M:3L Lab

arXiv preprint

Neural-field design of broadband Rayleigh-wave carpet cloaks under microstructure realisability constraints

A neural-field and differentiable-FEM framework for designing broadband Rayleigh-wave carpet cloaks within realizable Cauchy materials and mapping the optimized fields to explicit manufacturable microstructures.

Publication record

Status
Preprint published on arXiv
Authors
David Aznaurov, Davit Piliposyan, Danila Rukhovich, Sébastien Guenneau
Venue
arXiv preprint
Research area
Physics-Driven Machine Learning and Metamaterials
First arXiv submission

Abstract / summary

The paper formulates two-dimensional Rayleigh-wave carpet-cloak design as a PDE-constrained optimization problem within physically realizable Cauchy elasticity. A coordinate-based neural field and differentiable finite element solver optimize symmetric stiffness and density fields for single-frequency and broadband performance. The optimized fields are then mapped to explicit microstructures using homogenization, conditional diffusion, neural-field inverse design, and nearest-neighbor selection.

Methodology

  • PDE-constrained optimization of a two-dimensional Rayleigh-wave carpet cloak
  • Coordinate-based neural field coupled to a differentiable finite element solver
  • Single-frequency and broadband optimization of symmetric Cauchy stiffness and density fields
  • Microstructure-constrained realization using homogenization, conditional diffusion, neural-field inverse design, and nearest-neighbor selection
  • Validation with homogenized and fully resolved finite element simulations

Datasets and benchmarks

  • Database of homogenized microstructures

Principal results

  • The optimized Cauchy design approaches the ideal transformation-based cloak in finite element simulations.
  • After projection to explicit microstructures, the homogenized model recovers approximately 97% of the defect-free reference surface-displacement magnitude.
  • Direct simulation of the fully resolved microstructured geometry recovers approximately 76% at the design frequency.

Technologies and methods

  • Neural fields
  • Differentiable finite element methods
  • PDE-constrained optimization
  • Conditional diffusion
  • Homogenization
  • Finite element simulation

Related M:3L research

Evidence boundary

Benchmark claims and numerical values are paper-reported results for the named protocol. Potential applications are not evidence of completed deployments.