-
Notifications
You must be signed in to change notification settings - Fork 12
ARCHEReCSEJanuaryMinutes
Attending: Mayeul, Timm, Derek, Miguel.
- Completed Tasks 1-7. Task 8 is mostly complete and tested, lacking only integration into the main Heme loop.
- During the kickoff meeting it was decided to carry out task 9 after task 3. However, we have revisited that decision and we will carry it out as originally planned
- The use of Guo kernel requires adding a correction to the WSS computation. This correction is not currently present.
-
Task 5:
-
Various alternatives for inlet/outlet implementation were discussed. The issues to be overcome include:
-
The fact that we won't be able to use periodic boundary conditions as in Timm's code. This requires that particles are created at the inlet (at a rate given by the target volume fraction) and removed at the outlet.
-
The interaction between particles and fluid dynamics. It is no longer possible to assume that pressure is constant at any given plane orthogonal to the vessel centreline. This includes pressure inlets/outlets.
-
The proposed approach is based on:
-
For a given network geometry, add flow extensions to the inlets with VMTK. For any inlet, we work with two planes: the physical inlet (plane defining the inlet in the original geometry) and the numerical inlet (plane defining the inlet of the flow extension). The same goes for outlets.
-
Apply standard LB boundary conditions at the numerical inlets, since there's no particle flow interaction by construction.
-
Use the region of the flow to extension to:
-
(Inlets) Introduce spheres that get advected at the same time they grow. They must be fully grown by the time they reach the physical inlet.
-
(Outlets) Fade out particles such that they no longer exist at the numerical outlet. The fade-out could be implemented by ramping down the IBM spreading weight with travelled distance between the physical and the numerical outlet. Finally, once the weights reach zero, the particle is simply deleted.
-
Task 6/7:
-
We will go for a master-slave approach for RBC ownership, i.e. each RBC will be uniquely owned.
-
The parallel algorithm to be implemented is:
-
Interpolate (IBM) (This step will need a flow halo or a fancier strategy)
-
Collect nodes.
-
Update position.
-
Migrate cells.
-
Compute forces. (Timm's code starts with this step)
-
Distribute nodes
-
Compute particle-particle interaction.
-
Spread forces (IBM).
-
Solve LBM
-
Iterate.
-
Task 8:
-
Advanced particle initialisation methods will be based on growing particles (from an starting spherical configuration) using models of particle-particle and particle-wall interaction.
-
In parallel, given a target particle density, we will decide how many particles are required in each computational subdomain and solve it independently. At this stage we won't allow particles to cross interprocessor boundaries or walls.
- We won't submit an abstract to EASC15 since we don't foresee having results available in time for the meeting in April.
- We will submit a poster to ICCB15 (deadline 31 January). This will complement Timm's talk in the conference.
- Discuss journal papers in the next meeting:
- First paper describing implementation and technical details.
- More applied paper on RBC distribution in complex networks (in 2016?)
- Mayeul to look into options available (courses, workshops, photo competition, etc.) and sign up.
- Ideas:
- Oxygen delivery.
- Distribution of RBCs in complex networks and haemodynamics:
- Haematocrit distribution.
- WSS
- Basic understanding of design principles of microvasculature. What's it being optimised for?
- Link with angiogenesis in the context of cancer.
- In Edinburgh.
- Either w/c 4 May or 11 May depending on Miguel's !London/Edinburgh schedule.
- Add WSS correction to Guo kernel (MA).
- Check whether HemeLB keeps a flow halo in each computational subdomain (MB).
- Look into technical staff engagement with the ARCHER community (MA).
- Set date for next meeting (MB).