Why faults are hard to grid
A fault is a break where rock has slipped, so layers on one side are offset — the throw — against different layers on the other. That creates three tangled problems for a grid:
- Discontinuity: the geometry is not smooth, so cells cannot simply flow across the break.
- Juxtaposition: offset puts different layers face to face — a permeable sand may now sit against shale, or against another sand — which sets whether fluids can cross at all.
- Fault transmissibility: the fault rock itself, through smear and gouge, has its own conductivity, so the fault is a flow property, not just a surface. It can seal or conduct.
Non-conforming vs conforming
Structured corner-point grids keep a Cartesian index, so they represent a fault by stair-stepping the cell faces and wiring offset layers together with non-neighbor connections. The true fault plane is never an actual face, and the cells around it distort.
A fault-conforming grid does the opposite: it places cell faces exactly on the fault surface. The fault becomes a real internal boundary of the mesh, so:
- a fault transmissibility multiplier applies to genuine faces,
- juxtaposition is captured by the real cell adjacency across the fault, and
- cells stay orthogonal and well-shaped, protecting two-point flux accuracy.
Unstructured PEBI/Voronoi grids achieve this because you can place nodes so their perpendicular-bisector faces land on the fault plane.
The build problem
Making Voronoi cells conform to a surface used to require clipping cells against it, which spawns slivers and degenerate cells — right where quality matters. That is the barrier VoroCrust (Dr. Mohamed Ebeida et al., ACM Transactions on Graphics, 2020) removed, providing a provably conforming polyhedral Voronoi mesh with no clipping.
How AutoMesh-Geo helps
AutoMesh-Geo applies this VoroCrust-style approach to build grids whose faces sit on real fault and horizon surfaces, so faults are modeled as the flow features they are. It is a core capability for faulted oil & gas and subsurface models.