Can Numerical Simulation Revolutionize Nuclear Reactor Design?

[Astronuc]

This is a developing area in the nuclear industry involving large scale multiphysics computation.

For the past several years Argonne National Laboratory and Purdue University have been supported by the Department of Energy (DOE) and the Electric Power Research Institute (EPRI) to develop a next generation nuclear reactor simulator based on the commercial Computational Fluid Dynamics (CFD) code STAR-CD to solve the temperature/fluid field and the integral transport code DeCART to solve the Boltzmann Equation for the neutron flux distribution in the reactor core. Applications of the Numerical Nuclear Reactor (NNR) to practical Light Water Reactor problems has required more than 100 million mesh and tens of hours of computational time on Linux clusters for the solution of a single reactor state point.

Clearly computational efficiency must be improved. That will be quite a challenge.

Some background on DeCART - http://www.inl.gov/scienceandtechnology/cams/d/high_end_computing_star_simon_lo.pdf

Tom Downar is with PARCS (Purdue Advanced Reactor Core Simulator) which is part of the Purdue Nuclear Engineering program.
https://engineering.purdue.edu/PARCS

 

[PerennialII]

... fashinating work they got going, and somehow get the feeling that they can actually pull through it, they may not be too greedy what comes to their goals & have an idea of the immense work ahead - but large scale multiscale is spreading. Got to taunt our guys with this if aren't aware of the project.

 

[Astronuc]

Dave Weber from Argonne National Lab did a presentation this past Friday at a conference I was attending. I talked to him afterward, and he will be providing more information. I'd like to get more information from the PARC program.

One aspect in which I'm particularly interested is the ability to properly calculate local asymmetries locally within assemblies (pin-to-pin) and grossly (e.g. octant or quadrant, or axial power tilts) within the core.

Xe transients would be particularly interesting to model, as well as LOCA and RIA.

 

[jdsneeder]

Thanks for link, I did some undergrad work for Dr. Downar (great guy and very good prof) back in the late 80s (Integral Fast Reactor, Depletion Perturbation Theory) and it was good to see what they are up to.

 

[Astronuc]

http://www.eurosafe-forum.org/products/data/5/pe_385_24_1_seminar2_05_2005.pdf?PHPSESSID=4cd682861b80321facb9ca4f3345be75 (pdf, so use 'save target as' if bandwidth limited)

Abstract:

Incorporating full three-dimensional models of the reactor core into system transient codes allows for a “best-estimate” calculation of interactions between the core behavior and plant dynamics. Considerable efforts have been made in various countries and organizations on the development of coupled thermal-hydraulic and neutronics codes. Appropriate benchmarks have been developed in international co-operation led by NEA/OECD that permits testing the neutronics/thermal-hydraulics coupling, and verifying the capability of the coupled codes to analyze complex transients with coupled core-plant interactions. Three such benchmarks are presented in this paper – the OECD/NRC PWR MSLB benchmark, the OECD/NRC BWR TT benchmark, and the OECD/DOE/CEA V1000CT benchmark. In order to meet the objectives of the validation of best-estimate coupled codes a systematic approach has been introduced to evaluate the analyzed transients employing a multi-level methodology. Since these benchmarks are based on both code to code and code to data comparisons further guidance for presenting and evaluating results has been developed. During the course of the benchmark activities a professional community has been established, which allowed carrying out indepth discussions of different aspects considered in the validation process of the coupled codes. This positive output has certainly advanced the state-of the art in the area of coupling research.

 

[bhagyesh patil]

respected sir,
i want some details related to nuclear reactor modeling aspects...i will be thankfu lto u , if u guide me in the same...if possible send me good links realted to nuclear reactor modelling aspects...

 

[Morbius]

This is a developing area in the nuclear industry involving large scale multiphysics computation.

For the past several years Argonne National Laboratory and Purdue University have been supported by the Department of Energy (DOE)

Astronuc,

Programs are currently in the planning phases for the Dept. of Energy to conduct, under
the auspices of the "Global Nuclear Energy Partnership" GNEP;

http://www.gnep.energy.gov/

an initiative similar to the "Advanced Simulation and Computing" ASC program that has
for the past decade been very successful in increasing the computational capabilities
for the Dept. of Energy's responsibilities for the stewardship of nuclear weapons in an
era without nuclear testing:

http://www.sandia.gov/NNSA/ASC/

The ASC program has developed both advanced software, as well as hardware;
resulting in the development of several advanced computing platforms:

http://www.sandia.gov/NNSA/ASC/platforms.html

including the world's most powerful computer, BlueGene/L at Lawrence Livermore
National Laboratory:

http://www.llnl.gov/asc/platforms/bluegenel/

Dr. Gregory Greenman
Physicist

 

[Clausius2]

to develop a next generation nuclear reactor simulator based on the commercial Computational Fluid Dynamics (CFD) code STAR-CD to solve the temperature/fluid field and the integral transport code DeCART to solve the Boltzmann Equation for the neutron flux distribution in the reactor core. Applications of the Numerical Nuclear Reactor (NNR) to practical Light Water Reactor problems has required more than 100 million mesh and tens of hours of computational time on Linux clusters for the solution of a single reactor state point.

Clearly computational efficiency must be improved. That will be quite a challenge.

I don't truly know too much about this, but I can do imagine the kind of difficulty that the flow into a nuclear reactor has. I would like to know how much useful stuff do you get out of the simulations, I think you might be employing safety factors of order 1/2 or more. Solving the thermo-chemical-fluid problem seems unaffordable to me except if you run it on a Super Computer with full capacity and DNS codes. Solving it doing strong assumptions (as symmetry, steadiness, Continuum Flow, Equilibrium Flow, Chemical reduced mechanisms, Turbulence Models, decoupled wall Heat transfer and wall reactivity) is something more or less affordable (spending those tens of hours), but giving a lot of uncertaintity.

I am particularly not prone to believe in large scale simulations unless they are done with DNS (currently unemployable in large systems). "Modelling the models", as I use to refer for instance to the turbulence models, is something good for engineering because it gives you a 50% less of uncertaintity, but not very interesting for science. I am kind of pesimistic with the actual state of the art of CFD employed in industry. I am sat in front of a door of a guy who does a lot of DNS in UCSD, in problems now applied to Oceanography, involving moderate large scales, but not employing the typical 100km-cell-size of the bulk simulations using LES or K-epsilon models that some people even in science unfortunately still do.

 

[Astronuc]

Solving it doing strong assumptions (as symmetry, steadiness, Continuum Flow, Equilibrium Flow, Chemical reduced mechanisms, Turbulence Models, decoupled wall Heat transfer and wall reactivity) is something more or less affordable (spending those tens of hours), but giving a lot of uncertaintity.

That's why I am hoping to become more involved. We've had tentative discussion on coupling a thermo-mechanical code to this system , which will be way more complicated. One of the biggest challenges is fluid-structure interaction (FSI).

Proper benchmarking is the key. The gross uncertainties are not as large as one would think - it's something like bounded chaos. Locally at any given time, the uncertainties are larger - but on average are not so - such is the nature of turbulence.

Thermocouples can be used to monitor temperature distributions and laser doppler anemometry can be used to monitor/measure the flow distribution. The integrated energy distribution can also be measured by subsequent post irradiation examination (PIE), part of which is the measurement of radioisotopic distribution.

 

[Morbius]

I
I am particularly not prone to believe in large scale simulations unless they are done with DNS (currently unemployable in large systems).

Clausius,

The hydrodynamic flow in a reactor is NO WHERE NEAR the complexity that would
require a DNS code. We use DNS simulation to solve some of the problems in my
field.

I am sat in front of a door of a guy who does a lot of DNS in UCSD, in problems now applied to Oceanography, involving moderate large scales, but not employing the typical 100km-cell-size of the bulk simulations using LES or K-epsilon models that some people even in science unfortunately still do.

Actually one can do surprising well with LES, K-epsilon, k-L... models. It all depends
on how much the hydrodynamics is affected by turbulence. For the Reynolds
numbers one finds in a reactor - the answer is NOT MUCH.

Doing DNS simulation on the coolant flow in a reactor would be "gilding a lilly".
Dittus-Boelter analysis will suffice for reactor conditions.

Dr. Gregory Greenman
Physicist

 

[Clausius2]

That's why I am hoping to become more involved. We've had tentative discussion on coupling a thermo-mechanical code to this system .

OOps sorry, I don't remember that discussion, but next time we will have it-

The hydrodynamic flow in a reactor is NO WHERE NEAR the complexity that would
require a DNS code. We use DNS simulation to solve some of the problems in my
field. The flow is designed to be laminar.

Really? I must admit I'm having difficulties on believing in what you said. What kind of reactor are we talking about? Are you talking about a local (but very local) behavior of the flow? Are you guaranteeing me there is no single flow separation in the whole chamber? Have you ever simulated even the problem of the transversal flow over a periodic sequence of pipes? Is the Reynolds number so small to be laminar? I do NOT think so.

Actually one can do surprising well with LES, K-epsilon, k-L... models. It all depends
on how much the hydrodynamics is affected by turbulence. For the Reynolds
numbers one finds in a reactor - the answer is NOT MUCH.

Actually all the improvements of NASA and aeronautic companies on wing designs are in part due to the use of these models. The problem comes when used massively, without tunning the model coefficients, and moreover, when there is little o no capacity of interpreting correctly the results (and critically) by the engineer in charge.

The hydrodynamics once the flow is turbulent and statistically steady is almost no affected by the Reynolds Number, BUT if the flow is not turbulent, the switch to turbulent DOES change its characteristic a lot, and in particular it DOES change the heat transfer dynamics, which turns out to be the most important thing in a rector right?.

 

[Astronuc]

OOps sorry, I don't remember that discussion, but next time we will have it.

It's a very recent development.

Flow in a reactor - in the core - is turbulent. Re is on the order of 400,000 in a PWR core.

 

[Morbius]

Really? I must admit I'm having difficulties on believing in what you said. What kind of reactor are we talking about? Are you talking about a local (but very local) behavior of the flow? Are you guaranteeing me there is no single flow separation in the whole chamber? Have you ever simulated even the problem of the transversal flow over a periodic sequence of pipes? Is the Reynolds number so small to be laminar? I do NOT think so.

Clausius2,

Yes - I used to work in the field of nuclear reactor design. I am a computational
physicist and that is EXACTLY what I have done - simulated the flow in reactors.

As for your point about transversal flow over a sequence of periodic pipes; what
direction do you think the coolant flows in a reactor?

Actually all the improvements of NASA and aeronautic companies on wing designs are in part due to the use of these models. The problem comes when used massively, without tunning the model coefficients, and moreover, when there is little o no capacity of interpreting correctly the results (and critically) by the engineer in charge.

BIG difference between low density air at aircraft speeds and water flow in a reactor.

The hydrodynamics once the flow is turbulent and statistically steady is almost no affected by the Reynolds Number, BUT if the flow is not turbulent, the switch to turbulent DOES change its characteristic a lot, and in particular it DOES change the heat transfer dynamics, which turns out to be the most important thing in a rector right?.

If you are talking about an accident scenario wherein the hypothesis is that the
reactor vessel has been emptied, and we are calculating the re-fill by the ECCS;
then yes you have extremely turbulent conditions there.

If you are talking about steady-state operation of the reactor; then NO - the reactor
is designed so the flow is not turbulent to the degree to require DNS. The operator
is also required to operate the reactor so that you don't have the drop in heat transfer
that one would get with extreme turbulent flow.

Dr. Gregory Greenman
Physicist

 

[Clausius2]

Please, do you mind giving me a reference for what you are saying?

Or maybe we can work out the Reynolds number based on the fuel rod length (I don't care if it is a longitudinal flow), and work out if we have turbulen flow or not.

And with all my respect, you are a computational physicst, but I am a mechanical engineer, and usually Fluid Mechanics IS NOT the area of expertise of a physicist. Maybe I haven't been doing simulation of reactors (because as I said massive simulations are useless for science if you do it employing the yet well known shortcuts).

I still don't believe that stuff of laminar flow, and if you do some calculation or give me a reference I will be glad. Otherwise don't justify it saying that you have been doing that for a long time, because people also do wrong things for a long time in their jobs.

 

[Morbius]

And with all my respect, you are a computational physicst, but I am a mechanical engineer, and usually Fluid Mechanics IS NOT the area of expertise of a physicist.

Clausius2,

With all due respect; this IS my field - I am BOTH a Physicist and a Nuclear Engineer.

This IS what I do - I write the computer programs that do this type of calculation.

Maybe I haven't been doing simulation of reactors (because as I said massive simulations are useless for science if you do it employing the yet well known shortcuts).

Then you don't understand how computational physics is done with reactors.
You don't just do the calculation and be done with it. The calculations have to
match EXPERIMENTS!

All our calculation had to be benchmarked against actual experiments.

These consisted both of small-scale experiments that simulated conditions in a
reactor core, as well as experiments in test reactors.

When calculating nuclear reactors, nobody let's you calculate using incorrect models.
The assumption by the NRC is that your model doesn't work; and the burden of proof
is on the designers / analysts to prove the models DO WORK.

Dr. Gregory Greenman
Physicist

 

[Clausius2]

Then you don't understand how computational physics is done with reactors.
You don't just do the calculation and be done with it. The calculations have to
match EXPERIMENTS!

That's why I am asking you for a reference or calculation example. Give me order of magnitudes of the scales involved. It is pretty easy to figure it out.

These consisted both of small-scale experiments that simulated conditions in a reactor core, as well as experiments in test reactors.

Small scale? I hope you/or the board of engineers are using dynamical similarity, because it is laminar flow, don't forget that. Otherwise you will be saying with your own words that the Reynolds number doesn't matter in this business, and that's only true when it is so large for assuming statistically steady flow and turbulent flow.

When calculating nuclear reactors, nobody let's you calculate using incorrect models.
The assumption by the NRC is that your model doesn't work; and the burden of proof
is on the designers / analysts to prove the models DO WORK.

Dr. Gregory Greenman
Physicist

This whole thread goes about until what extent these models work?. I'll put you an example. I know that some heat engine companies use codes for simulating the thermodynamic conditions in the combustion chamber. They are crowded with equations, every of them unidimensional (see the book of Heywood on this topic), giving you the maximum pressure, temperature of exhaust, pressure of exhaust, heat released, work done and all this bulk magnitudes. For sure it serves to the designer to have a first order of magnitude of the machine, and if you test it with an experimental model it does work for these purposes. But imagine I want to design the chamber or a intake valve, or the piston, or the exhaust pipe, do you think all those equations are useful for me? do you think that model is capturing the physics I need? Maybe Westinghouse is interested on knowing the bulk properties of your reactor, doing a lot of assumptions to obtain them, but you are not definitely capturing the physics for a detailed calculation, or at least this is my impression without knowing yet what kind of calculation you do.

 

[Morbius]

Maybe Westinghouse is interested on knowing the bulk properties of your reactor, doing a lot of assumptions to obtain them, but you are not definitely capturing the physics for a detailed calculation, or at least this is my impression without knowing yet what kind of calculation you do.

Clausius2,

You have absolutely no idea about the types of computer models that are use.

They are not just filled with dimensionless correlations that somebody pulled out of the
air. There's been LOTS of research by the national laboratories on validating these
computer models so that reactor calculations are done only on CERTIFIED codes.

Universities, national laboratories, in addition ot the manufacturers have all conducted
experiments to certify these computer models. There is an entire body of knowledge.

Perhaps you should check out "Nuclear Science Abstracts" to see the TONS of work
that goes on to validate these models.

Nobody licenses a reactor based on calculations using the type of computer model
you gave with your combustion chamber example.

Dr. Gregory Greenman
Physicist

 

[Morbius]

That's why I am asking you for a reference or calculation example. Give me order of magnitudes of the scales involved. It is pretty easy to figure it out.

Then you can do it. I can't openly discuss design details on things that I've worked
on.

Dr. Gregory Greenman
Physicist

 

[Clausius2]

Ok, I will take a look at some of those abstracts.