torch-fem 0.2.0

Simple finite element assemblers with torch.

torch-fem: differentiable linear elastic finite elements

Simple finite element assemblers for linear elasticity with PyTorch. The advantage of using PyTorch is the ability to efficiently compute sensitivities and use them in structural optimization.

Basic examples

The subdirectory examples->basic contains a couple of Jupyter Notebooks demonstrating the use of torch-fem for trusses, planar problems, shells and solids.

Simple cantilever beam: There are examples with linear and quadratic triangles and quads.

Optimization examples

The subdirectory examples->optimization demonstrates the use of torch-fem for optimization of structures (e.g. topology optimization, composite orientation optimization).

Simple topology optimization of a MBB beam: You can switch between analytical sensitivities and autograd sensitivities.

Simple topology optimization of a 3D beam: The model is exported to Paraview for visualization.

Simple shape optimization of a fillet: The shape is morphed with shape basis vectors and MMA + autograd is used to minimize the maximum stress.

Simple fiber orientation optimization of a plate with a hole: Compliance is minimized by optimizing the fiber orientation of an anisotropic material using automatic differentiation w.r.t. element-wise fiber angles.

Installation

Your may install torch-fem via pip with

pip install torch-fem

Minimal code

This is a minimal example of how to use torch-fem to solve a simple cantilever problem.

from torchfem import Planar
from torchfem.materials import IsotropicPlaneStress

# Nodes and elements
nodes = torch.tensor([[0.0, 0.0], [1.0, 0.0], [2.0, 0.0], [0.0, 1.0], [1.0, 1.0], [2.0, 1.0]])
elements = torch.tensor([[0, 1, 4, 3], [1, 2, 5, 4]])

# Create model
cantilever = Planar(nodes, elements, material)

# Load at tip
cantilever.forces[5, 1] = -1.0

# Constrained displacement at left end
cantilever.constraints[[0, 3], :] = True

# Show model
cantilever.plot(node_markers="o", node_labels=True)

This creates a minimal planar FEM model:

# Solve
u, f = cantilever.solve()

# Plot
cantilever.plot(u, node_property=torch.norm(u, dim=1))

This solves the model and plots the result:

Benchmarks

The following benchmarks were performed on a cube subjected to a one dimensional extension. The cube is discretized with N x N x N linear hexahedral elements, has a side length of 1.0 and is made of a material with Young's modulus of 1000.0 and Poisson's ratio of 0.3. The cube is fixed at one end and a displacement of 0.1 is applied at the other end. The benchmark measures the forward time to assemble the stiffness matrix and the time to solve the linear system. In addition, it measures the backward time to compute the sensitivities of the sum of displacements with respect to forces.

Apple M1 Pro (10 cores, 16 GB RAM)

Python 3.10 with Apple Accelerate

N	DOFs	FWD Time	FWD Memory	BWD Time	BWD Memory
10	3000	0.75s	23.59 MB	0.59s	0.09 MB
20	24000	3.37s	439.23 MB	2.82s	227.05 MB
30	81000	3.09s	728.31 MB	1.47s	0.06 MB
40	192000	7.84s	807.41 MB	3.99s	217.56 MB
50	375000	16.29s	1211.27 MB	9.50s	433.30 MB
60	648000	30.78s	2638.23 MB	19.17s	1484.67 MB
70	1029000	55.14s	3546.22 MB	34.41s	1997.77 MB
80	1536000	87.83s	5066.81 MB	56.23s	3594.27 MB
90	2187000	131.09s	7795.83 MB	107.40s	5020.55 MB

AMD Ryzen Threadripper PRO 5995WX (64 Cores, 512 GB RAM)

Python 3.12 with openBLAS

N	DOFs	FWD Time	FWD Memory	BWD Time	BWD Memory
10	3000	1.42s	17.27 MB	1.87s	0.00 MB
20	24000	1.30s	160.49 MB	0.98s	62.64 MB
30	81000	2.76s	480.52 MB	2.16s	305.76 MB
40	192000	6.68s	1732.11 MB	5.15s	762.89 MB
50	375000	12.51s	3030.36 MB	11.29s	1044.85 MB
60	648000	22.94s	5813.95 MB	25.54s	3481.15 MB
70	1029000	38.81s	7874.30 MB	45.06s	4704.93 MB
80	1536000	63.07s	14278.46 MB	63.70s	8505.52 MB
90	2187000	93.47s	16803.27 MB	142.94s	10995.63 MB