3d visualization in fulcrum csas-shift sequences

ORNL is managed by UT-Battelle, LLC for the US Department of Energy

3D Visualization in FulcrumCSAS-Shift Sequences

B.J. Marshall and Rob Lefebvre

SCALE Users’ Group WorkshopORNLAugust 19, 2019

2

Outline

• 3D visualization in Fulcrum

• Brief intro to Shift

• Integration of Shift into SCALE

• CSAS and CSAS-Shift examples

3

3D Visualization in Fulcrum

• Why is 3D visualization in Fulcrum part of a CSAS-Shift tutorial?– The ray tracing algorithms used in Fulcrum for 3D visualization are the

same as those used in Shift.– This reuse of components is a hallmark of the modernized approach to

SCALE development.

• The first release of 3D visualization in Fulcrum will provide most of the same features as KENO3D.

• Additional capabilities are included such as opacity control and undoing cuts.

4


• Open 2_rocky_flats.inp

5


• Click view – Or the view drop down → View geometry– Or Ctrl+Shift+V

• Launches viewer in 2D, like Fulcrum in SCALE 6.2

6


• Select 3D from view orientation drop down

• Click controls button from top bar

7

3D Visualization in Fulcrum – Materials Controls

• Materials, camera, and cuts tabs are active and useful

• Opacity control on materials tab is a new feature!

• Mixture toggle (checkbox) is a new feature!

8


• Single click in any column will generate a description of the mixture in the lower pane of the panel– Single click directly on “Show” checkbox will toggle mixture without

changing focus to toggled mixture

• Double click will allow modification of color or opacity– Color options vary based on platform, as with 2D viewer in SCALE 6.2– Opacity selected as percentage where 100% is totally opaque

• Image on next slide has void (mixture 0) turned off and concrete (mixture 3) opacity set to 33%

9


10

3D Visualization in Fulcrum – Camera Controls

• Camera controls are probably most different from KENO3D

• The view is controlled by positioning and orienting the camera relative to the “scene”

• Position block controls where the image is in the window– Arrows move image in those directions

• Orientation block controls rotation of image– Arrows rotate image in those directions

11

3D Visualization in Fulcrum – Camera Controls

• Presets provide the same 4 preset scenes that KENO3D provides

• Miscellaneous pane:– Orthographic check box disables perspective

and makes edges straight and right angles 90°– Zoom factor zooms in and out and is also

available on the top bar of visualization window– Outline threshold is tolerance for determining if

something is a boundary or not

• Ray tracing happens again after each change, no live ray tracing

12

3D Visualization in Fulcrum – Cuts Controls

• Central portion of the pane presents the controls to add a new cut– Can name cuts so you know what each cut

does– Block and pie slice options are available just

as in KENO3D– Cut planes can be moved with the offset

controls provided– Rotation of cuts controlled in the same way

as rotation in KENO-VI– Click “Add” once the cut is set up

13


• Set up a cut to remove front right quarter of Rocky Flats model– Cut name: Front right quarter– Click add

• Image changes (see next slide)

• Cut added to list of cuts in top pane of control panel (at right)

14


15


• Cuts can be removed in Fulcrum – NEW FEATURE!

• Simply click on cut in list and then on Remove button

End of 3D visualization slides. Any questions? Let’s take a few minutes and just try some 3D visualization on your own…

1717

Shift overview – Exnihilo code system

• Flexible, high-performance Monte Carlo radiation transport framework

• Shift is physics agnostic– SCALE CE physics – SCALE MG physics

• Shift is geometry agnostic– SCALE geometry (K5 and K6)– Exnihilo RTK geometry– MCNP geometry– DagMC-CUBIT CAD geometry

• Fixed-source and eigenvalue solvers

• Integrated with Denovo for hybrid methods

• Multiple parallel decompositions and concurrency models

• Shift is designed to scale from supercomputers to laptops

Slide courtesy of Greg Davidson and Tom Evans

1818

Status Updates

• Added missing capabilities in FY19– Improved Geometria/HYAS to support most k5/k6 geometry

features– Added capability to calculate volumes• k6-geometry, stochastic volume calculation with ray-tracer• k5-geometry, analytic volume calculation

– Designed CSAS output builder and extended the output edits,– Added flux and fission density tallies

geometry improvements

volume calculations

output edits

flux and fission densities

FY19

FundamentalCSAS featuresimplemented

Previous FYs

1919

Development Plans

• Focus areas for FY19 – Geometry improvements

• FY16 100% of VALID ran, regression to FY18 and only 82% ran• Currently 93% of VALID runs

– Add automatic volume calculation capability for the new random geometry– Enable calculation of real variance of k-eff by generations in Shift– Complete the missing output edits– Enable Shift/Sourcerer method in CSAS-Shift

• Focus areas for FY20 (upon available funding)– Reaction tallies,– Tally fission neutrons,– Enable start types available in Shift

20

Strategy for Shift integration into SCALE

• The goal is to be as transparent to the user as possible

• Ultimately, the only input change that should be required is changing “=csas5” or “=csas6” to “=csas5-shift” or “=csas6-shift”

• The CSAS input is processed and translated for use in Shift

• The output is received and processed to be similar to CSAS output– Key outputs are being formatted into similar or, where possible, identical

edits to minimize impact on post-processing scripts

• Shift integration funded by both NRC and NCSP

So much for background – let’s run some jobs!

22

CSAS5 & CSAS5-Shift calculations

• We’ll start with the Rocky Flats problem from KENO training, run it in CSAS5 with KENO, and then run it in CSAS5-Shift with Shift

• The input should be open from the 3D visualization practice

• To run the input, click into the input and click the Run button on the top bar

– Messages panel reports progress of job

23

Examine output

• Right click on 2_rocky_flats.inp in navigation panel, then select “Open associated files” and then 2_rocky_flats.out

• Entire output file now open for review

• Best estimate k-eff ~1.0022 ± 0.0021

24

CSAS5-Shift

• Convert CSAS5 to CSAS5-Shift– Click immediately after =csas5– Type “-shift”

• Save file with a new name– File → Save as…

• Run the new CSAS5-Shift input

25

CSAS5-Shift Output

• As before, right click in navigation panel and “Open associated files”

• Look through text output from CSAS5-Shift

• Best estimate k-eff ~1.001114 ± 0.004273– Statistically equivalent to CSAS5 (KENO) result

• Comments, questions, thoughts before we move on to another model?

26

CSAS6 and CSAS6-Shift Calculations

• Similar process as with Rocky Flats, but with 4_castor_cask.inp

• Open in Fulcrum, select appropriate processor and run– KENO k-eff ~0.6274 ± 0.0019

• Modify sequence to CSAS6-shift, save as new file, select appropriate processor and run– Shift k-eff ~0.621619 ± 0.002519

• Results within 2 standard deviations of each other so statistically equivalent

• Questions before we discuss parallel calculations?

27

Parallel Monte Carlo calculations

• Many factors effect parallel performance– Physics: MG and CE have different performance– Tallies: Cost per transport step scales linearly with number of tallies

• Reaction rate tallies can be particularly expensive, especially in CE– Geometry: Geometric complexity can be a factor for extremely

complex models– Problem type: Eigenvalue problems are less efficient than fixed-source

problems because of fission rebalance and k-eff estimates– Hardware: Speed and performance of interconnect within a node and

between nodes is a huge factor in parallel performance

• Larger batch sizes are more efficient in parallel

28

Parallel Monte Carlo calculations

• Generally Shift parallelism is better than KENO, and should generally scale better, especially on large numbers of cores for large problems

• Many k-eff problems will not be large enough to see great parallel efficiencies

• Table on the next slide provides some results based on castor cask problem– These results are only for a single problem and not necessarily

indicative of how KENO or Shift will perform for any specific problem

29

Sample parallel results for castor cask model

NPG 100,000 1,000,000Serial Parallel (16 cores) Parallel

KENO Shift KENO Shift KENO Shiftkeff 0.625659 0.625621 0.625659 0.625461 0.625680 0.625345σ 0.000099 0.000100 0.000099 0.000100 0.000080 0.000117

Run time (s) 51162.7 70287.3 5345.63 6980.44 10237.4 7396.34FOM (1/σ2T) 1994 1423 1193 895 954 617Core*time/

particle (ms) 0.759 0.662 1.269 1.134 1.085 0.784

Reported efficiency N/A N/A 79.31% N/A 87.56% N/A

30

Older results from ORNL/SR-2016/401

That’s it… Any questions?

3d visualization in fulcrum csas-shift sequences

Documents