Aperi CMC

Solver

Aperi-Mech

Aperi-Mech is our flagship solver, designed to run efficiently with an AI agent. It includes an internal mesher designed for AI, so an agent can generate meshes efficiently without wasting tokens.

Capabilities

What it does

  • Supports unstructured meshes. Generate analysis-ready meshes from imperfect geometry or raw scans without manual cleanup.
  • Explicit transient dynamics, implicit dynamics, and statics. All done with matrix-free solvers, no global assembly required.
  • Performance portable: single codebase runs natively on CPUs and GPUs with no vendor lock-in.
  • Meshfree and finite element methods in a single framework.

Method

Conforming Reproducing Kernel

The Conforming Reproducing Kernel Method (CRK) is the result of extensive research focusing on the application of mesh-free methods to expedite simulation development. In most simulation processes, a significant bottleneck lies in the effort required to generate a suitable discretization, known as “meshing,” from the given geometry. This task is particularly critical for dynamic structural analyses, as computational costs are directly linked to mesh quality.

CRK offers an effective solution for simulating scenarios involving complex geometries, low-quality meshes, and nearly incompressible materials, all without increasing the computational cost to obtain accurate solutions.

Examples

Worked examples

This landslide simulation uses real terrain data from a lidar scan. No manual geometry cleanup was involved. The simulation uses explicit dynamics with a Drucker-Prager material model for the muddy soil in the middle of the hill. The topographic dataset came from the Geometrics Lab at Oregon State University, using a ground filter developed by EZDataMD to strip out vegetation, structures, and other noise.
Time series of a simulated landslide on the slope of the Moon's Tycho crater, at initial time, 80, 120, and 160 seconds
A simulation of a landslide on the Tycho crater, performed with Aperi-Mech on an STL pulled from NASA's 3D resources. Performance testing was done on an NVIDIA H100 GPU and AMD EPYC 7V12 processors. For reference, on this problem a single H100 GPU ran the simulation about as fast as 32 CPU cores.

References

Published work

  1. An agile design-to-simulation workflow using a new conforming moving least squares method
  2. Conforming window functions for meshfree methods
  3. Treatment of near-incompressibility in meshfree and immersed-particle methods