Enhancing Beamline Shielding Analysis with ADVANTG 3

1. Evaluation of SNS Beamline Shielding Configurations Using MCNPX Accelerated by ADVANTG Joel M. Risner Seth R. Johnson Igor Remec Kursat B. Bekar RNSD / Radiation TransportOak Ridge National LaboratoryOak Ridge, TN, USA

2. Background and Objectives • Beamline shielding analyses for SNS are performed using MCNPX. The typical development of variance reduction parameters for SNS MCNPX models is a challenging and time consuming process, and often increases modeling complexity due to the use of cell-based geometry splitting and rouletting. The cell-based splitting approach, which must be ‘hand tuned’ for each model variation, has the additional limitation of lacking energy-dependent variance reduction. • ORNL has pioneered the development of hybrid deterministic/stochastic transport methods, which utilize ‘moderate fidelity’ discrete ordinates transport calculations to generate variance reduction parameters (weight windows and consistently biased sources) that can significantly accelerate Monte Carlo solutions. We have incorporated this hybrid methodology into the SCALE MAVRIC sequence (which employs the Monaco neutron/photon Monte Carlo code) and into ADVANTG, which generates weight windows and biased sources that can be used with MCNP and MCNPX.

3. Background and Objectives • MAVRIC and ADVANTG each implement the Consistent Adjoint Driven Importance Sampling (CADIS) and Forward-Weighted CADIS (FW-CADIS) methods. CADIS is designed for single-tally applications, while FW-CADIS can be effectively applied to multiple tallies and mesh tallies. • We previously used a developmental version of ADVANTG (referred to as ADVANTG 2) to generate weight windows and biased sources for shielding analyses of SNS Beamline 1B, which contains the Nanoscale-Ordered Materials Diffractometer (NOMAD) experimental arrangement. Our primary goals were to 1) obtain well-converged ‘nearly global’ solutions, 2) reduce the run time required for the MCNPX calculations (which was typically several thousand CPU hours), and 3) reduce the time-intensive process of developing variance reduction parameters by hand.

4. Background and Objectives • ORNL has subsequently developed an enhanced version of ADVANTG (referred to as ADVANTG 3) which is significantly more straightforward to use. ADVANTG 3 is integrated with a parallel version of the Denovo discrete ordinates code, which eliminates the need to perform a multi-step process in which Denovo inputs were created and then separately used with Python codes to ‘drive’ the forward and adjoint deterministic calculations. ADVANTG 3 also has an option to collapse weight windows in space and energy. • A directional- and energy-dependent plane source option has been added to better model the neutron source at the upstream part of the beamline, close to the proton beam target.

5. SNS Beamline Shielding Calculations Using MCNPX • Previous Methodology: • Assign particle splitting parameters ‘by hand’ for variance reduction. • Splitting parameters are very problem dependent – changes in either the actual model or the desired analysis output can require a different set of parameters. • The additional cells required for geometry splitting make the model more complex and can adversely affect tracking speed. Hybrid Methodology: • Generate weight windows for variance reduction using the FW-CADIS method in ADVANTG. • Weight windows can be generated to provide well-converged results over large (nearly global) extents of the model. • Because the weight windows are independent of the geometry surfaces and cells, no additional geometric complexity is needed. This simplifies the model and can further improve calculational speed.

6. Sketch of NOMAD Beamline (BL-1B)

7. Vertical Cut Through the Centerline of the MCNPX Model of BL-1B (NOMAD) Secondary Shutter/ Optics Carousel Sample Access Room T0 Chopper Beam Stop Detector Vessel Bandwidth Choppers Example of Geometric Complexity Used for Cell-Based Splitting and Rouletting

8. Horizontal Cut Through the Midplane of the MCNPX Model of BL-1B (NOMAD) Secondary Shutter/ Optics Carousel Detector Vessel NOMAD Beamline (BL-1B) T0 Chopper Beam Stop Bandwidth Choppers USANS Beamline (BL-1A)

9. Unique Beamline Shielding Challenges • While the CADIS and FW-CADIS methods have been applied to a fairly wide range of shielding applications, beamline shielding poses challenges that set it apart from ‘typical’ shielding scenarios: • High-energy neutrons (up to 300 MeV in this study) • A highly forward-peaked plane source at the beamline opening (10 x 12 cm) with a distance of nearly 30 meters to the beam stop • A combination of deep penetration shielding and a beamline which can be completely open or can be partially blocked by neutron choppers or a secondary shutter.

10. FW-CADIS Calculations • SNS MCNPX model is approximately 9 x 17 x 30 meters • ADVANTG creates a Denovo input by meshing the MCNPX model. ADVANTG then applies the FW-CADIS methodology to run a forward Denovo calculation, create adjoint sources, run an adjoint Denovo calculation, and construct weight windows and a biased source for MCNPX. • Denovo calculations • ~ 3M spatial cells • 44 groups (33 neutron, 11 gamma) (collapsed from HILO2K library) • P3 scattering • Lobatto quadrature • 86 CPU hours forward, 83 CPU hours adjoint (open beamline model) • Weight-window file is approximately 1.8 GB

11. The FW-CADIS Method Illustrated 1: Construct Denovo model 2: Solve forward Denovo problem 3: Construct importance source 4: Solve adjointDenovo problem 5: Construct weight windows

12. MCNPX Results – Hybrid Solution Sample Access Room Results for these three planes are compared to the original geometry splitting approachon the following slides. Detector Vessel Red Contour Line is 0.25 mrem/hr

13. First Set of Comparisons:Open Beamline, Full SourceHigh-Energy and Low-Energy Components(Accident Condition)

14. Open beamline, full sourceHorizontal tally at the centerline of the beam line, vicinity of detector vessel NOTE: the “region of interest” for the original MCNPX runs is outlined in blue. Geometry Splitting Total Dose Rate (mrem/hr)(red contour line at 0.25 mrem/hr) Hybrid Total Dose Rate (mrem/hr)(red contour line at 0.25 mrem/hr)

15. Open beamline, full sourceHorizontal tally at the centerline of the beam line, vicinity of detector vessel NOTE: the “region of interest” for the original MCNPX runs is outlined in blue. Hybrid: ~1000 CPU Hours Relative Error (1s) Geometry Splitting: ~5000 CPU Hours Relative Error (1s)

16. Open beamline, full sourceVertical tally at the centerline of the beam line, vicinity of detector vessel Hybrid Total Dose Rate (mrem/hr)(red contour line at 0.25 mrem/hr) Geometry Splitting Total Dose Rate (mrem/hr)(red contour line at 0.25 mrem/hr) NOTE: the “region of interest” for the original MCNPX runs is outlined in blue.

17. Open beamline, full sourceVertical tally at the centerline of the beam line, vicinity of detector vessel NOTE: the “region of interest” for the original MCNPX runs is outlined in blue. Hybrid: About 1000 CPU Hours Relative Error (1s) Geometry Splitting: About 5000 CPU Hours Relative Error (1s)

18. Open beamline, full sourceVertical tally through the sample position, perpendicular to the beam line Hybrid Total Dose Rate (mrem/hr)(red contour line at 0.25 mrem/hr) Geometry Splitting Total Dose Rate (mrem/hr)(red contour line at 0.25 mrem/hr) NOTE: the “region of interest” for the original MCNPX runs is outlined in blue.

19. Open beamline, full sourceVertical tally through the sample position, perpendicular to the beam line NOTE: the “region of interest” for the original MCNPX runs is outlined in blue. Geometry Splitting: About 5000 CPU Hours Relative Error (1s) (red contour line at 0.25 mrem/hr) Hybrid: About 1000 CPU Hours Relative Error (1s) (red contour line at 0.25 mrem/hr)

20. Second Set of Comparisons: Open Beamline, Low-Energy Source –Predominantly 0.1 - ~3 Angstroms(Normal Operation – Applies to Locations Downstream of the T0 Chopper)

21. Open beamline, low-energy sourceHorizontal tally at the centerline of the beam line, vicinity of detector vessel NOTE: the “region of interest” for the original MCNPX runs is outlined in blue. Geometry Splitting Total Dose Rate (mrem/hr)(red contour line at 0.25 mrem/hr) Hybrid (ADVANTG 2) Total Dose Rate (mrem/hr)(red contour line at 0.25 mrem/hr)

22. Open beamline, low-energy sourceHorizontal tally at the centerline of the beam line, vicinity of detector vessel NOTE: the “region of interest” for the original MCNPX runs is outlined in blue. Hybrid (ADVANTG 2): About 480 CPU Hours Relative Error (1s) (red contour line at 0.25 mrem/hr) Geometry Splitting: About 1300 CPU Hours Relative Error (1s) (red contour line at 0.25 mrem/hr)

23. Open beamline, low-energy sourceVertical tally through the sample position, perpendicular to the beam line Hybrid (ADVANTG 2) Total Dose Rate (mrem/hr)(red contour line at 0.25 mrem/hr) Geometry Splitting Total Dose Rate (mrem/hr)(red contour line at 0.25 mrem/hr) NOTE: the “region of interest” for the original MCNPX runs is outlined in blue.

24. Open beamline, low-energy sourceVertical tally through the sample position, perpendicular to the beam line NOTE: the “region of interest” for the original MCNPX runs is outlined in blue. Hybrid (ADVANTG 2): About 480 CPU Hours Relative Error (1s) (red contour line at 0.25 mrem/hr) Geometry Splitting: About 1300 CPU Hours Relative Error (1s) (red contour line at 0.25 mrem/hr)

25. Carousel Shutter Closed • For the original calculations a model with no USANS beamline was used to evaluate radiation levels for a configuration in which the NOMAD secondary shutter (the carousel) is closed. That model required geometric changes to support the use of cell-splitting parameters that are specific to this configuration. • For the hybrid analysis we simply rotated the carousel shutter and repeated the ADVANTG/MCNPX calculational sequence.

26. Carousel Shutter Closed – with Geometric Splitting Note the changes in the geometry that are driven by cell-splitting regions. The corresponding importance values must also be determined by the analyst.

27. Carousel Shutter Closed – with Hybrid Method For the hybrid analysis, we simply rotated the carousel and generated new variance reduction parameters with ADVANTG.

28. Carousel closed, full sourceHorizontal tally at the centerline of the beam line Hybrid Relative Error (1s) (red contour line at 0.25 mrem/hr) Hybrid Total Dose Rate (mrem/hr)(red contour line at 0.25 mrem/hr)

29. Carousel closed, full sourceVertical tally at the centerline of the beam line Hybrid Total Dose Rate (mrem/hr)(red contour line at 0.25 mrem/hr)

30. Carousel closed, full sourceVertical tally at the centerline of the beam line Hybrid Relative Error (1s) (red contour line at 0.25 mrem/hr)

31. Carousel closed, full sourceHorizontal tally at 180 cm (top of shielding blocks) Hybrid Total Dose Rate (mrem/hr)(red contour line at 0.25 mrem/hr) Hybrid Relative Error (1s) (red contour line at 0.25 mrem/hr)

32. A Few Observations … • The convergence of the ADVANTG-accelerated MCNPX calculations is somewhat poor in the beam stop area. This is probably due to target weight variations that cause rouletting followed by splitting along the beamline just prior to the beam stop. A separate analysis of the beam stop (rather than an essentially global approach) may be a better approach for this region. • The selection of angular quadrature for this analysis was crucial. We applied a Lobatto quadrature with 16 ‘polar’ coordinates. This set provides an ordinate along the beamline axis. If a more typical level symmetric quadrature set is used, the target weights change significantly along the proximal end of the beamline, which leads to oversplitting. One could re-normalize the weight windows, but there is no single value that could be applied that is appropriate for all energies.

33. Note the increase in target weights preceding the beam stop. The beam stop poly begins at ~2690 cm.

35. Conclusions • While we achieved substantial reductions in MCNPX run times, the benefits to using ADVANTG go well beyond those run time savings. • The use of ADVANTG resulted in ‘nearly global’ Monte Carlo solutions for this complex model. These solutions were obtained without the need to develop variance reduction parameters by hand, saving substantial amounts of analyst time as well as achieving well-converged solutions in substantially less computer time. • The use of ADVANTG 3, which has a substantially improved user interface, made this analysis much more straightforward than our original ADVANTG 2 study. We also applied a new directional plane source option and did limited testing of weight window collapsing.

36. Supplemental Slides

37. MCNPX Results Sample Access Room Detector Vessel Red Contour Line is 0.25 mrem/hr

38. MCNPX Results Sample Access Room Detector Vessel Red Contour Line is 0.25 mrem/hr

39. Horizontal Cut Through the Centerline of the MCNPX Model of BL-1B (NOMAD) – Detector Vessel Vicinity Beam Stop Detector Vessel Sample Position USANS Beamline (BL-1A) Secondary Shutter/ Optics Carousel T0 Chopper NOMAD Beamline (BL-1B)

40. Vertical Cut Through the Centerline of the MCNPX Model of BL-1B (NOMAD) – Detector Vessel Vicinity Secondary Shutter/ Optics Carousel Sample Access Room T0 Chopper Detector Vessel Sample Position

41. Vertical Cut Through the Detector Vessel at the Sample Position – Perpendicular to the Beamline Sample Access Room USANS Beamline (BL-1A) Detector Vessel Sample Position

42. Example mesh tally relative error distributions: geometry splitting caseOpen beamline HE source; horizontal tally at the beamline center

43. Example mesh tally relative error distributions: hybrid caseOpen beamline HE source; horizontal tally at the beamline center