[jankovz]

Good morning! This is my first multiphase CFD project, and I'd like to lay out all of my choices and ask if they seem reasonable. The case is physically similar to the dam break tutorial, except in 3D and with more complex geometry.

Background:
I have written a simplified internal room flooding code, and am trying to validate it for an example case in Fluent. The geometry is shown below with initial density (red water, blue air) and mesh, and is only approximate. No smoothing of geometry has been done.

The two large tanks are initially filled with water, which can leak into a nearby building. This building contains a ground floor connected to a basement via a hole in the floor (representing a staircase), two basement rooms connected to each other by a door, and a final sump. I am interested only in the flow between rooms, so I will be using a mass flow monitor across each of the six flow paths.

Current Setup:
- ANSYS Workbench 15, double precision, parallel
- 14 Frozen bodies, meshed using automatically-detected connections and default settings.
- 775,244 elements, 189,246 nodes, 0.988 maximum skewness, 54.6 maximum aspect ratio
- Using VOF (also trying mixture) model with implicit body force, transient, gravity on, with air as primary phase and water as secondary. Operating density is set to 1.2.
- Laminar flow model for now, eventually k-e once the model is more stable.
- The top of the tank closest to the building and the top of the ground floor use a pressure outlet boundary condition. The intent is an open vent, that air may move freely across the boundary.
- All other boundaries are either interior, interface, or wall.
- Fixed time steps of 0.0001s. Once the pipe between the tank and room starts to fill, continuity residual skyrockets with even 0.0005s steps.

Problem:
I realize this is a complex geometry with a large transient, and it gets worse: I'd like to go out to about an hour of problem time. It feels like using a micrometer to measure a building.
Any suggestions that might improve the mesh, or better choices of solver settings?