abstract

haremboingAI and Robotics

Oct 20, 2013 (3 years and 5 months ago)

70 views

NTHAS
5
:
F
ift
h

Korea
-
Japan

Symposium on Nuclear Thermal Hydraulics and Safety

Jeju
,
Korea

,
November 2
6
-

29
, 200
6

NTHAS5
-
A007



F
UZZY

L
OGIC

R
EPRESENTATION OF FLOW REGIME
M
AP

OF

T
WO
-
PHASE FLOW



J
ae Young Lee
*

and Nam Y
ee

Kwak



School of Mechanical and C
ontrol Engineering, Han Dong Global University

Heunghae, Pohang,
K
yungbuk, Korea

E
-
mail :
jylee7@handong.edu




ABSTRACT


A method to represent the flow regime map of two
-
phase flow in the way of fuzzy logic is developed. The
conventional

flow regime tra
nsition criteria have been developed in the way of crisp logic.
H
owever, as noted
the regime transition is gradually made so near the boarder of the regimes, the constitutive
relation
s have been
determined through interpolation method.
The key idea of the
present method is to determine the interpolation
me
thod based on the fuzzy logic.
The fuzzy logic is constructed by the objective neural
networks

which can
remove the observer

s subjective opinion. The excitement level of the output nodes of neural network
s are the
measure of the confident level of a certain flow regime map. In the
present

study, the self organized neural
network is employed for the purpose. The impedance signals for the void fraction
are transformed into the
cumulative

probability density
function and input to the neural network. A

flow test facility of
two pipes
with
inner diameter of 2
5.4
mm and
50
.8mm

is
constructed

to simulate vertical two
-
phase flow and harness the time
sequential void fraction data.
The present method successfully cons
truct three dimensional map by adding an
axis of the confidential level on the plane of liquid and gas superficial velocities. It was found that the results
showed a good agreement of the
conventional

Mishima
-
Ishii Criteria.
Also, it is natural to distingu
ish the
discrete bubble and cap bubble also stable slug and unstable slug.
I
t may be said that the
present

representation
may
replace

the current relaxation method in the determination of the interfacial
transfer

terms in the safety
analysis code.



1. IN
TRODUCTION


The
water cooled nuclear power plants including pressurized
water reactor, boiling water reactor, and pressurized heavy
water reactors are often in the condition of two
-
phase flow
during normal operation and hypothetic incident conditions.
Its

deterministic safety has been analyzed by the safety codes
furnished by the two
-
fluid model in which
gas and liquid are
separately treated with the constitutive
relations

to describe
the interfacial transfer terms. The merit of two
-
fluid
model
(Ishii 1975)

is the general capability to simulate all flow
conditions including counter current flow which cannot be
simulated in the
homogeneous

equilibrium
model and the drift
flux model.

The interfacial transfer depends highly on the
flow patterns, the shape of i
nterface in two
-
phase flow. The
complicated relation of the flow regime and related
constitutive models could be a cause of spurious instability
when the flow condition is near the boarder of two flow
regimes. If the flow regimes are changed in each time
step in
numerical simulation, all interfacial transfer terms are
changing suddenly and it maybe a cause of unrealistic
instability and mislead the results into the unphysical situation.
Therefore, safety analysis codes developed their own way of
relaxatio
n in determination of interfacial transfer terms. For
instant, RELAP5
-
MOD3 has a special logic in determination
of the interfacial area

concentration
.
I
n the bubbly flow regime,
it is determined by the critical Weber number and Sauter mean
diameter of the

discrete bubble. However, for the slug flow
regime, it is determined by
interpolating

those of bubbly and
annular flow regimes with the function of exponential.
However, the results showed a sharp change of
IAC(Interfacial area
concentration
) near the tr
ansition
criterion for bubbly
-
slug ( void fraction =0.3) as shown in
Fig.1.




Fig.
1
. The interfacial area concentration predicted by
the RELAP5
-

MOD3 with the experimental data at
2 m/sec.


RELAP find the way to determine th
e IAC in the slug flow
regime on the behalf of those in the bubbly flow regime and
annular flow regime. However, it is not experimentally
verified yet. C
onsidering the uncertainties lie in the interfacial
transfer term, it maybe acceptable but to enhance t
he
prediction capability we need to construct more realistic
method to determine it. Not only IAC, the logic for flow
regime map
highly affects on
the other terms such as the
interfacial drag force, lift force, all transient forces, mass
transfer for boili
ng and condensation, and heat transfer
.


Therefore, at least to relax such sudden change of the
interfacial transfer terms near the boarder of competing flow
regimes, the present study was
performed

to find the
confidence level of competing flow regimes in

a certain flow
condition. The confidence level of a certain flow regime can
be functionized in the plane of flow regime map with the
superficial gas and liquid velocities and it is correspondent
with the membership
functions

in fuzzy logic( ). The
fu
zzy logic is different from the
conventional

crisp logic in
which only the member of 100% confidence are allowed, so
even though a flow condition has 70% confidence of
membership of A flow regime and 30% confidence of
memebership of B flow regime, it is co
unted as the A flow
regime with 100% confidence. It should not be constructed in
the
arbitrary

ways, it should be
constructed

based on the real
experimental

data and in the way of objective to remove out
the diversity among the
researchers
.
In the

present
paper, we
employed a neural network.
For the feedforward neural
network, the output nodes are excited differently, so we can
produce the confidence level
according

to the degree of
excitement (Lee and Ishii, 2006), however, there is a chance of
subjective
in the selection of the
reference

data for training the
neural network. Therefore, in the present study, the self
-
organized neural network is employed
because

it can learn by
itself so the chance of designer

s en
gagement could be ruled
away.



2.
Methodolo
gy of flow regime identification



Noted the subjective character in determination of flow regime,
Mi et al developed a way by adopting the power of the
artificial neural network and input the statistical characteristics
of the time sequential data of cro
ss sectional void fraction.
They selected two or three variables such as mean, deviation,
skewness as inputs. Lee et al.(2003)

improve the method by
input the sorted data directly as shown in F
i
g.2. It identifies
the flow regime using data measured in ver
y short period time.
Therefore, it can be called an instantaneous and objective
method

. In the present section, brief introduction of the
method and the self org
anized neural network is made.



Fig.
2
. The schematic diagram of the supervised
neural netwo
rk for the instantaneous and objective
flow regime identification



2.1
D
ata harnessing and
preprocessing


2.
1
.
1 Experimental Facilities


In the present study, in order to produce the time sequential
data of cross
sectional

void fraction, a vertical flo
w loop is
constructed with the impedance meter. Fig.3 shows the picture
of the test facility. There are two pipes with diameters of 25.4
mm and 50.8mm. The mixing chamber with three directional
immersed jet nozzles mixes air and water to homogenize the
mix
ture of air
-
water.




Fig.3 the picture of test loop for the flow regime identification


.

A
s a sensor to measure the time sequential data of void
fraction, the present test facilities equips impedance meter. In
order to
enhan
ce

the
quality
of the data, we design the electric
circuit based on the constant current supply.





2.
1
.
2
Cumulative probability density function


The initial and final value of the time sequential data as
shown

in Fig.4(a) can be changed in each observat
ion. Therefore
there is a chance of confusion to the neural
network
. Therefore,
the construction of the probability density function, Fig. 4(b),
gave us more solid figure. However, pdf needs to sati
sfy

the
statistical requirement. As an alternative, the
cu
mulative

pdf,
the sorted data according to its magnitude, can be a index to
identify the flow regime. As shown in Fig.4(c), only
few
number of data produce the
pdf which will be input to the
neural
network
.




Fig.
4
. The signal characteristics for the c
ap bubbly flow
(a) The void
-
impedance signal; (b) The probability
density function for 60 seconds data observation; (c)
The probability distribution function for 1 second data
observation




2.2
Self organized neural network and voting to determine
confide
nce level


2.2.1
Self organized Neural network


A neural network with two layers is employed as a self
organized neural
network

as shown in Fig. 5. T
he self
organized neural network has been developed




Fig.
5
.
Structure of

the Kohonen self organized ne
ural
network

which has only two layer without any
supervising in traing.



2.2.2
Determination of the membership



The feed forward neural network quantifies confidence level
of output nodes, so procedure to construct membership is
relatively easy. Howe
ve
r, SOM notifies only the winner node
.
Therefore, in the present study we determine the confidence
using the voting method which determines the probability of
appearance

of a
certain

flow regime in a
certain

period of
observation

period. For instance, we ca
n vote 60 times from
the data of one minute when we divide data in one second
period as shown in Fig.6. The flow patterns identified has a
certain shape of distribution and it can be used as
the the
membership function


Fig.
6
.
Voting method to determine
the confidence level.


After finding such a confidence level, we use the linear
regression method to
determine

the fuzzy
boundary

with the
98% confidence level in the assumption that the transition has
the linear
dependency

between the
superficial

liquid a
nd gas
velocities.





3.
RESULTS AND DISCUSSIONS


T
he time sequential data
from the

test rig are used to
determine the flow regime map and fuzzy
representation in the
three dimensional space




3
.1
Fuzzy Flow regime map for the
vertical

upward flow in
th
e pipe with ID of 25.4mm



The present method is applied to the data of the coccurent
vertical two phase flow in the pipe with ID of 25.4mm. Since
we set up five output nodes, the present Kohonen Neural
network

classified five flow regimes: discrete bubbl
y flow,
cap bubbly flow, slug flow, churn flow, and annular flow.
As
shown in Fig. 7, flow regime map is represented in the space
of the
superficial

gas and liquid velocity and confidence level.






Fig. 7 Three dimensional fu
zzy
representation

of the flow
regime map of vertical upward two

phase flow



Figure 8 showed the overlapped surface of the fuzzy
membership

functional flow regime map.
I
t represents the
confidence

level change near the boarder of the flow regimes.
Fig. 8(
b) showed the contour map which is correspondent
with
the

traditional flow regime map. It showed clearly the
transition regions in between flow regimes. Previously these
transition area has been
presumed

arbitrarily but it is the first
work to identify the

transition area based on the experimental
data and the objective decision method.




Fig. 8 The surface of the flow regime map and the contour
map of the fuzzy flow regime map


In the F
i
g. 9, comp
arison of the present result with the
previous work is performed. The well known Mishima
-
Ishii
model shows a good agreement with the present work.
The
transition of bubbly
-
and
-
slug is in side of the transition area
between the cap bubbly and slug flow regi
me. It has been
known that traditional flow regime identification has been
made that bubbly flow includes cap bubbly flow as noted by
Titel and Dukler. We also identify the transition of stable slug
and unstable slug flow. As for the churn flow regime, in
the
fast superficial liquid velocity, we cannot clearly determine
the well known argument on the definition of churn: fully
agitated slug of Ishii and flooding of Jianti and Hewitt.



Fig. 9 The transition area in the map and comparison with
the Mishima
-
I
shii transition criteria for 25.4mm ID pipe.



3
.1
Fuzzy Flow regime map for the
vertical

upward flow in
the pipe with ID of 50.8mm


Figure 10 showed the membership functions for each flow
regime for 50.8mm ID pipe. Five flow regimes were
successfully

converted into the fuzzy membership
function

with the present method. .





Fig.
10

Three dimensional fuzzy
representation

of the flow
regime map of
50.8mm ID pipe


Figure 11 showed the overlapped surface of the fuzzy
members
hip

functional flow regime map.
I
t represents the
confidence

level change near the boarder of the flow regimes.
Fig. 11(b) showed the contour map which is correspondent
with the

traditional flow regime map. It showed clearly the
transition regions in betwe
en flow regimes.






Fig.
11

The surface of the flow regime map and the
contour map of the fuzzy flow regime map

for 50.8 mm ID
pipe



In the F
i
g. 12, comparison of the present result with the
previ
ous work is performed. The well known Mishima
-
Ishii
model shows a good agreement with the present work. The
transition of bubbly
-
and
-
slug is in side of the transition area
between the cap bubbly and slug flow regime. It has been
known that traditional flow

regime identification has been
made that bubbly flow includes cap bubbly flow as noted by
Titel and Dukler. We also identify the transition of stable slug
and unstable slug flow. As for the churn flow regime,
the
present flow regime map identify its trans
ition in a little bit
small

gas superficial velocity than Mishima
-
Ishii requirement.




Fig.
12

The transition area in the map and comparison
with the Mishima
-
Ishii transition criteria for
50.8
mm ID
pipe.



4.

CONCLUSIONS



In the present study, a syst
ematic method to construct a fuzzy
membership
function

of flow regime which is capable of
determining realistic interfacial transfer rate by fuzzy logic
interpolation of competing values of flow regimes related. For
this, the time
sequential

impedance data

were harnessed from
the vertical upward flow loop




ACKNOWLEDGEMENT
S


The

work is supported by BAERI(Basic Atomic Energy
Research Institute) in the program of
Nuclear research of
Ministry of Science and Technology of Korean Government.


NOMENCLATURE


j

superficial velocity [m/sec]




[
m
2
]


Greek Letters



void fraction


Subscripts

g



gas
-
phase

f
liquid
-
phase


REFERENCES


[1] M. Ishii, Thermo
-
fluid dynamic theory of two
-
phase flow,
Eyrolles, Paris, 1975.

[2] M. Ishii, Objective characterization of interfacial structure
in two
-
phase flow, Keyno
te Speak, NURETH
-
10, Seoul, 2003,
KL
-
01.

[3] P. Griffith, G.B. Wallis, Two
-
phase slug flow, Journal of
Heat Transfer 83 (1961) 307
-
320.

[4] Y.
Taitel, D. Barnea, A.E. Dukler,
Modeling flow pattern
transitions for steady upward gas
-
liquid flow in vertical t
ubes.
AIChE Journal 26 (1980) 345
-
354.

[5] R.C.
Fernandes, R. Semiat, A.E. Duckler,
Hydrodynamic
model for gas
-
liquid slug flow in vertical tubes, AIChE
Journal 32 (1983) 981
-
989.

[6] A.R. Hassan, C.S. Kabir, Two
-
phase flow in vertical and
inclined annuli.
, International Journal of Multiphase Flow 18
(1992) 279
-
293.

[7] M.A. Vince, R.T. Lahey, On the development of an
objective flow regime indicator. International Journal of
Multiphase Flow 8 (1982) 93
-
124.

[8] O.C. Jones, N. Zuber, The interrelation betwee
n void
fraction fluctuations and flow patterns in two
-
phase flow,
International Journal of Multiphase Flow 2 (1975) 273
-
534

[9] Y. Mi, M. Ishii, L.H. Tsoukalas, Flow regime
identification methodology with neural networks and two
-
phase flow models, Nuclear
Engineering and Design 204
(2001) 87
-
100.

[10] Y. Mi, M. Ishii, L.H. Tsoukalas, Vertical two
-
phase flow
identification using advanced instrumentation and neural
networks, Nuclear Engineering and Design 184 (1998) 409
-
420.

[11] J.Y. Lee, N.S. Kim, M. Ishii,

An instantaneous flow
regime identification using probability distribution function
and feed forward neural network, Proceedings of 10th Int.
Topical Meeting on Nuclear Reactor Thermal Hydraulics
(NURETH
-
10), Seoul, 2003, A00309.

[12] K. Mishima, M. Ishii
, Flow regime transition criteria for
upward two
-
phase flow in vertical tubes. International Journal
of Heat and Mass Transfer 27 (1984) 723
-
737.

[13] H. Goda, S. Kim, Y. Mi, J. Finch, M. Ishii, J. Uhle, Flow
regime identification of co
-
current downward tw
o
-
phase flow
with neural network approach, Proceedings of 10th
International Conference of Nuclear Engineering (ICONE
-
10),
Arlington, VA, 2002, 22088.

[14] D. Barnea, O. Shnam, Y. Taitel, Flow Pattern transition
for vertical downward two
-
phase flo
w, Chemical Engineering
Science 37 (1982) 741
-
744.

[15] K. Usui, Vertically downward two
-
phase flow, (II) Flow
regime transition criteria, Journal of Nuclear Science and
Technology 26 (1989) 1013
-
1022.