All entries for March 2007
March 30, 2007
Yesterday I ran rs01-02-07 without wall functions to see if it would replicate rs01-02-05. It did. On thinking more about it, I needed to use the wall distance, ‘wy’ for the ufric estimate rather than dy. Also I think I needed to set as(i,j,k)=0 otherwise this will double account the velocity within the wall. However, I think this may need more thought….
I will run it as rs01-02-07a and think about the implementation.
- Run rs01-02-07a
- Think about the implementation
- Finish reading the Wu & Squires paper
- Print out the review by Simpson of turbulent separation
March 28, 2007
Milla & I have come up with a simple but rather addictive kickstart drink. It goes…
1 part vodka, 2 parts cranberry juice, 2 parts pineapple juice, 2 parts lemonade
and is highly addictive!
rs01-02-07 still appears to be stable although the recirculation development is different to the other sims (although stable). Will just need to leave that running for now.
In the meantime, I can get on with the MT03-01 simulations in preparation for Juice’s work.
I also need to assess the full Reynolds number grid requirements before the meeting on MondayTask-list:
- Calculate the grid requirements for full and 0.1 Re simulations
- Look at MT03-01 simulations
Have set running mt03-01-18 again which uses 50% fluctuations. Once this is established, I will run 19 & 20 with 75% & 100% fluctuations resp. (As detailed in the recent paper on LES BF Step)
Will now get on with working out grid requirements for LES runs for full and 0.1 Re in preparation for Monday’s meeting.
For the full Re case, Bernard (2003) gives the friction velocity at the apex of the bump (I.e. the point of maximum wall shear) to be of the order of 0.535m/s. This would equate to a wall unit size of around 2.719×10-6m or 27 microns.
As LES requires the following:
dx+ < 100, dy+ < 1, dz+ < 20
this gives the following grid limits:
dx < 2.7mm, dy < 27 microns, dz < 0.54mm
Assuming that these grid spacings will be used throughout the domain (which is true as a starting point and we will try to compress the grid later to save resources):
nx = 10m/2.7mm = 3700
ny = 1m/27microns = 15000
nz = 0.5m/0.54mm = 925
This equates to a grid mesh of 5.134×1010 (that is 51,340 million nodes!) Clearly, even with some grid expansion, it would be almost impossible to bring the requirements down to the region of 1-10 million nodes.
Using semi-empirical theory, the wall shear stress at the inlet is likely to be around 0.002 Pa which will probably increase by a factor of 2.5 near the bump apex to around 0.0057 Pa. This equates to a friction velocity of 0.068m/s thus giving a wall unit as 0.2mm.
Using the same grid requirements as above, we can estimate the grid spacings to be:
dx < 20mm, dy < 0.2mm, dz < 4mm
This gives the number of grid nodes as:
nx = 10m/20mm = 500
ny = 1m/0.2mm = 2000
nz = 0.5m/4mm = 125
Thus giving a total mesh size of 125 million. Still big but much closer to feasible.
Savings can be found by relaxing the dy+ requirement to <3 rather than <1. This will reduce the mesh count by a factor of 3 to 41.6 million.
Further savings can be found by reducing the length of the domain to 8m so reducing the mesh count by 1.2 to 34.7 million.
By also reducing the domain in z from 0.5m to 0.3m (slightly larger than the boundary layer thickness) this reduces the mesh count by 1.67 to 20.7 million.
This is still too large but by assuming that we can relax the wall shear stress estimates at the apex to be equal to those at the inlet, this gives a mesh of 300×1000 x 75 so a count of 22.5 million. Applying the above savings (which add up to a factor of 6, we get a mesh count of around 3.5-4 million.
Using a mesh of 4.5 million for extra nodes gives an estimated run time of 180 days (6 months) without the use of 2D preliminary start-up simulation.
March 27, 2007
Milla & I are celebrating our engagement and 24th/26th birthdays this weekend on Saturday 31st March at the Cryfield Sports Pavilion. Pictures will be posted up here next week!
The plan for today is to finish reading that paper on LES over a BFS (hopefully by 1000) the get into the office and get on with setting up the combinations of methods in 2D. Once this is done I can try and set up a 3D simulation using some of the 2D data.
Finished reading that paper so can get on with simulations…
Had an idea. Although running a full LES sim of the bump+jet is out of the question, what about a bump alone simulation?
- Estimate grid requirements for bump using LES
- Set-up ‘combinations’ of methods
- Check WF + IBM simulation (rs01-02-07) is working ok
Simulation rs01-02-07 does not appear to be having any wall function application going on. Will need to rethink how the IBM interacts with the WF approach.
Previously, the WF approach would only be applied if both the nblv points above AND below were fluid but this would negate the WF approach throughout. Have changed this statement to above OR below.
Right, I think I can get an LES simulation of the full bump working for 4.5M nodes. This will take approx. 6 months to run assuming I DON’T use a preliminary 2D simulation for initial conditions. After lunch, I will carry on with my assessment then email Peter & Yongmann the results.
Had a long lunch as I needed to service the car and go shopping but am back in the office and am going to sort my space out. I’m going to keep work stuff here and try to work from here much more.
Sorted books & papers out. Can get on with work now.
- Email Lille
- Check rs01-02-07
- Read paper from Juice
- Decide on combinations (although I think rs01-02-07 is all that will be necessary)
- Look at setting up a 3D BL profile (MT03-01) but with very high levels of disturbance. (around 100%)
With regards to rs01-02-07, I changed the dimension line at the beginning of coeffu3 from nblu(0:500,0:500) to nblu(0:id,0:jd). This has changed the WF application points so we shall see!
March 16, 2007
After a few days of debugging the wall function method I actually have some results to show for it!
The plot below shows the base simulation. Note that on the run up the profile is turbulent but following the acceleration and separation, the profile re-laminarises.
Adding in the wall function approach allows the boundary layer to reattach whilst retaining its turbulence as shown below.
The following set of (2D) simulations are being run:
rs01-02-01: Base simulation. 2D RANS, No IBM, No WF, No TI DONE
rs01-02-04: Wall Functions. 2D RANS, No IBM, With WF, No TI
rs01-02-05: Immersed Boundary Method. 2D RANS, With IBM, No WF, No TI
rs01-02-06: Detached Eddy Simulation. 2D DES, No IBM, No WF, No TI
These should take the weekend to run so I’ll have some results to compare on Monday then I can set the 3D simulations running. (Based on the run times of these 2D sims)
Just realized that rs01-02-04 was running using a new simulation each time so hasn’t really moved on any further. Will need to re-run.
March 14, 2007
After three days of hefty debugging, have finally managed to get the wall function approach working for the bump. Apparently I was reducing the source term for an increase in wall shear stress rather than increasing it. It seems to work much better (and without any capping or dodgy fixes to get it working!)
Right, tomorrow, I will sort out my results and start writing into Chapter 7.
March 12, 2007
I started watching this thinking it was going to be the classic ‘disaster’ style space movie of late (I.e. Deep Impact, Armageddon) and the first hour seemed to be such. Especially unneccesary was the scene in which three of the astronauts were killed tampering with the ‘face’. A rock to the face and being torn to pieces by being spun in the centre of a vortex all seemed a bit far fetched but if this was a ‘disaster’ movie then that’s what you expect.
However, the final 45 mins seemed to switch to a much more realistic exploration of the martian surface. The logic behind the rest of the story (no spoilers!) seemed plausible but due to the initial over-the-top drama, it didn’t sit well.
In all, a moderate film but little is learned about Mars and it is strangely inconsistent.
Right, the goals are as follows:
1) 2D simulation using only RANS, no IBM, no WF and no TI
2) Add in WF
3) Add in IBM
4) Add in 2D DES
5) 3D simulation using only RANS, no IBM, no WF and no TI
6) Add in WF
7) Add in IBM
8) Add in 3D DES
9) Add in TI
So, once I have done this, we have the bump set-up without any flow control. We are at stage 1 with 2 underway. 3 & 4 should not present any problems as they are already done. Likewise, 5 should be simple after 1-4 and 6-8 should have been sorted from 2-4. I will need to work on 9 in the background from JJs work (which should be debugged anytime now)
Therefore, at present, I need to get 2 working. This will probably be by using old results for the no WF simulation in order to make more stable. But I will try some other methods whilst I wait for 1 to finish running.
rs01-02-01: As per objective 1 above
rs01-02-02: As per objective 2 above
March 05, 2007
Am sorting out simulations on my CS disk. For RS01, I will run A04 again to get a steadier mean.