Deep material network

  • Novel machine learning approach based on network structure and mechanistic building block

  • Materials: rubber composite, polycrystalline materials, CFRP (e.g. UD and woven composite)

  • Small-strain and finite-strain formulations in 2D&3D

  • Advantages

    • Avoiding extensive offline sampling stage

    • Eliminating the need for extra calibration and micromechanical assumptions

    • Efficient online predictions without the danger of extrapolation

    • Arbitrary material laws in online prediction stage

    • Linear computational complexity to the number of degrees of freedom.

  • Application​s

    • Topology learning of RVE​

    • Seamless structure-property relationship and material design

    • Scale linking via direct network concatenation (e.g. three-scale CFRP)

Self-consistent clustering analysis

  • Mechanistic RVE model reduction based on clustering technique and micromechanics theory

  • Materials: Particle-reinforced composite, polycrystalline materials

  • Small-strain and finite-strain formulations in 2D&3D

  • Advantages

    • Avoiding extensive offline sampling stage

    • Explicit mapping between clusters and RVE parts

  • Applications​​

    • Concurrent multiscale simulation ( integrated with LS-DYNA)