FENIAX is an aeroelastic toolbox written in Python using JAX. It acts as a post-processor of commercial software such as MSC Nastran.
Some of the key features of the software are:
- Arbitrary FE models built for linear aeroelastic analysis are enhanced with geometric nonlinear effects, flight dynamics and linearized state-space solutions about nonlinear equilibrium.
- Leveraging on the numerical library JAX and optimised algorithms, a high performance is achieved that leads to simulation times comparable to the linear counterparts on conventional platforms.
- The software runs on modern hardware architectures such as GPUs in a addition to standard CPUs.
- Algorithm differentiation (AD) of the aeroelastic response is available via JAX primitives.
- Concurrent simulations for multiple load cases have been developed.
- Currently the code has been tested and is developed in Linux and MacOS.
- A minimum installation into the current environment is possible by navigating to the main directory and
pip install .
- However developer mode is recommended and also installing the full set of packages which include testing and visualisation capabilities:
pip install -e .[all]
-
see pyproject.toml file for the options available. Python 3.10+ is required.
-
To install with GPU support install jax first:
pip install -U "jax[cuda12]"
pip install -e ".[all]"
Available at https://acea15.github.io/FENIAX/
The most relevant examples in the code base are shown here, these and more can be found in the folder /examples
They are also part of a large test suite that is integrated into the development using CI/CD.
!!! tip Navigate to the code of the various examples, including the simulation input settings and postprocessing of the simulation --exactly as it was used for the articles backing the software. See examples
!!! success - Validated with MSC Nastran nonlinear solution (sol 400) - AD differentiation of the response verified against finite-differences
!!! note Take a liner FE model of arbitrary complexity from your favourite FE solver, and turn it into a fully geometrically nonlinear model. You just need a condensation step into the main load paths and the resulting linear stiffness and mass matrices.
!!! success - Validated with MSC Nastran nonlinear solution (sol 400) - Runs over x100 faster than Nastran - AD differentiation of the response verified against finite-differences
This example first appeared in the work of Juan Carlos Simo (see Bio) , a pioneer in the field of computational structural mechanics and the
!!! success - Nonlinear aeroelastic response in our solvers takes similar times or less to the linear Nastran solution!! - Concurrent simulations for various loading settings achieve unparalleled computational times. - CPU VS GPU benchmarks available.
- Extremely large deformations
- Validation of concurrent solution
- 300 modes in the solution, 8 different loading scenarios running in parallel each with 11 substeps, 24 seconds in total on A100 GPU
- Nonlinear effects: follower aerodynamic forces, geometric stiffening, wing shortening.
- Steady manoeuvre varying flow conditions and AoA for a total 256 cases in 14 seconds.
- 512 different gust cases run on A100 GPU NVIDIA in 38 seconds!
- Rigid body modes included, rigid/elastic nonlinear couplings accounted for.
- Load envelopes available from the simulation.
