Computation in nuclear engineering

[hiraku0n]

To what extent is computation used in nuclear reactor design? For example in aerospace CFD and FEM softwares are frequently used in the design process; the days of wind tunnel testing is limited to its absolute minimum. What are major softwares used in NE? Are there softwares such as NX, that combines various analysis into one (CAE) in the nuclear industry?

 

[vanesch]

In all generality, I'd say that computation (and validation of that computation) is probably just as much used in nuclear reactor design than in airospace design, simply because of the expense and the risk of error. You cannot just build some prototypes and see if they behave more or less correctly, and then tune by trial and error. It is far cheaper to find a design malfunction in simulation than in prototyping.

As to software, there's usually a combination of a radiation transportation program and a hydrodynamics program. I don't know much about these combinations. The program I use regularly (but not for reactor design, which is not at all my job) is a French monte-carlo called Tripoli (somewhat equivalent to MCNP). It is only concerned with the radiation transport and nuclear reactions (not with heating and hydrodynamics and so on, nor with changes in material composition).

 

[Xnn]

Of course codes are used, but eventually the design has to be put into service.

So, fuel designs tend to be incremental, but have come a long way in 50 years.
Lead test assemblies are used to some degree, but then when the first batch of a new design is loaded, it's a bigger question as to how well it behaves.
That's why there are Engineers involved with both operation and design.

Every Fuel vendor has their own codes; some are better than others.
They are fairly old and the NRC does have concerns about that.
Designs typically use 9-10% margin to thermal limits. Actual operation is closer.

GE uses Panacea. Not sure about Westinghouse; it may be called Powerplex.
Areva is French and who knows about Hitachi the Swedes & Russians.

 

[Astronuc]

GE uses TGBLA (lattice code) with PANACEA (core simulator). GE was working on a transport-theory based code, and may have resumed development.
Westinghouse uses PHOENIX (lattice code) with ANC and more advanced SP-NOVA (core simulator).
AREVA uses CASMO (lattice code) with MICROBURN-B/P (core simulator).

BNFL/British Energy have used WIMS / PANTHER

I believe Swedes (Nordic countries) used PHOENIX/POLCA-7 or CASMO-4/POLCA-7 for BWRs, but perhaps use CASMO/SIMULATE or vendor codes corresponding to their fuel supply (or both). Some utilities use vendor methods or independent methods, to either design the cores or do an independent over-check.

Studsvik-Scandpower produces CASMO (lattice code) and SIMULATE (core simulator).

EPRI has developed the ARROTA core simulator, and they sponsor VIPRE & RETRAN T/H code.

There are universitiy codes like Purdue's PARCS and NC State's (Turinsky's) FORMOSA core simulators.


Traditionally, thermal-hydraulics codes are decoupled from core simulator codes.

ANL is working on a Numerical Nuclear Reactor that attempts to marry Core Simulation (Nuclear methods) with a T/H capability.

 

[gmax137]

the structural design uses FEM codes like ANSYS. CFD codes are slowly becoming more popular for use in complicated thermal-hydraulics. reactor vendors use their own codes for thermal hydraulics within the core (channel & fuel pin codes). For a long time, the industry was kind of static in the methods used (because most of the work was for existing customers/plants, and they resist the need to develop new methods). This is beginning to change.

 

[hiraku0n]

Thank you all.

So it appears there are various codes. I would think they would be tested against known data but has there been studies comparing it against each other? How much computing power would be necessary in running these code?
Are there any open source development in this area?

 

[Xnn]

Actual response of cores are checked against predictions in a required test called Reactivity Anomalies; which is just 1 of many aspects of a core design of interest to operators. Acceptance criteria is typically +- 1% Dk/k. There has been at least 1 instance where this was not met. Believe it involved changing fuel vendors/codes and a whole lot of finger pointing.

Also, some limited work has been done to compare gamma scans of bundles against predictions; but that is more costly and not always required. NRC is pushing for this
but it is debateable if there is a need to go beyond that.

Computing power is not much of an issue as modern computers have more than enough power to run most codes in a reasonable time.

Not aware of any open source development that is worth anything. Vendors like to keep valuable information propietary.

 

[Astronuc]

I would think they would be tested against known data but has there been studies comparing it against each other? How much computing power would be necessary in running these code?

Are there any open source development in this area?

There have been limited comparisons, e.g. between a vendor method and CASMO/SIMULATE. The comparisons were done by the vendor and utility and are not available in the public domain. There have been attemps at gamma-scanning, but results have been problematic.

One core simulation (depletion case) can be run on a single workstation, however the set up is complicated since one needs to go backwards as far as possible with respect to the reinsert fuel. Cores simulations can be run in octant, quadrant or full core depending on the assymetries being modeled. A typical 4-loop PWR (193 assy) running 17x17 fuel contains 50952 fuel rods, and they are typical nodal with 24 x 6'' (15.24 cm) axial nodes, or greater numbers if axial zoning of burnable absorbers are used with blankets of more than 6''.

Some codes are available through IAEA or universities, but usually one must register and pay a fee. Access to source code is restricted.

Application of CFD in the nuclear industry is relatively recent, and the two main codes are CFX and Star-CD. AFAIK, no one has integrated numerical reactor simulators with a CFD code. The core simulators basically contain relatively simple thermal hydraulic models which provide coolant temperature, density and void coefficients. Depletion calculations presume steady-state, although there have been some attempts to look in more detail at downpowers and power ascensions (and associated Xe-swings), control blade history effects, and various transients, e.g. rod-drop or rod-ejection accident.

 

[hiraku0n]

What PDE(s) would these codes be solving? Neutron Transport? Multigroup diffusion equation?

 

[QuantumPion]

What PDE(s) would these codes be solving? Neutron Transport? Multigroup diffusion equation?

Lattice codes (e.g. casmo, phenix) use various approximations to the transport equation with anywhere from 2 to several hundred energy groups. Core simulators like simulate use the diffusion equation, with 2 energy groups.

 

[Xnn]

Some codes are also used to figure out things like void fraction, nodal and pin powers as well as thermal limits like maximum fraction limiting critical power ratio (MFLCPR), maximum fraction limiting power density (MFLPD) and maximum average power density ratio (MAPRAT).

The critical power ratio is a measure of how close any node of a fuel bundle is to transition boiling; an important parameter during transients.

We also make predictions of how the core will respond to changes in flow and control rod positions in order to assure adequate margin for normal operations, anticiptated transient and design base accident conditions at all times.

 

[vanesch]

What PDE(s) would these codes be solving? Neutron Transport? Multigroup diffusion equation?

In fact, I know more about monte carlo solvers for radiation transport. Are these codes not used in the design of power plants ? In other words, does one use mainly what's called deterministic codes ?
I would think that monte carlo codes require less approximations and can implement much more detailed physics (there's no need to do homogenisations for instance: nothing stops you from implementing the actual geometry, at a reasonable price in calculation speed).

Of course, at a certain point, one needs to "bin" the results of a monte-carlo.

 

[QuantumPion]

In fact, I know more about monte carlo solvers for radiation transport. Are these codes not used in the design of power plants ? In other words, does one use mainly what's called deterministic codes ?
I would think that monte carlo codes require less approximations and can implement much more detailed physics (there's no need to do homogenisations for instance: nothing stops you from implementing the actual geometry, at a reasonable price in calculation speed).

Of course, at a certain point, one needs to "bin" the results of a monte-carlo.

Monte-carlo codes are used for certain aspects of power plants, such as radiation shielding, spent fuel pool criticality, etc. But the core design itself uses nodal codes because doing a monte-carlo analysis for the large variety of possible core conditions which are time-dependent would be exceedingly time consuming.

 

[Morbius]

To what extent is computation used in nuclear reactor design? For example in aerospace CFD and FEM softwares are frequently used in the design process; the days of wind tunnel testing is limited to its absolute minimum. What are major softwares used in NE? Are there softwares such as NX, that combines various analysis into one (CAE) in the nuclear industry?

hiraku,

The use of computation in nuclear engineering is EVEN MORE extensively used than in the
aerospace industry. The nuclear industry didn't have the luxury of having something analogous and as
useful as a wind tunnel. For example, all the SAFETY calculations that are done for nuclear reactors
are done on computers.

There are LOTS and LOTS of computer codes used in the nuclear field and RSICC at Oak Ridge
National Laboratory is a central repository for the software:

http://www-rsicc.ornl.gov/

http://rsicc.ornl.gov/rsiccnew/CFDOCS/qryPackage.cfm

Dr. Gregory Greenman
Physicist

 

[Morbius]

In fact, I know more about monte carlo solvers for radiation transport. Are these codes not used in the design of power plants ? In other words, does one use mainly what's called deterministic codes ?
I would think that monte carlo codes require less approximations and can implement much more detailed physics (there's no need to do homogenisations for instance: nothing stops you from implementing the actual geometry, at a reasonable price in calculation speed).

Of course, at a certain point, one needs to "bin" the results of a monte-carlo.

vanesch,

You can't do full core multi-year burnup calculations in a Monte Carlo code as yet - that field the
deterministic codes have "locked up" at the present time.

When I attended the American Nuclear Society's Mathematics and Computation Section's
topical meeting in Monterey, California in April of 2007; Prof. William Martin, the Chairman
of the Nuclear Engineering Dept at the University of Michigan led a discussion as to when
computers were going to be powerful enough to supplant deterministic methods. It's going to
be another COUPLE DECADES before Monte Carlo codes can do some of the calculations
now done by deterministic methods.

Present day computers are NOT SUFFICIENT for nuclear weapons calculations - those are
going to take at least 10 Petaflops or up to 100 Petaflops. The current most powerful computer
[ #1 on the Top 500 list ] is Los Alamos' Roadrunner computer which is just a little more than 1 Petaflop:

http://www.top500.org/list/2008/11/100

http://www.lanl.gov/roadrunner/

Dr. Gregory Greenman
Physicist

 

[Astronuc]

I forgot about APROS, which is a Finnish code. This might be of interest, or at least a starting point.

www.vtt.fi/inf/pdf/publications/1999/P376.pdf

 

[libertad]

GE uses TGBLA (lattice code) with PANACEA (core simulator). GE was working on a transport-theory based code, and may have resumed development.
Westinghouse uses PHOENIX (lattice code) with ANC and more advanced SP-NOVA (core simulator).
AREVA uses CASMO (lattice code) with MICROBURN-B/P (core simulator).

BNFL/British Energy have used WIMS / PANTHER

I believe Swedes (Nordic countries) used PHOENIX/POLCA-7 or CASMO-4/POLCA-7 for BWRs, but perhaps use CASMO/SIMULATE or vendor codes corresponding to their fuel supply (or both). Some utilities use vendor methods or independent methods, to either design the cores or do an independent over-check.

Studsvik-Scandpower produces CASMO (lattice code) and SIMULATE (core simulator).

EPRI has developed the ARROTA core simulator, and they sponsor VIPRE & RETRAN T/H code.

There are universitiy codes like Purdue's PARCS and NC State's (Turinsky's) FORMOSA core simulators.


Traditionally, thermal-hydraulics codes are decoupled from core simulator codes.

ANL is working on a Numerical Nuclear Reactor that attempts to marry Core Simulation (Nuclear methods) with a T/H capability.

Hi Astronuc,
Is there any references in internet to introduce the nuclear codes which are developing or using by the major companies in nuclear industry?

 

[Astronuc]

The vendors like Toshiba/Westinghouse, GE/Hitachi (GEH) or Global Nuclear Fuel (GNF), AREVA, Mitsubishi, ENUSA, etc use proprietary nuclear and T/H codes, so the details will not be found in the literature.

Independent codes like CASMO (lattice) and SIMULATE (core simulator/depletion code) are more open with respect to how they are used, but again the details (the guts of the code) are proprietary.

Let me look into some of them and see what's available.

Some information is available from IAEA where some older codes have been put in the public domain.

Also, one can add GARDEL (by Studsvik) to the list. Just search Google with "GARDEL","core simulator".
Studsvik's products page - http://studsvikscandpower.com/products/

This program gives one an idea of what's available - http://meetingsandconferences.com/anfm2003/pdf/papers.pdf

http://article.nuclear.or.kr/jknsfile/v37/JK0370079.pdf

This is a good resource - http://rpd.ans.org/

 

[libertad]

Thank you in advance Astronuc

 

[Astronuc]

A more recent development

Supercomputers study reactor core behaviour
http://www.world-nuclear-news.org/NN-Supercomputers_study_reactor_core_behaviour-2701104.html
27 January 2010

A new computer algorithm that enables scientists to view nuclear fission in much finer detail has been developed by researchers at the US Department of Energy's (DoE's) Argonne National Laboratory (ANL). The code could be used in the development of new reactor designs.

The neutron transport code UNIC is being developed by a team of nuclear engineers and computer scientists at ANL which they said enables researchers to obtain a highly detailed description of a nuclear reactor core for the first time.

The calculations required to model the complex geometry of a reactor core requires massive computer memory capacity, far higher than most computers can handle. . . . .

 

[Hologram0110]

Astronuc's link didn't work for me, but I think this is to a similar article.
http://www.anl.gov/Media_Center/News/2010/news100121.html

As the question about how much computation there is nuclear engineering, the answer is a lot. Numerical methods are used all over the place because geometry and other complications prevent analytical solutions in many cases.

What PDE's do they solve?
A lot. Which ones depend on the code purpose. For example, a fuel modeling code would have to simultaneously solve PDE's for heat transport, thermal-expansion and deformation, fission gas production & release, sheath effects ect. By comparison a fuel management code would need to solve many neutron transport PDEs on a much larger volume.

 

[Astronuc]

Astronuc's link didn't work for me, but I think this is to a similar article.
http://www.anl.gov/Media_Center/News/2010/news100121.html

The World Nuclear New site is temporarily down.

Nuclear codes like SIMULATE solve the diffusion equation, normally collapsed into two energy groups. More advanced methods solve the transport equation. There are simple fuel rod models used to calculate fuel temperature.

Thermo-mechanical codes for modeling fuel rods are either finite difference or FEM. If one has proper materials properties, which including effects of radiation, than ANSYS or ABAQUS could be used to model fuel rods/elements.

FEM is used extensively for mechanical analysis of fuel assemblies and components. CFD is becoming more widely used.

 

[Viviz]

Hello everyone,

I know this is a pretty old thread, but I think somebody searching desperately for a free lattice code would bless me for this intervention.

Not aware of any open source development that is worth anything. Vendors like to keep valuable information propietary.

In fact, there's one. The lattice code Dragon is without any doubt the better lattice code freely available, given that the only other one is WIMS-D5 (which is now pretty old, 60's are far away...).

It has very similar function of a commercial product, but is open-source (LGPL).
It is developed by Ecole Polytechnique de Montreal, and is mainly used for CANDUs.

EDF (in France) also keep an eye on it (you can count some works of them on Dragon), but their main lattice code is still Apollo (like Areva and CEA).

So : if you need a free lattice code, go for Dragon. If you can pay, you can hesitate between CASMO/APOLLO/DRAGON/[insert here any other lattice codes which could compete].

 

[mjlaris]

There have been limited comparisons, e.g. between a vendor method and CASMO/SIMULATE. The comparisons were done by the vendor and utility and are not available in the public domain. There have been attemps at gamma-scanning, but results have been problematic.

One core simulation (depletion case) can be run on a single workstation, however the set up is complicated since one needs to go backwards as far as possible with respect to the reinsert fuel. Cores simulations can be run in octant, quadrant or full core depending on the assymetries being modeled. A typical 4-loop PWR (193 assy) running 17x17 fuel contains 50952 fuel rods, and they are typical nodal with 24 x 6'' (15.24 cm) axial nodes, or greater numbers if axial zoning of burnable absorbers are used with blankets of more than 6''.

Some codes are available through IAEA or universities, but usually one must register and pay a fee. Access to source code is restricted.

Application of CFD in the nuclear industry is relatively recent, and the two main codes are CFX and Star-CD. AFAIK, no one has integrated numerical reactor simulators with a CFD code. The core simulators basically contain relatively simple thermal hydraulic models which provide coolant temperature, density and void coefficients. Depletion calculations presume steady-state, although there have been some attempts to look in more detail at downpowers and power ascensions (and associated Xe-swings), control blade history effects, and various transients, e.g. rod-drop or rod-ejection accident.

I have a copy of GEs Panacea that has been ported to the PC and I run it on my laptop. Panacea is a steady state core simulator used to design rod patterns (cycle step out calculations during cycle design) and calculate core thermal limits during operation.

Mark Laris
Chief Nuclear Engineer

 

[libertad]

Hello everyone,

I know this is a pretty old thread, but I think somebody searching desperately for a free lattice code would bless me for this intervention.



In fact, there's one. The lattice code Dragon is without any doubt the better lattice code freely available, given that the only other one is WIMS-D5 (which is now pretty old, 60's are far away...).

It has very similar function of a commercial product, but is open-source (LGPL).
It is developed by Ecole Polytechnique de Montreal, and is mainly used for CANDUs.

EDF (in France) also keep an eye on it (you can count some works of them on Dragon), but their main lattice code is still Apollo (like Areva and CEA).

So : if you need a free lattice code, go for Dragon. If you can pay, you can hesitate between CASMO/APOLLO/DRAGON/[insert here any other lattice codes which could compete].

I have the source code of Dragon but I don't know how to compile it on Windows OS.
I've tried to compile it using Intel Visual Fortran Compiler 11 but I got many errors.
do you know any solution?

 

[libertad]

I have the source code of Dragon but I don't know how to compile it on Windows OS.
I've tried to compile it using Intel Visual Fortran Compiler 11 but I got many errors.
do you know any solution?

I compiled it

 

[edgepflow]

In the early 90s, there was a code called TWODANT (two dimensional diffusion accelerated neutron transport). I remember it would run on a modest PC and would give pretty good results. Don't know if it is used much anymore.

 

[Hologram0110]

In the early 90s, there was a code called TWODANT (two dimensional diffusion accelerated neutron transport). I remember it would run on a modest PC and would give pretty good results. Don't know if it is used much anymore.

As far as I've observed, the modeling codes tend to grow in complexity with capability of the times. For example, while a 2D model might have been used in the early 90s, if it were used now additional factors would be incorporated to provide more detailed analysis.

The easiest example I can think of is one of my professors in nuclear engineering used to use a 2D neutron diffusion model with about 35 points in the radial direction. Now in class when we write similar codes, they can comfortably be solved with 1000's of points.

Check out http://www.physorg.com/news183389151.html" work being done at Argon with super-computers. Normally to try and get similar results a multi-cell model would need to be used to cut down on ram and processing requirements but with a super computer and clever programing they don't need it.

 

[Viviz]

I have the source code of Dragon but I don't know how to compile it on Windows OS.
I've tried to compile it using Intel Visual Fortran Compiler 11 but I got many errors.
do you know any solution?

So sorry, I only read your question more than a month later.
The easiest way (IMHO) to perform that is to install a http://en.wikipedia.org/wiki/Cygwin" , and install gcc4 core package, and gcc4-fortran.
Then you can compile it using gfortran. Readme explains it all.

Pretty recently, the possibility has been added to compile it directly under a Windows environment, but I didn't try that at all. Read more at :
http://www.polymtl.ca/merlin/faq.htm
http://www.polymtl.ca/merlin/archives.htm

Don't forget to download ENDF libraries, you'll need it :
http://www.polymtl.ca/merlin/libraries.htm

Learn to use/write/read bash scripts, you'll also need it.

Hope that my very late answer will help you.

 

[Viviz]

I compiled it

Darn. Again 1 month late xD
Sorry for the mess, and happy you did it ;)