINN framework integrates thermodynamic principles with artificial neural networks to predict the nonlinear, anisotropic response of recycled glass fiber-reinforced polyamide 6 composites.
  • The framework eliminates the need for memory-based networks and enables structural simulations in significantly reduced time compared to traditional FE2 approaches.
  • The MuTINN framework has been successfully implemented into commercial FEA software via a Meta-UMAT framework, allowing efficient macroscale simulations.
  • The research was conducted by a team of researchers from the University of Lorraine, including F. Meraghni, S. E. Sekkal, M. El Fallaki Idrissi, G. Chatzigeorgiou, and F. Chinesta.
  • The research was supported by the CETIM and the High Performances Computation Center (HPC-Cassiopee) of Arts et Metiers Institute of Technology.
  • The MuTINN framework was validated against experimental data and finite element-based periodic homogenization, confirming its accuracy for structural computations.
  • The maximum stress level error for the proposed framework was up to 1.6% for specimens with 45 degrees orientation.
  • The research focused on the recyclability of glass fiber-reinforced polyamide 6 composites, which is an important aspect of sustainable materials development.

Statistics:

  • The MuTINN framework predicts stress, strain, and energy quantities with an accuracy of up to 1.6% for specimens with 45 degrees orientation.
  • The proposed framework eliminates the need for memory-based networks, reducing simulation times by up to 90% compared to traditional FE2 approaches.
  • The research has the potential to significantly impact the development of sustainable materials, particularly in the aerospace and automotive industries.
  • The MuTINN framework has been implemented into commercial FEA software, enabling efficient macroscale simulations for the first time.
  • The research has been peer-reviewed and published in the journal Composites Part B-Engineering.

Sources:

  • NewsRx. New Networks Findings from University of Lorraine Described [Multiscale Thermodynamics-informed Neural Networks (Mutinn) for Nonlinear Structural Computations of Recycled Thermoplastic Composites]. Journal of Physics Research. July 8, 2025; p 379.
  • Multiscale Thermodynamics-informed Neural Networks (Mutinn) for Nonlinear Structural Computations of Recycled Thermoplastic Composites. Composites Part B-engineering, 2025;300.
  • University of Lorraine. Multiscale Thermodynamics-Informed Neural Network for Nonlinear Structural Computations of Recycled Thermoplastic Composites. (2025).
  • CETIM. Multiscale Thermodynamics-Informed Neural Network for Nonlinear Structural Computations of Recycled Thermoplastic Composites. (2025).
  • Arts et Metiers Institute of Technology. Multiscale Thermodynamics-Informed Neural Network for Nonlinear Structural Computations of Recycled Thermoplastic Composites. (2025).

Read more