Projects - Tessella

Tessella: Automated FEA Battery Analysis Pipeline

(Still in active development and defense.) A computational Python pipeline built to analyze the ion transport behavior in solid-state battery materials

Problem Solved

Statistically determine the most likely interface diffusion value from pre-generated probability density functions (PDF) generated from a kernel density estimation (KDE) fitting.

Approach / Physics

To determine the PDF plots (see the associated photos), the following five steps procedure is followed:

Material Characterization: From an EBSD scan, determine the seed and boundary regions sizes for computational model generation
CPVT Generation: Using the seed region information, the seed points of the resulting Voronoi tessellation are generated using the Controlled Poisson Voronoi Tessellation algorithm.
Voronoi Tessellation Generation: From the seed points, the Voronoi tessellation is generated using the computational geometry Python library Shapely
Mesh Generation and Diffusion Simulation: The resulting Voronoi tessellation is translated to the Gmsh geometry format, which is then meshed. This mesh is then solved with appropriate boundary conditions using the Python finite element anaysis framework Simple Finite Elements (SFePy).
Effective Diffusion Calculation: After solving the domain, using homogenizing volume-averaging methods, the effective diffusion tensor is determine after solving two independent simulations with affine boundary conditions.

This pipeline is repeated around 1000 times, extracting the components of the effective diffusion tensor each time to form the PDF to be determined using KDE.

Outcome / Validation

KDE generated, giving a broad range of interface diffusion mean values with a list of coefficients of variance.

Stack / Tools

Resulting probability density functions for interface diffusion means (0.025, 0.05, 0.075, 0.1 nm^2/s)

Voronoi tessellation generated with a large seed size of 1.8 nm, small size of 0.9 nm, and mean size of 1.2 nm. The percentage of seed within the large and small seed sizes is 85%. The domain size is 10 x 10 nm, with the boundary size of 0.12 nm. Note that each subsequent generation with the same parameters produces a different structure.