Upload
chrissbans
View
223
Download
0
Embed Size (px)
Citation preview
7/29/2019 Flowin St Bility
1/18
Nuclear Engineering and Design 215 (2002) 153170
On the analysis of thermal-fluid-dynamic instabilitiesvia numerical discretization of conservation equations
Juan Carlos Ferreri a,1*, Walter Ambrosini b
a Autoridad Regulatoria Nuclear, A6. del Libertador 8250, 1429 Buenos Aires, Argentinab DIMNP, Facolta di Ingegnera, Uni6ersity of Pisa, Via Diotisal6i 2, 56126 Pisa, Italy
Abstract
The numerical aspects involved in the prediction of fluid-dynamic instabilities are shortly reviewed, mainly
addressing single-phase natural circulation flows. In particular, some open literature, produced since the pioneering
work of Welander [J. Fluid Mech. 29 (1967) 17 30] and related to single-phase thermosyphon loops, is firstly
considered; then the analysis turns to summarize some early calculations and examples of code verification results
dealing with the prediction of single-phase natural circulation. The work previously performed by the Authors in
relation to Welanders problem and other relevant single-phase thermosyphon loop cases is reviewed, with main
emphasis on the adopted methodology, based on numerical discretization of governing equations. These results serve
to illustrate how time domain codes may be used to analyze such problems and also to highlight the effect of
considering different constitutive laws in the computed linear stability map. Reference to the application of this
methodology to boiling channel stability analysis is also made to demonstrate its generality. The results show boththe qualitative and quantitative details of the changes that stability maps undergo as a consequence of numerical
discretization of governing equations. The recent attempt to make easier the application of the methodology by the
use of automatic FORTRAN code differentiation tools is finally reported, showing quantitative results concerning
calculation of the sensitivity of simulation results to physical and numerical parameters. 2002 Published by Elsevier
Science B.V.
www.elsevier.com/locate/nucengdes
1. Introduction
The need for modeling of natural circulation
flows in hydraulic circuits is the result of the
importance of the related phenomena in engineer-ing sciences and applications. References concern-
ing this type of flows in the last 20 years are so
numerous as to provide a trivial proof of this
assertion. As a matter of fact, there are several
relevant motivations to deal with this subject: in
addition to the traditional widespread utilization
in industry, natural circulation is receiving atten-
This paper is an updated and condensed joint version of
the invited lecture: Single-phase Natural Circulation in Simple
CircuitsReflections on its Numerical Simulation by J.C.F.
and the paper Parameter Sensitivity in Single-Phase Natural
Circulation via Automatic Differentiation of FORTRAN Codes
by the authors, both presented at the Eurotherm63 Seminar.
* Corresponding author.
E-mail address: [email protected] (J.C. Ferreri).1 Member of CONICET, Argentina.
0029-5493/02/$ - see front matter 2002 Published by Elsevier Science B.V.
PII: S 0 0 2 9 - 5 4 9 3 ( 0 2 ) 0 0 0 4 8 - 1
mailto:[email protected]:[email protected]7/29/2019 Flowin St Bility
2/18
J.C. Ferreri, W. Ambrosini/Nuclear Engineering and Design 215 (2002) 153170154
tion in the nuclear field for passive heat removal
systems in innovative reactors, which makes it a
subject deserving even greater attention from the
technical community.
The problem of simulating natural circulation
flow is a rather complex one. For instance, it is
difficult to establish which are the appropriate
values of system parameters, such as singular anddistributed pressure losses in loop components,
because the flow in pipes and restrictions, like
fittings and valves, is usually non-developed. The
usual practice in this field, consists in adopting
well established results obtained for steady-state
conditions, such as using Idelchik (1966) hand-
book for estimating singular pressure drops; but
there are well known limitations in the application
of these relationships to non-developed or time
dependent flows (these aspects have been reviewed
in many references, e.g. Zvirin (1981)). The samereasoning applies to distributed friction resistance
laws. The balance between friction and driving
forces determines steady-state conditions in natu-
ral circulation flows, then, depending on the
choice for closure laws, the system may present a
different behavior. Obviously, this represents a
key point in determining the capabilities of mod-
els; it will be noted in the following that in the
case of simplified analyses based on dimensionless
equations, the choice of adequate independent
variables may put constraints in the selection of
constitutive laws that have an impact on results.
Scaling natural circulation flows to get an ap-
propriate experimental apparatus representative
of real life installations is not trivial and has been
the subject of many analyses up to the present
time. Relevant information on this subject may be
found in Zvirin (1981), Ishii and Kataoka (1984),
DAuria et al. (1991), Vijayan and Austregesilo
(1994), Wulff and Rohatgi (1999). This is indeed
an important aspect because it defines if a model
will reasonably represent a given physical appara-
tus. Sometimes, scaling leads to the adoption of
the 1-D approximation; this may, in turn, hide
important aspects of the system physics. A simple
example of this situation consists in keeping the
height of the system unchanged to get the same
buoyancy; then, if the system is scaled accordingly
to the power/volume ratio, the cross section area
of the volume will be reduced; this leads to a
much smaller pipe diameter that makes the 1-D
representation reasonable, at the cost of eliminat-
ing the possibility of fluid internal recirculation. A
workaround for this situation is providing paths
for recirculation, in the form of additional, inter-
connected components; however, this solution
may impose the flow pattern in the system and thebalance between these aspects is a challenge to
any practitioner in natural circulation modeling.
An additional important characteristic of natu-
ral circulation flows is that they are non-linear in
essence. The problem is particularly challenging
since no matter how simple the hydraulic loop can
be, its behavior is usually very complex: depend-
ing on the ratio of driving forces (heat sources/
sinks strength) to the integrated friction along the
loop, the flow may become oscillatory, showing
flow reversals. The modeling of the dynamic be-havior of non-linear systems is presently accom-
plished by different techniques, depending on the
characteristics of the problem under consider-
ation. We leave apart lumped parameter models
despite their usefulness for studying, for example,
the system behavior beyond the stability
threshold. It is our interest to focus on the solu-
tion of natural circulation flows as represented by
non-linear partial differential equations (PDEs),
which are the only feasible choice for adequately
modeling the behavior of distributed parameter
systems. Then, appropriate techniques must be
applied to get a reasonable picture of both stabil-
ity conditions and unstable transient behavior.
Discretization of PDEs by numerical schemes
leading to systems of algebraic equations is the
usual technique in complex cases, in order to
provide a realistic prediction of plant behavior.
The large thermal-hydraulic (TH) system codes
adopted for safety analyses of nuclear power
plants are relevant examples of this choice. How-
ever, it is well known that numerical schemes
usually adopted in the simulation of natural circu-
lation flows introduce damping in the solution;
therefore, unstable flows may be turned stable
because of this added diffusion. Numerical
schemes may affect the expected convergence of
computed results, even in simple situations; these
effects may originate, for instance, in the numeri-
7/29/2019 Flowin St Bility
3/18
J.C. Ferreri, W. Ambrosini/Nuclear Engineering and Design 215 (2002) 153170 155
cal coupling of the discretised equations or in the
criteria adopted for variable lumping in finite
volume schemes.
Therefore, when applied to fluid-dynamic insta-
bilities, most of the adopted numerical methods
are conditioned by the above mentioned effects of
discretization, that introduce unwanted numerical
diffusion and other spurious effects which maystrongly change the predicted dynamic behavior
of the system. These effects are often found to
make difficult a reliable stability analysis by nu-
merical means (see e.g. Bau and Torrance, 1981;
Misale and Tagliafico, 1987; Vijayan et al., 1995;
Wu and Almenas, 1994; Ambrosini and Ferreri,
1998). Examples of accurate estimates of the ef-
fect of numerical diffusion on the predicted degree
of stability or instability are not so frequent. Basic
references to the effects of the numerics on com-
puted results are the works of Stuhmiller andFerguson (1979), Barre et al. (1993), Mahaffy
(1993), Yih-Yun Hsu (1994), Ginestar Peiro and
March-Leuba (1998). It is, therefore, important to
quantify the sensitivity of the results obtained
from numerical simulations to physical and nu-
merical discretisation parameters.
This paper focuses on the numerical aspects of
1-D calculations of fluid-dynamic instabilities,
with main reference to single-phase natural circu-
lation flows in the context of nuclear safety evalu-
ations. We, therefore, consider systems amenable
to one-dimensional (1-D) modeling, thus disre-
garding a wide class of problems for which this
assumption would be questionable. The aspects
related to the effect of numerical discretisation of
PDEs on the computed system behavior are given
particular attention, summarizing previously pub-
lished work performed with a systematic method-
ology of analysis and presenting recent
developments related to the evaluation of sensitiv-
ity to physical and numerical parameters. The
effect of the choice of constitutive laws for evalu-
ating friction is also addressed. To complete the
panorama of obtained results the extension of the
methodology of analysis also to boiling channel
stability problems is mentioned.
Though the goal of the work is to discuss the
capabilities of system codes in predicting natural
circulation, other aspects relevant in this frame
are not directly addressed. One of these aspects is
the so-called user effect, i.e. the effect that the
modeling choices of a given code user may have
on the computed results. This problem received
careful attention quite recently in the field of
nuclear safety evaluations (e.g. see the papers in
IAEA, 1998). However, the problems here ad-
dressed are generally simple enough as to mini-mize the effect of user choices on the results of
code application.
2. Numerical simulation of 1-D fluid-dynamic
instabilities
2.1. Early experience on single-phase natural
circulation
Single-phase natural circulation flows may bemodeled via the use of two equations:
the integral momentum balance equation:
dW
d~=CB BTCF f(W) WW (1)
the distributed energy balance equation:
(q
(~+W
(q
(s=Q+F
(2q
(s2(2)
In these equations, written in dimensionless
form, W represents some measure of the flow rate,
CB is a coefficient to weight the buoyancy term
(BT) which is usually the integral of temperature
along the loop length, CF is a coefficient in the
friction law and f(W) gives the form of the fric-
tion law. Moreover, q is some measure of the fluid
energy, usually temperature, Q the energy gain/
loss by wall transfers (it includes boundary condi-
tions), F is a coefficient to weight the influence of
axial diffusion and, finally, ~ and s are the non-di-
mensional time and space co-ordinates,
respectively.
In the papers by Ferreri and Doval (1984, 1985,
1988), Doval and Ferreri (1988), one of the au-
thors started to document his experience on the
numerical modeling of natural circulation flows in
nuclear safety related problems. This subject, to-
7/29/2019 Flowin St Bility
4/18
J.C. Ferreri, W. Ambrosini/Nuclear Engineering and Design 215 (2002) 153170156
gether with the well-posedness of the governing
equations in thermal-hydraulics, used to be (and
continues to be) a field of active work. The prob-
lem faced at the time was using TH system codes
to compute the transition from forced circulation
to natural circulation flow in a nuclear power
plant. The codes used at the time, as described in
Ferreri and Doval (1984), failed to converge to-
ward a solution in the last stage of the simulation.
This fact and the need to get results lead to the
development of a 1-D, single-phase, single-loop
simulation code, including concentrated and dis-
tributed heat sources and sinks to compute natu-
ral circulation flows (see Ferreri and Doval, 1984,
1985, 1988). The obtained results (quoted in Fer-
reri and Doval, 1985) pointed out the damping of
flow perturbations as a function of the adopted
discretization in a closed loop.In Ferreri and Doval (1984), the developed
code consisted in a 1-D version of the iterative,
semi-implicit SOLA algorithm. It adopted an im-
plicit finite-difference method for the energy equa-
tion in the primary loop and heat exchangers. The
momentum equation was solved in two ways: a
fully discrete finite-difference method using the
forward time, upwind-space (FTUS) approxima-
tion and a formulation obtained by the usual
integration along the loop. The cases considered
were:(a) the results in Zvirin and Greif (1979), making
reference to a previous fundamental paper by
Welander (1967), concerning a very idealized
thermosyphon loop with point heating source
and sink;
(b) the results of Zvirin et al. (1981), concerning
the natural circulation flow in a nuclear
loop, formed by parallel loops and plenty of
fittings;
(c) the computational simulations of toroidal
loops by Mertol (1980).These simulations were aimed at providing
some hints on the convergence of the numerical
results to theoretical and experimental results. In
Zvirin and Greif (1979), no instabilities were ob-
tained, even in cases of theoretically unstable
flows. These authors attributed these results to the
adopted analysis technique.
The tests reported in Ferreri and Doval (1984)
allowed for reproducing Welander (1967) results
and studying the effect of the friction correlation
and of the number of nodes (i.e. of the numerical
diffusion) in the transient behavior of the loop.
Consideration was given to the minimum number
of cells providing enough damping to keep the
results stable (in stable cases) and rapidly conver-gent. The economy on cell number was manda-
tory because of CPU time and memory
restrictions by these years.
Keeping in mind the final application of the
code, the results in Zvirin et al. (1981) were then
considered. Surprisingly, the results obtained by
Ferreri and Doval (1984) provided a better ap-
proximation of the experimental data than the
ones by Zvirin et al. (1981), doubtless due to a
fortuitous choice of better system parameters, the
critical aspect mentioned in the Section 1.Concerning the work by Mertol (1980), the task
was determining the minimum number of cells
necessary to keep the system in neutral oscillatory
state around the theoretical steady-state flow rate.
A rule of thumb was learnt from that work: one
to one ratio up to one to three between cell length
and diameter usually assured catching flow oscil-
lations. The time step was set to keep the cell
Courant number near to 0.8, a basic tenet to get
low diffusion while keeping the calculations stable
under the forward time upwind space (FTUS)
approximation. The analysis of non-linear sys-
tems, as applied to natural circulation flows in
loops, was beginning with a theoretical analysis
by Hart (1984, 1985). The conceptual explanation
of flow instabilities in Welanders paper (We-
lander, 1967) remained valid in these analyses.
Bau and Wang (1992) reviewed the literature re-
lated to this way of analysis, that kept growing up
to date.
The aforementioned process of verification and
partial validation of the code allowed to the sin-
gle-phase natural circulation flows in a simplified
version of a nuclear power plant. It must be
mentioned that in this plant (Atucha-I pressur-
ized, heavy-water reactor) the core consists in
nearly 250 vertical channels immersed and packed
in a moderator tank, therefore, the 1-D approxi-
mation was a reasonable one. The first stage of
7/29/2019 Flowin St Bility
5/18
J.C. Ferreri, W. Ambrosini/Nuclear Engineering and Design 215 (2002) 153170 157
the transient consisted in a pump trip without
depressurization; then, after the reactor scram, the
flow decreased up to the point in which natural
circulation flow was established. At this point, the
system code adopted at the time (RELAP4/
MOD5) ceased to converge. The values of the
system parameters were then fed to the in-house
developed natural circulation flow code and, aftersome minor oscillations (in the order of 10%),
stabilized quite nearly to the system code values.
After this, a transient was simulated, whose re-
sults in the long term, i.e. the steady-state flow
rate for residual power, fairly agreed with the
ones known by design, as to testify the appropri-
ate simulation of the ratio between driving forces
to loop friction. Then, despite the limited code
capabilities, results were quite good as a conse-
quence of the time spent during the stage of
verification and partial validation. This code wasnot applied to the case of the Embalse CANDU
pressurized, heavy-water, nuclear power plants;
however, up to recently, this type of configura-
tions (see Vijayan and Date, 1990, 1992) have
been the subject of active research.
The use of TH system codes is essential for the
safety assessment of nuclear power plants but
their application in the analysis of fluid-dynamic
instabilities raises the question of the validity of
the results, as already pointed out in previous
studies (see for e.g. the works by DAuria et al.,
1991, 1997; DAuria and Galassi, 1998; Wulff et
al., 1992). The effects of numerical discretization
is the subject of main interest here and in what
follows attention will be focused on this impor-
tant aspect of code utilization.
2.2. Systematic analysis of truncation error
effects in single-phase flow
Based on the experience gained in the afore-
mentioned early developments, it was decided to
check the applicability of the new series of TH
systems codes like RELAP5/MOD3.2 (see Carl-
son et al., 1990) in the calculation of unstable flow
problems. The motivation for pursuing the work
was the suspicion that the code, making use of
first order numerical methods, should behave as
the programs adopted in previous experience. Ad-
ditional interest for the subject was found consid-
ering information on the observation of similar
behavior in blind inter-comparison experiments
(e.g. the final report by Ferraz Bastos, 1997) and
simultaneous work (Wu and Almenas, 1994)
This study provided the occasion to cope with
the problem at a very basic level, developing an
activity that included the following phases: analysis of the behavior of RELAP5 in the
application to a particular single-phase natural
circulation problem;
study of the same problem by the use of ana-
lytical (modal expansion) technique and by dif-
ferent numerical methods;
systematic analysis of truncation error effects
by linearization of numerically discretized
equations;
consideration of other single-phase ther-
mosyphon loop geometry by the samemethodology;
extension of the methodology to deal with
boiling channel instabilities.
All this work was summarized in Ferreri et al.
(1995), Ambrosini and Ferreri (1997a,b, 1998),
Ferreri and Ambrosini (1999), Ambrosini et al.
(1999, 2000). What follows is a brief mention of
these results, giving explicit consideration to these
references.
In Ferreri et al. (1995), Welanders problem
(Welander, 1967) was put under analysis using the
RELAP5.x series of codes. The problem deals
with natural circulation flow in a very simple
situation: namely, flow driven by buoyancy in a
loop constructed with two parallel, adiabatic, ver-
tical tubes. The loop has a point heat source at its
bottom and a point heat sink at its top. In fact,
the mode of heat transfer is assumed to be such
that the product of the heat transfer coefficient in
the source by the heated length is kept constant
when the heated length tends to zero. The same
holds for the heat sink so that the physically finite
heated lengths become points. Welander per-
formed his analysis only for laminar flow; how-
ever, his results were generalized to consider
turbulent flow. Chen (1985), among many other
authors, performed such analysis for a more gen-
eral loop considering heated (cooled) horizontal
tubes of finite length. In Ferreri et al. (1995), the
7/29/2019 Flowin St Bility
6/18
J.C. Ferreri, W. Ambrosini/Nuclear Engineering and Design 215 (2002) 153170158
analysis in Welanders paper was generalized in
order to keep strictly the same hypotheses of the
original derivation.
In its form generalized to turbulent flow, the
problem is expressed by Eqs. (1) and (2) where the
Blasius law is adopted to quantify friction in
momentum equation, thus resulting in a depen-
dence of the friction term on flow rate raised tothe power 1.75. Appropriate boundary conditions
are then imposed to translate the effect of source
and sink on the fluid passing through them.
The application of RELAP5 to an unstable case
of Welanders problem showed that (Fig. 1): depending on the node size in the legs and the
adopted time-step the semi-implicit and the
nearly-implicit numerical methods available in
the code may predict both stability and
instability;
the semi-implicit numerical method is less dif-fusive than the nearly-implicit one;
when sufficient detail is adopted in discretizing
the adiabatic legs, unstable behavior is pre-
dicted showing, as expected, successive flow
reversals.
The results shown in Ambrosini and Ferreri
(1998), that reports most of the material obtained
in the study on Welanders problem, dealt with
several of the relevant aspects mentioned in the
Section 1. In particular, it was shown that conver-
gence to exact steady-state in Welanders problem
could only be achieved by giving consideration to
a special form of heat sources/sink boundary con-
dition. This occurs because the single nodes
adopted to simulate the source and the sink can-
not provide the correct boundary conditions at
the inlet of each adiabatic leg owing to the poor
description of the actual temperature distribution
in the heater and cooler. This result had been
previously documented by Grandi and Ferreri
(1991) showing the inter-dependence of boundary
conditions application and the numerical scheme
used.But the crucial topic in Ambrosini and Ferreri
(1998) was to ascertain the reasons for the influ-
ence of nodalization on stability predictions ob-
served in RELAP5. To help in testing and in
having reference results, some simple codes were
Fig. 1. RELAP5 results for an unstable case of Welanders problem (Dt=0.5 s).
7/29/2019 Flowin St Bility
7/18
J.C. Ferreri, W. Ambrosini/Nuclear Engineering and Design 215 (2002) 153170 159
Fig. 2. The flow rate for the FTUS scheme using 1000 nodes and a modal expansion with 500 modes and the FTUS O(1 /1000)
numerical diffusion coefficient.
developed to solve natural circulation flow in this
simple system. One code version was based on the
modal approximation, to get diffusion-free solu-
tions. Other finite-differences code versions im-
plied: the forward-time, upwind method (FTUS)
being first order accurate in both space and time,
an implicit-time upwind method (ITUS), again
first order accurate, and the second order McCor-
mack method and Warming-Beam explicit
methods.
The effects of nodalization on the computed
neutral stability boundaries were quantified and
put into perspective against diffusion-free results.
It is interesting now to compare the results ob-
tained using the explicit upwind method with the
results of a modal solution with a second order
term simulating numerical diffusion. With this
aim, the diffusion coefficient is defined as:
D=q Ds
2
1
q DtDs
(3)
where q is the loop volumetric flow-rate, Ds is the
space interval and Dt is the time step, all ex-
pressed in dimensionless form. As it can be easily
recognized, the particular form adopted for D
comes from the analysis of truncation error in the
FTUS method. Fig. 2 (taken from Ferreri and
Ambrosini, 1999) shows the predicted loop flow
rate variation with time. The diffusion coefficient
was the one corresponding to 1000 nodes under
the FTUS approximation; the number of modes
considered in the modal expansion was 500. As
may be observed, both approximations behave
similarly. Exact coincidence is precluded by the
non-linearity of the system and small differences
in starting conditions. It may be shown that the
predicted degree of instability is very similar for
the nodal and the modal solution with equivalent
dissipative effects and the agreement is improved
by increasing the number of nodes. The impor-
tance of the second order term alone in determin-
ing the overall truncation error effect on stabilitypredictions is once again emphasized.
To get quantitative information on the effect of
nodalization on the neutral stability boundary of
a given 1-D flow system in Ambrosini and Ferreri
(1998), the linearization of numerically discretized
partial differential equations was adopted. This
methodology shares with the usual techniques
adopted for studying stability of physical systems
(e.g. direct linearization of PDEs or of ODEs
obtained by modal or nodal expansion) the idea
to locally perturb the solution around steady-stateconditions to achieve information on linear sta-
bility, i.e. on stability to infinitesimal perturba-
tions. However, the fact that the numerically
discretized equations are perturbed allows to get
information on stability as predicted by the nu-
merical method, including the effects of trunca-
tion error and discretization in general. This is an
7/29/2019 Flowin St Bility
8/18
J.C. Ferreri, W. Ambrosini/Nuclear Engineering and Design 215 (2002) 153170160
important information when judging about the
acceptability of a given nodalization for transient
stability analyses, as it clearly shows the effect of
numerical discretization in stability maps.
The methodology of analysis addresses the per-
turbations in the variables defining the state of the
considered system. Once the vectors of perturba-
tions of state variables at the nth and (n+1)thtime levels and at each node are defined, a lin-
earized relationship is found between them in the
form:
(ly6
)n+1=(J66 SSn+1)1J6
6 SSn (ly
6
)n (4)
that involves the Jacobeans of the N components
of the vector function representing the numerical
method and the related boundary conditions with
respect to the system state variables. The Ja-
cobeans are evaluated at the steady-state (fixed)
point as a function of grid parameters (in thepresent case the grid step Ds and the time step Dt)
and physical parameters.
In order to set up stability maps, the concept of
a margin in excess to neutral stability is then
introduced. Two different quantities were used in
this purpose:
Dz=z(A66
)1 zR=ln z(A6
6
)
Dt(5)
which are based on the spectral radius z(A66
) of the
matrix A66
=(J66SSn+1)1J
66SSn and are positive for
unstable conditions and negative for stable ones.
The rank of matrix A66
is in the order of the
product of the number of nodes by the number of
independent variables defined at each node.
In a recent paper, Doster and Kendall (1999)
adopted a similar methodology of analysis for
studying single-phase natural circulation flows.
Their technique is based on the assumption of a
complex exponential trend for the space distribu-
tion of state variable perturbations, leading to the
discussion of stability by consideration of the
eigenvalues of a matrix as a function of a wave
number. On the other hand, the methodology
referred to in the present paper is based on the
linearization of the full set of space and time
discretized equations and deals with the overall
amplification or damping of perturbations at each
node.
In the case of Welanders problem (Welander,
1967), the relevant physical parameters are h, the
coefficient corresponding to the buoyancy term
and m, the coefficient corresponding to the friction
term in the momentum equation. Both coefficients
determine the system steady-state conditions and
can be taken as coordinates of a map ofDz or zR.
In this purpose, Dz or zR are calculated through-
out a selected hm rectangular domain, thus iden-
tifying, with the aid of contour plots, regions with
a different degree of stability. This method is easy
to implement in computer programs and has the
advantage to provide a great deal of information,
at the price of a reasonable computing effort. Fig.
3 shows the results obtained for the FTUS
method with 30, 40, 50 and 100 nodes and Dt=
104. The effect of numerical diffusion on the
marginal stability boundary predicted by the first
order explicit numerical method are clear from
the figure and qualitatively justifies the behavior
observed by RELAP5. In relation to the other
numerical methods, it was shown that, as ex-
pected, the implicit first order method is much
more diffusive than the explicit one (as it is in
RELAP5 for the nearly-implicit method with re-
spect to the semi-implicit numerical method) and
Fig. 3. The effect of the number of nodes on the neutral
stability curve using FTUS.
7/29/2019 Flowin St Bility
9/18
J.C. Ferreri, W. Ambrosini/Nuclear Engineering and Design 215 (2002) 153170 161
that second order methods have a superior perfor-
mance in stability analyses, owing to the lack of
the second order derivative term in the modified
equation.
Other effects on the neutral stability boundary
considered in Ambrosini and Ferreri (1998) were
the ones due to the cell Courant number and to
the time discretization of buoyancy and frictionterms in momentum equation. Low values of
Courant number implied increased damping in
the first order explicit method, what is usual in
these finite-differences approximations. Quantita-
tive results in relation to the time discretization of
buoyancy and friction are not so usual (see Gines-
tar Peiro and March-Leuba, 1998; Doster and
Kendall, 1999, for example). In Ambrosini and
Ferreri (1998), it was shown that explicit coupling
between energy and momentum implied a change
in the way of convergence of the neutral stabilityboundary at practical values of cell Courant num-
bers (i.e. $0.8). Implicit coupling allowed recov-
ering the expected diffusive way of map
convergence.
The effect of closure relations was explicitly
mentioned in the Section 1. Ambrosini and Fer-
reri (1997b) dealt with this problem to quantify
their effect on the neutral stability boundary and
to show how the choice of adequate dependent
variables may facilitate this analysis. The analysis
of natural circulation flows in toroidal loops with
imposed heat flux is quite appropriate for this
task. This problem has been studied, both theoret-
ically and experimentally, by several authors con-
sidering different kinds of boundary conditions
(see e.g. Hart, 1984; Sen et al., 1985). Ambrosini
and Ferreri (1997b) considered the specific
boundary conditions addressed by Sen et al.
(1985), involving imposed heat flux heating and
cooling and selected the Reynolds number as
main flow variable in the dimensionless equations
for the toroidal loop making use of three different
frictions laws, including one accounting for a
realistic transition between laminar and turbulent
flow (Churchill, 1977).
Taking advantage of the fact that for this prob-
lem only three ordinary differential equations
fully represent the system dynamics, stability
maps for the spacetime continuous mathemati-
Fig. 4. Effect of friction law on the neutral stability boundary:
modal method vs. FTUS method and 200 nodes (Churchill
friction law with relative roughness 0.0001, torus length/di-
ameter ratio=100, D~=107).
cal problem have been obtained by plotting the
real part of the eigenvalue of the linearized ODE
system having maximum norm, zR
(to be directly
compared with the quantity defined in Eq. (5) for
numerically discretized equations) as a function of
a Grashof number and of the angle of asymmetry
in heating, k. In Fig. 4 it can be noted that the use
of a simple power law for the Fanning factor,
which is the usual choice in literature on ther-
mosyphon loops, gives a very poor representation
of the situation occurring when a realistic law is
used and the regime transition falls in the domain
of dependence of parameters. The curve corre-
sponding to the FTUS scheme with a considerablenumber of nodes shows reasonably converged
results. It is a direct poof of the dramatic influ-
ence of the closure relations, no matter what the
adopted numerical scheme is. It may be argued
that it is an obvious consequence of the modeling
criteria; however, to the authors knowledge, it is
not so usual to find a quantitative determination
of these effects.
The authors also analyzed with the same
methodology the results of Vijayan and Austrege-
silo (1994) for a rectangular thermosyphon loopsobtaining similar results in relation to the effect of
truncation error on stability prediction.
2.3. Numerical analysis of boiling channel
stability
Given the interesting results obtained by the
7/29/2019 Flowin St Bility
10/18
J.C. Ferreri, W. Ambrosini/Nuclear Engineering and Design 215 (2002) 153170162
methodology developed for studying stability of
single-phase thermosyphon loops, an effort was
then spent to extend its applicability to other
instability problems. Among the several ones hav-
ing an interest in engineering, boiling channel
instability plays a key role in nuclear reactor
safety and has been the subject of in depth investi-
gations (see e.g. Wulff et al., 1992; DAuria et al.,1997 among many others).
In order to test the capabilities of the methodol-
ogy when applied to boiling channel stability, a
simple homogeneous equilibrium model was se-
lected for a single boiling channel with imposed
pressure drop, overall heat flux and inlet subcool-
ing (Ambrosini et al., 1999, 2000). The adoption
of this simple model also allowed comparison
with plenty of results available from previous
works. The problem is anyway far more complex
than in the case of the single-phase thermosyphonloops, as in each node three independent variables
must be now defined, being the local pressure,
specific fluid enthalpy and volumetric flux.
Both nonlinear and linear stability conditions
were addressed setting up, as in the case of ther-
mosyphon loops, twin computer programs based
on a same semi-implicit numerical method: a tran-
sient program provides information on nonlinear
behavior starting from selected initial conditions;
a linearized program supplies the basic informa-
tion about linear stability of steady-state flow
conditions. The adopted numerical method makes
use of standard choices in nuclear reactor system
codes as staggered meshes and upwind
differencing.
Fig. 5 shows a stability map obtained as a
function of the phase change and the subcooling
numbers as defined by classical relationships (see
e.g. Ambrosini et al., 2000):
Npch=
P
Whfg
6fg
6f Nsub=
hfhin
hfg
6fg
6f (6)
where P is the channel power, W the mass flow
rate, hin the inlet fluid enthalpy and hf, hfg, 6f, 6fgare the saturated liquid and differential vapor-to-
liquid enthalpies and specific volumes. The map
was obtained for a relevant boiling channel case,
highlighting the regions of density-wave and
Fig. 5. Typical quantitative boiling channel stability map (96
nodes, max Courant number=0.9): the neutral stability
boundary obtained with 48 nodes is also reported for purpose
of comparison (from Ambrosini et al., 1999).
Ledinegg instabilities. The spatial discretization of
the channel includes 96 nodes and a maximum
Courant number of 0.9 was adopted. For purpose
of comparison, the neutral stability boundary ob-
tained with a lower number of nodes is also
reported, pointing out the effect of numerical
diffusion at low inlet subcooling. The map is a
contour plot of the previously defined variable zR,
representing the degree of damping or amplifica-
tion of perturbations as predicted by the numeri-
cal method. This quantity is obviously strictly
related to the predicted decay ratio of perturba-
tions, i.e. the ratio between the amplitude of
oscillations at two subsequent cycles.
By the use of the linear analysis program, it waspossible to set up a number of stability maps
providing information on the sensitivity of stabil-
ity to different physical and numerical parameters
(Ambrosini et al., 1999). As an example of such
Fig. 6. Quantitative boiling channel stability map (12 nodes,
max Courant number=0.9): the neutral stability boundary for
48 nodes is also reported for purpose of comparison (from
Ambrosini et al., 1999).
7/29/2019 Flowin St Bility
11/18
J.C. Ferreri, W. Ambrosini/Nuclear Engineering and Design 215 (2002) 153170 163
analyses, Fig. 6 reports the results obtained for
the same case as in Fig. 5 by decreasing the
number of nodes down to 12. As it can be noted,
the use of a coarse nodalization has a dramatic
effect on the shape of the neutral stability
boundary, giving rise to alternating stripes of
stable and unstable conditions. These changes can
be related to the interaction of the boilingboundary with node boundaries, as it can be
understood considering that the stripes of alter-
nated stability and instability approximately fol-
low lines of constant boiling boundary position,
represented by the relationship:
NsubLbb
L
=
Lbb
LNpch (7)
where L and Lbb are the channel length and the
boiling boundary position.
This is a well-known effect occurring whenusing the homogeneous equilibrium model and
has been reported in previous works (see e.g.
Podowski and Rosa, 1997) as requiring a special
treatment to be mitigated. The stability map in
Fig. 6 gives a very clear and understandable pic-
ture of this effect, capable to justify previous
observations of local behavior in parameter space.
3. Role of automatic differentiation tools
In the previous sections, it was shown how a
methodology of analysis based on numerical dis-
cretization of partial differential equations gov-
erning fluid-dynamic problems may be useful to
get information on the capabilities of numerical
methods in accurately predicting stability. A key
point of this methodology is the evaluation of
Jacobean matrices of the algebraic relationships
characterizing the numerical method and the re-
lated boundary conditions. This is a basic but
somehow boring and error-prone process, which
represents a difficult task to be completed when
dealing with complex models. It is in this perspec-
tive that the use of automatic FORTRAN differenti-
ation tools may play a considerable role.
In this section, as a first step towards this goal,
the sensitivity of stability results to physical and
numerical discretization parameters is analyzed
using a tool for the automatic differentiation of
FORTRAN codes. ADIFOR (meaning Automatic
DIfferentiation of FORtran), version 2.0D that
has been developed by C. Bischof from the ANL,
USA and A. Carle from Rice University, USA
(Bischoff et al., 1994) is the adopted tool that
allows for evaluating the derivatives of model
variables with respect to model parameters. Re-lated bibliography may be found at the ADIFOR
WWW site (ANL, 1999). Some introductory ex-
amples may also be found in the WWW (see for
example ANL, 1999; Goenka, 1999; Rightley,
1997). This is not the only tool available to auto-
matically calculate derivatives and a comparison
between two of them may be found in Rightley
(1997). Its use is becoming widespread to calculate
derivatives of dependent variables with respect to
parameters. Why it is important to know the
derivatives of dependent variables is closely asso-ciated with the construction of response surfaces
to determine the uncertainties of computed re-
sults, as has been clearly stated by Isukapalli
(1999).
ADIFOR is a pre-processor code which, given a
FORTRAN 77 code that computes a function, auto-
matically generates another, augmented, FOR-
TRAN 77 code. It must be considered that any
code may be put in the form of a function, simply
by introducing a call to a main routine after
setting parameter values. The latter computes the
function and the derivatives with respect to a list
of variables. The user must specify the list of
dependent and independent variables. After gen-
erating the augmented code that calculates the
specified derivatives via ADIFOR, the user must
provide a new driving FORTRAN 77 code that
takes into account the new set of variables. Even
when the codes must comply with FORTRAN 77
standards, ADIFOR accepts some common lan-
guage extensions.
Some simple hints to help on the very first use
of ADIFOR may be readily found in Rightley
(1997). Many references document the accuracy of
the derivatives calculated in this way. This is not
the only way to apply ADIFOR; the reader is
referred to the aforementioned literature. The use
ofADIFOR is straightforward. However, the origi-
nal code must be sometimes prepared to get the
7/29/2019 Flowin St Bility
12/18
J.C. Ferreri, W. Ambrosini/Nuclear Engineering and Design 215 (2002) 153170164
Fig. 7. Map ofDz for the first-order, explicit momentum, implicit temperature coupling, cell Courant number 0.8, turbulent flow
and Ds=1/10, see Ambrosini and Ferreri (1998).
appropriate functional dependence between the
independent parameters and the dependent ones.This fact turns essential to take some time to verify
the automatically computed derivatives. One im-
portant feature ofADIFOR is the careful treatment
of derivative exceptions when handling intrinsic
functions. Its reporting capability is also of great
importance.
We are now interested in showing the sensitivity
of the neutral stability boundary to nodalization
and system parameters in natural circulation flows.
It may be argued that the effects of discretization
are known a priori for a given numerical model.However, this knowledge may not be simply
achievable in quantitative terms. In the frame of
such studies, the use of automatic differentiation of
FORTRAN codes may be especially useful. The
results reported here deal again with the pioneering
work of Welander (1967), because its stability
properties have been the subject of the above-men-
tioned analytical and theoretical studies.
The analysis was performed using the dimension-
less volumetric flow rate and temperature as system
variables and is focused on the results obtainedusing the implicit coupling of the momentum and
energy equations and the forward time, upwind-
space finite-difference method (FTUS) for the mo-
mentum equation. It may be argued that the
consequences of using this approximation on the
results are well known, i.e. obtaining a first order
accurate solution in the space and time increments.
However, what is particularly addressed here is the
quantification of these effects (and those of thecoupling) on the stability limits in systems working
in the natural circulation flow regime.
Fig. 7 shows the stability map for Welanders
problem in the plane ofh and m, the system physical
parameters. It has been obtained using 11 nodes
(Ds=1/10), cell Courant number (COU)=0.8 and
a value of 1.75 for the exponent of flow rate in
momentum equation x (defined as 2-b, where b is
the exponent in the friction law, assigned in the
present case to 0.25 as in the Blasius relationship
for turbulent flow in smooth pipes). The selectionof the value 0.8 for COU was not arbitrary: despite
the very low number of nodes, it ensures low
numerical diffusivity while keeping the calculations
stable. The theoretical neutral stability boundary
corresponds to Welanders analysis (Welander,
1967) modified for turbulent flow conditions and is
coincident with the one obtained from modal
analysis (see Ambrosini and Ferreri, 1998).
The interest in the analysis to follow is showing
how the solution depends on the cell Courant
number (COU) and x. The above mentioned set ofparameter values in the dependence domain will be
denoted as ()O. This implies calculating the follow-
ing derivatives:
(Dz
(COU
O
and(Dz
(x
O
7/29/2019 Flowin St Bility
13/18
J.C. Ferreri, W. Ambrosini/Nuclear Engineering and Design 215 (2002) 153170 165
In these cases, the value of the space increment,
Ds, will be fixed; however, Ds was also varied in
turn trying to investigate how the computed
trends confirm the results of the authors previous
analyses on the sensitivity to nodalization. In
order to improve the reliability of the obtained
conclusions, it was decided to check every ex-
ploratory calculation made with the help of ADI-
FOR using finite-difference approximations of the
derivatives; this prudent approach was rewarding,
because it allowed eliminating some reasonable
looking, wrong results.
The influence of x is considered first. Fig. 8
shows the map of (#Dz/#x)O as a function of h
and m considering Ds=1/10. It may be observed
that this derivative is not always negative; this is
due to the shift in the neutral stability boundary
for increasing values of x. Fig. 9 shows the same
map, but considering Ds=1/30. Now the negativearea is more smoothly defined. A reasonably con-
verged solution with respect to nodes number
corresponds to Ds=1/100, which was obtained
also by finite-differences as a further check. In this
case, the zero curve remains unchanged, with
small changes in the values of the derivative.
The change ofDz as a function of COU, (#Dz/
#COU)O, obtained using the same set of parame-
ters, is shown in Fig. 10. Decreasing COU should
lead to further damping in zones of positive
derivatives; this is coherent with the analysis
shown in Ambrosini and Ferreri (1998), where a
set of calculations was made imposing a fixed time
interval thus leading to more damped solutions.
Fig. 11 shows the map obtained assuming Ds=1/
30. Now, a shift in the zero curve towards the
right of the map may be observed. Once again, it
is interesting to resort to converged maps; the
map so obtained showed further bending of the
zero curve, leading to positive and negative zones
in the map.
As already mentioned, the effects of Ds on the
neutral stability boundary were also addressed. In
Ambrosini and Ferreri (1998) work was mainly
devoted to this aspect of the computations and
the effects of different approximations were as-sessed. The derivative of Dz in terms of Ds for
Ds=1/100, COU=0.8 and x=1.75 showed con-
vergence. However, the same derivatives com-
puted using ADIFOR were not coincident. Further
analysis is needed to clearly understand the rea-
sons of this discrepancy, but it is likely that the
discrete nature of the node spacing should impose
a limitation to its computation.
Fig. 8. Sensitivity map ofDz to the friction law exponent (#Dz/#x) for the first-order, explicit momentum and implicit temperature
coupling. [COU=0.8, x=1.75 and Ds=1/10].
7/29/2019 Flowin St Bility
14/18
J.C. Ferreri, W. Ambrosini/Nuclear Engineering and Design 215 (2002) 153170166
Fig. 9. Sensitivity map ofDz to the friction law exponent (#Dz/#x)O for the first-order, explicit momentum and implicit temperature
coupling. [COU=0.8, x=1.75 and Ds=1/30].
4. Conclusions
This paper discussed the general aspects of
natural circulation flows in 1-D systems; and then
focused on some early results by the Authors
concerning the development of calculation tools.
The philosophy adopted in these simulations was
to gain experience from the use of in-house devel-
oped codes, adapted to simple situations, in order
to understand the behavior of large system codes
in predicting fluid-dynamic instabilities.System codes must be robust, in the usual
nomenclature; unfortunately, it sometimes means
that they should produce good results (or, simply,
results) in most cases. Experience shows that
avoiding numerical difficulties usually implies the
use of numerical damping, the basic tenet being
the use of full upwinding. Due to the intrinsic
non-linearity of natural circulation flows, coupled
problems of heat and fluid flow use linearization
and/or averaging that imply changing the time
constant of the computed results, adding damp-
ing. Then, the use of sensitivity analysis tech-
niques or looking for convergence of results in a
brute force way is sometimes rewarding. Time
domain codes may be used to analyze unstable
flows, if an adequate nodalization is used.
The methodology adopted for analyzing the
drawbacks involved in the use of numerical dis-
cretization of partial differential equations while
dealing with stability problems provided a quanti-
tative assessment of the major difficulties occur-
ring in the use of transient codes to this purpose.
Most of the observed qualitative trends were ex-
pected, based on the knowledge about the charac-
teristics of the numerical schemes used. The
quantitative information presented provides a
clear picture of the changes that truncation error
and, in general, discretization may induce in pre-
dictive capabilities about stability. In particular,the results obtained can be summarized as
follows:
first order explicit numerical methods are less
diffusive than first order implicit ones, but
attention must be anyway paid to the choice of
the nodalization and of the time step;
second order methods are better suited for
stability analyses and should be considered for
use in system codes wherever possible;
other details of the discretization, e.g. the cou-
pling in time between momentum and energy
equations, may also have an effect on stability
predictions;
additional discretization effects may appear in
two-phase flow stability problems, as it was
observed for the homogeneous equilibrium
model of a boiling channel; of course, for
two-fluid models with complete non-equi-
7/29/2019 Flowin St Bility
15/18
J.C. Ferreri, W. Ambrosini/Nuclear Engineering and Design 215 (2002) 153170 167
librium the situation is likely to be different
and a specific analysis should be made to clar-
ify the effect of discretization choices on stabil-
ity predictions.
A new development of the research has been
introduced here. It consists in the use of an auto-
matic differentiation tool: ADIFOR, version 2.0D,
which proved to be useful to determine the influ-
ence of parameters on the neutral stability
boundary of natural circulation flows. This is not
the most usual use of this tool, since it involves
tracing the influence of parameters on the spectral
radius of the discrete problem. The results ob-
tained confirmed exactly the numerically simu-
lated derivative maps. In this way, the influence of
the cell Courant number and a variable related to
the exponent in the friction law was assessed.
Experience showed that checking results withfinite-differences might be rewarding to get insight
on the results.
The use of ADIFOR is straightforward when
applied to predictive codes that give a result as a
function of a set of parameters. However, it is
essential to know the details of the code to be
differentiated and to explicitly calculate the
steady-state conditions corresponding to the ad-
dressed fixed point. This is particularly difficult
when dealing with unstable steady-state condi-
tions, which cannot be achieved asymptotically by
transient calculations with constant boundary
conditions.
In this respect, the goal of applying this tech-
nique to large TH systems codes seems a really
challenging one. This is partly due to multiple
sources of derivative discontinuities, coming from
physical correlations and intrinsic functions. Tab-
ular data may pose further difficulty to the analy-
sis, not to speak about numerically induced
oscillations like the ones observed on flow rate.
The task would surely require a huge effort.
At present, the short-term perspective of this
research consists in developing more sophisticated
tools for analyzing, in a realistic way, complex
stability problems making use of the described
numerical methodology. In particular, a code for
analyzing general single-phase single-loop naturalcirculation flows with a variety of boundary con-
ditions is being developed. At the time of writing
(June 2000), the transient program is already
completed and the linearized version will follow
soon, possibly owing to the use of ADIFOR. It
is the goal of the Authors to be capable to pro-
duce in the next future an example of coherent
linear and non-linear stability analysis tools for
natural circulation loops being enough flexible
and realistic to be reliably applied to a variety of
Fig. 10. Sensitivity map ofDz to cell Courant number (#Dz/#COU)O for the first-order, explicit momentum and implicit temperature
coupling. [COU=0.8, x=1.75 and Ds=1/10].
7/29/2019 Flowin St Bility
16/18
J.C. Ferreri, W. Ambrosini/Nuclear Engineering and Design 215 (2002) 153170168
Fig. 11. Sensitivity map ofDz to cell Courant number (#Dz/#COU)O for the first-order, explicit momentum and implicit temperature
coupling. [COU=0.8, x=1.75 andD
s=1/30].
conditions of interest for experiments and
applications.
Acknowledgements
The work summarized here was possible due to
the support of institutions and the collaboration
of many colleagues. However, J.C. Ferreri must
specially mention his early work in collaborationwith Alicia S. Doval at CNEA, Argentina. He is
also deeply grateful by the warm and generous
friendship of the staff of the DIMNP, University
of Pisa, along all these years of informal collabo-
ration. Last but not least, comments and criti-
cisms by referees allowed the authors improving
the referred papers.
References
Ambrosini, W., Ferreri, J.C., 1997a. Numerical analysis of
single-phase, natural circulation in a simple closed loop. In:
Proceedings of11th Meeting on Reactor Physicsand Thermal
Hydraulics, Pocos de Caldas, M.G., Brazil, 1822 August
1997, pp. 676681.
Ambrosini, W., Ferreri, J.C., 1997b. Stability analysis of single-
phase thermo-syphon loops by finite-difference numerical
methods. In: Proceedings of Post SMIRT 14 Seminar 18 on
PassiveSafety Features in Nuclear Installations,Pisa,2527
August 1997, pp. E2.1E2.10. Nuclear Eng. Design, to be
published.
Ambrosini, W., Ferreri, J.C., 1998. The effect of truncation error
on numerical predictions of stability in a natural circulation
single-phase loop. Nuclear Eng. Design 183, 5376.
Ambrosini, W., Di Marco, P., Susanek, A., 1999. Prediction of
Boiling Channel Stability by a Finite-Difference Numerical
Method, 2nd International Symposium on Two-Phase Flow
Modelling and Experimentation, Pisa, Italy, 2326 May,
1999.
Ambrosini, W., Di Marco, P., Ferreri, J.C., 2000. Linear and
non-linear analysis of density-wave instability phenomena.
Int. J. Heat Technol. 18 (1), 2736.
ANL, 1999, Computational Differentiation Technical reports,
Argonne Nat. Laboratory, http://www.mcs.anl.gov/
autodiff/techreports.html.
Barre, F., Parent, M., Brun, B., 1993. Advanced numerical
methods for thermalhydraulics. Nuclear Eng. Design 145,
147158.
Bau, H.H., Torrance, K.E., 1981. Transient and steady behavior
of an open symmetrically-heated, free convection loop. Int.
J. Heat Mass Transfer 24, 597609.
Bau, H.H, Wang, Y.Z., 1992. Chaos: a heat transfer perspective.
In: Tien, L. (Ed.), Annual Review of Heat Transfer, vol. 4.
Hemisphere Publishing Co.
Bischoff, C., Carle, A., Khademi, P., Mauer, A., 1994. TheADIFOR 2.0 System for the automatic differentiation of
FORTRAN 77 programs, ANL/MCS-P481-1194.
Carlson, K.E., Riemke, R.A., Rouhani, S.Z., Shumway, R.W.
and Weaver, 1990. RELAP5/MOD3 Code Manual. Vol. I:
Code structure, System Models, and Solution Methods,
EG&G Idaho, Inc., NUREG/CR-5535, EGG-2596.
Chen, K., 1985. On the oscillatory instability of closed loop
thermosyphons. J. Heat Transfer, Trans. ASME 107, 826
832.
http://www.mcs.anl.gov/autodiff/tech_reports.htmlhttp://www.mcs.anl.gov/autodiff/tech_reports.html7/29/2019 Flowin St Bility
17/18
J.C. Ferreri, W. Ambrosini/Nuclear Engineering and Design 215 (2002) 153170 169
Churchill, S.W., 1977. Friction equations span all the fluid
flow regimes. Chem. Eng. 84 (24), 9192.
DAuria, F., Galassi, G.M., 1998. Code validation and uncer-
tainties in system thermahydraulics. Progr. Nuclear Energy
33 (1/2), 175216.
DAuria, F., Galassi, G.M., Vigni, P., Calastri, A., 1991.
Scaling of natural circulation flows in PWR systems. Nu-
clear Eng. Design 132, 187205.
DAuria, F., et al., 1997. State-of-the-art report on boilingwater reactor stability, NEA/CSNI/R(96)21, OCDE/
GD(97)13, January 1997.
Doster, J.M., Kendall, P.K., 1999. Stability of one-dimen-
sional natural-circulation flows. Nuclear Sci. Eng. 132,
105117.
Doval, A.S, Ferreri, J.C., 1988. A comparison of post-test
calculations using TRAC-PF1 with PMK-NVH experi-
mental results. In: Proceedings of Technical Committee/
Workshop on Computer Aided Safety Analysis, IAEA,
Poland, IAEA-TC-560.02, pp. 7278.
Ferraz Bastos, J.L., 1997. Tema Especial en Termo-hidraulica
(in Portuguese), Report to 11th Meeting on Reactor
Physics and Thermal Hydraulics, Pocos de Caldas, M.G.,
Brazil, 1822 August, 1997.
Ferreri, J.C., Doval, A.S., 1984. On the effects of discretiza-
tion in the computation of natural circulation in loops (In
Spanish). Semin. Argentine Committee Heat Mass Trans-
fer 24, 181212.
Ferreri, J.C, Doval, A.S., 1985. Experience in the Implementa-
tion, Use and Development of Thermal-hydraulic Codes
for Safety Analysis. In: Proceedings of Technical Commit-
tee and Workshop, IAEA, Portoroz, Yugoslavia, 1985,
IAEA-TC-560, pp. 3439.
Ferreri, J.C., Doval, A.S., 1988. On the Effects of Discretiza-
tion in the Time Evolution of Perturbations in a Closed
Loop, presented to the Technical Committee/Workshop on
Computer Aided Safety Analysis, IAEA, Poland, IAEA-TC-560.02, pp. 7278.
Ferreri, J.C., Ambrosini, W., 1999. Verification of RELAP5/
MOD3 with theoretical and numerical stability results on
single-phase, natural circulation in a simple loop, United
States Nuclear Regulatory Commission, NUREG IA/151.
Ferreri, J.C., Ambrosini, W., DAuria, F., 1995. On the Con-
vergence of RELAP5 Calculations in a Single-Phase, Natu-
ral Circulation Test Problem, Proceedings of X-ENFIR,
711 August 1995, Aguas de Lindoia, S.P., Brazil, pp.
303307.
Ginestar Peiro, D., March-Leuba, J., 1998. Sensitivity of BWR
Stability Calculations to Numerical Integration Tech-
niques, Informal report to NURETH, Bethesda, MD,USA, 1920 October.
Goenka, P., 1999, Example derived from a code used at
General Motors Research in friction and lubrication analy-
sis of automotive engines, http://www-unix.mcs.anl.gov/
autodiff/ADIFOR/adifordemo.
Grandi, G.M., Ferreri, J.C., 1991. Limitations of the use of a
Heat Exchanger Approximation for a Point Heat Source,
Internal Memo, CNEA, Gerencia Seg. Rad. y Nuclear,
Div. Modelos Fsicos y Numericos, Argentina.
Hart, J.E., 1984. A new analysis of the closed loop thermo-
syphon. Int. J. Heat Mass Transfer 27 (1), 125136.
Hart, J.E., 1985. A note on the loop thermosyphon with mixed
boundary conditions. Int. J. Heat Mass Transfer 28 (5),
939947.
IAEA, 1998. IAEA Specialist Meeting on User Qualification
for and User Effect on Accident Analysis for Nuclear
Power Plants, Vienna, 31 August3 September, 1998.
Idelchik, I.E., 1966. Handbook of Hydraulic Resistance, Co-efficients of Local Resistance and of Friction, 1960, En-
glish version, AEC-TR-6630.
Ishii, M., Kataoka, I., 1984. Scaling laws for thermal-hy-
draulic systems under single-phase and two-phase natural
circulation. Nuclear Eng. Design 81 (3), 411425.
Isukapalli, S.S., 1999. Uncertainty analysis of transport-trans-
formation models, Ph.D. Thesis, Rutgers, The State Univ.
of New Jersey, New Brunswick, USA.
Mahaffy, J.H., 1993. Numerics of codes: stability, diffusion
and convergence. Nuclear Eng. Design 145, 131145.
Mertol, A., 1980. Heat Transfer and Fluid Flow in Ther-
mosyphons, PHD Dissertation, Univ. of California at
Berkeley.Misale, M., Tagliafico, L., 1987. The transient and stability
behavior of single-phase natural circulation loops, Heat
and Technology, 5 (1-2) 101116.
Podowski, M.Z., Rosa, M.P., 1997. Modelling and numerical
simulation of oscillatory two-phase flows, with application
to boiling water nuclear reactors. Nuclear Eng. Design 177,
179188.
Rightley, M.L.J., 1997. Automatic Differentiation Tools: A
Working Application of ADIFOR and GRESS, Los
Alamos Nat. Laboratory, Los Alamos, NM, USA, http://
csnl-www.lanl.gov/rightley/papers/1dad/diffusion.html.
Sen, M., Ramos, E., Trevino, C., 1985. The toroidal thermo-
syphon with known heat flux. Int. J. Heat Mass Transfer28 (1), 219233.
Stuhmiller, J.H., Ferguson, R.E., 1979. Comparison of Nu-
merical methods for Fluid Flows, Electric Power Res.
Inst., EPRI NP-1236.
Vijayan, P.K, Date, A.W., 1990. Experimental and theoretical
investigations on the steady-state and transient behavior of
a thermosyphon with throughflow in a figure-of-eight loop.
Int. J. Heat Mass Transfer 33 (11), 24792489.
Vijayan, P.K, Date, A.W., 1992. The limits of conditional
stability for single-phase natural circulation with
throughflow in a figure-of-eight loop. Nuclear Eng. Design
136, 361380.
Vijayan, P.K, Austregesilo, H., 1994. Scaling laws for single-phase natural circulation loops. Nuclear Eng. Design 152,
331347.
Vijayan, P.K, Austregesilo, H., Teschendorff, V., 1995. Simu-
lation of the unstable behavior of single-phase natural
circulation with repetitive flow reversals in a rectangular
loop using the computer code ATHLET. Nuclear Eng. De-
sign 155, 623641.
Wu, Cheng-Chi, Almenas, K., 1994. RELAP5 computations
of flow instabilities in a circular toroidal thermosyphon,
http://csnl-www.lanl.gov/~rightley/papers/1dad/diffusion.htmlhttp://csnl-www.lanl.gov/~rightley/papers/1dad/diffusion.htmlhttp://www-unix.mcs.anl.gov/autodiff/ADIFOR/adifor_demohttp://www-unix.mcs.anl.gov/autodiff/ADIFOR/adifor_demo7/29/2019 Flowin St Bility
18/18
J.C. Ferreri, W. Ambrosini/Nuclear Engineering and Design 215 (2002) 153170170
RELAP5 International User Seminar, Baltimore (Mary-
land), 29 August1 September, 1994.
Welander, P., 1967. On the oscillatory instability of a differen-
tially heated fluid loop. J. Fluid Mech. 29 (1), 1730.
Wulff, W., Rohatgi, U.S., 1999. System Scaling for the West-
inghouse AP600 Pressurized Water reactor and Related
Test Facilities-Analysis and results, NUREG/CR-5541 and
BNL-NUREG-52550.
Wulff, W., Cheng, H.S., Mallen, A.N., Rohatgi, U.S., 1992.BWR Stability Analysis with the BNL Engineering Plant
Analyzer, NUREG/CR-5816, BNL-NUREG-52312, Octo-
ber 1992.
Yih-Yun, H., 1994. Some challenges to thermal-hydraulic
codes. Nuclear Eng. Design 151, 103111.
Zvirin, Y., 1981. A review of natural circulation loops in
pressurized water reactors and other systems. Nuclear Eng.
Design 67, 203225.
Zvirin, Y., Greif, R., 1979. Transient behavior of natural
circulation loops: two vertical branches with point heat
source and sink. Int. J. Heat Mass Transfer 22, 499504.
Zvirin, Y., Jeuck, P.R. III, Sullivan, C.N., Duffley, R.B., 1981.Experimental and analytical investigation of a natural cir-
culation system with parallel loops. J. Heat Transfer,
ASME 103, 645652.