Sunteți pe pagina 1din 14

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/267579000

A Standard Method to Determine Loss Coefficients of Conduit Components


Based on the Second Law of Thermodynamics

Conference Paper · July 2012


DOI: 10.1115/ICNMM2012-73249

CITATIONS READS

9 363

2 authors:

Bastian Schmandt Heinz Herwig


Audi AG Technische Universität Hamburg
43 PUBLICATIONS   167 CITATIONS    576 PUBLICATIONS   2,897 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Devaluation of Energy in Various Flow Situations View project

All content following this page was uploaded by Heinz Herwig on 01 December 2015.

The user has requested enhancement of the downloaded file.


Proceedings of the ASME 2012 10th International Conference on Nanochannels,
Microchannels, and Minichannels
ICNMM2012
July 8-12, 2012, Rio Grande, Puerto Rico

ICNMM2012-73249

A STANDARD METHOD TO DETERMINE LOSS COEFFICIENTS OF CONDUIT


COMPONENTS BASED ON THE SECOND LAW OF THERMODYNAMICS

Bastian Schmandt; Heinz Herwig


Institute of Thermo-Fluid Dynamics
Hamburg University of Technology
21073 Hamburg, Germany

ABSTRACT NOMENCLATURE
Losses in conduit components of a pipe system can be ac- a m channel width
counted for by using component specific loss coefficients K. Es- Ĉ1 , Ĉ2 , m — fitting constants for K(Re)-ansatz
pecially in mini- and micro-systems an exact knowledge of these Dh m hydraulic diameter
loss coefficients (which in laminar flow strongly depend on the ĖL W exergy loss rate
Reynolds number) is important. Limited space will generally f — friction factor
lead to a high loss-contribution of single components compared g m s−2 gravitational acceleration
to the contribution of the straight channels. The determination K — head loss coefficient
of K-values of single components based on a numerical simula- KE — exergy loss coefficient
tion using the Second Law Analysis (SLA) has turned out to be Lu , Ld m upstream/downstream length
a very attractive method. The simulation of the flow field shows LVu , LVd m length of computational domain
the distribution of losses and upstream and downstream lengths L̂u , L̂d m upstream/downstream position of
of impact (Lu , Ld ) where the otherwise fully developed flow is pressure tapping
affected by the component. The numerical SLA-Method is in- ṁ kg s−1 mass flow rate
troduced as a standard method, illustrated and validated with n — exponent in force-velocity-
highly accurate measurements in a 90 deg bend with a square correlation
cross section. The local entropy generation rates based on the ∆p Pa pressure difference
numerical simulation of the flow field are computed and care- ∆piJ Pa pressure difference between up-
fully interpreted. Component specific values of K, Lu are Ld are stream and downstream tapping
collected in a table and illustrated by plots of the entropy gen- ∆p jk Pa pressure difference of fully devel-
eration rate distribution along the bend’s centerline. Validation oped flow between two upstream
is achieved with experimental results from a test facility exclu- tappings
sively built for this purpose: Laminar flow in a 90 deg bend is Re — Reynolds number
induced by a controlled gear pump with polydimethylsiloxanes Ṡ W K−1 entropy generation rate
of different viscosities as working fluids. t s time
T K,◦C temperature

1 Copyright
c 2012 by ASME
Tm K,◦C mean temperature and heat transfer processes is the quantitative measure of the
um m s−1 cross section averaged velocity losses involved and can be used to determine them.
u∆p Pa uncertainty of pressure measure- Since entropy is quantified and balanced by the second law
ment of thermodynamics the determination of losses based on an anal-
uDh m uncertainty of hydraulic diameter ysis that takes into account the entropy generation is called Sec-
uL̂ m uncertainty of pressure tapping lo- ond Law Analysis or with the popular three letter acronym the
cation SLA-Approach.
uṁ kg s−1 uncertainty of mass flow rate The physical background of this approach can be found in
uν mm2 s−1 uncertainty of viscosity standard thermodynamics textbooks like [1,2,3]. The systematic
uD — uncertainty of experimental data consideration of entropy generation in conjunction with losses in
unum — numerical uncertainty flow and temperature fields goes back to early studies by Bejan,
uinp — uncertainty of model inputs like [4], see also the comprehensive presentations in [5, 6]. Later
uval — validation uncertainty these ideas were adopted and developed further towards an as-
sessment strategy and the already mentioned SLA-approach was
Greek letters applied to a wide variety of problems, see for example [7, 8, 9].
α — correction factor for kinetic energy Here, we want to establish the SLA-approach as a standard
in a one-dimensional model method for the determination of losses in an internal flow field of
ϕ J kg−1 specific dissipation single conduit components. For that purpose we
∆ϕ J kg−1 additional specific dissipation
Φ W dissipation rate 1. demonstrate, how to apply it
ν m2 s−1 kinematic viscosity 2. apply it to a 90 deg bend as a benchmark case
µ kg/(ms) dynamic viscosity 3. validate the SLA-approach for this benchmark case
ρ kg m−3 density
τ s time constant of pressure measure- These three steps are the content of the next three chapters.
ment

Subscripts and superscripts LOSS COEFFICIENTS AND THE SLA-APPROACH


0 1/m value per length From a fluid mechanics point of view single conduit com-
000 1/m3 value per volume ponents of an internal flow path like straight channels, bends,
(fictitious) undisturbed flow trijunctions, channel entrances, and exits can be characterized by

within a component so-called head loss coefficients. In standard textbooks like [10]
c
downstream they are introduced as
d
u
upstream
0
ambient conditions
(i, j,k)
upstream position index b ≡ 2∆p
K (1)
downstream position index ρu2m
J

with ∆p as pressure drop due to the conduit component and


INTRODUCTION
ρu2m /2 as dynamic pressure in the flow field. Here the symbol K b
Losses in a flow field, from a thermodynamic point of view
is used, since later on we will introduce an alternative definition
are losses of exergy1 , also called available work. By these losses
of a head loss coefficient K. When a straight channel of length L
energy is devaluated through an internal conversion of exergy
and with a hydraulic diameter Dh is treated as such a component
into anergy2 . In a fluidic system this may happen in a flow field
due to a dissipation process or in a temperature field due to heat
conduction with finite temperature gradients, as part of an irre-
versible heat transfer process. b =f L
K (2)
This energy devaluation manifests itself in an increase of en- Dh
tropy by its generation. Thus, entropy generation in such flow
is replaced by the friction factor f.
K
b and f are widely used in technical applications of all kinds,
1 exergy: the maximum theoretically obtainable work from the energy inter- though their definition is slightly problematic. Specific data for
acting with the environment to equilibrium K
b and f can be found in comprehensive collections like [11], [12],
2 anergy: energy-exergy

2 Copyright
c 2012 by ASME
[13] and in the famous Moody chart, see [14] as far as the friction Φ000 (in W m−3 ) follows from the mechanical energy differential
factor f is concerned. equation for the flow and reads, see [3]:
Without exception they are based on experimental data, of-
ten collected many decades ago like the still widely used data
of [15] for sand roughened pipes or the data of [16] for pipes with
"  #
∂u 2
  2 
000 ∂v ∂w 2
a special regular roughness called Loewenherz thread. Losses Φ =µ 2 + +
∂x ∂y ∂z
in commercial pipes have been reinvestigated in [17], for exam-
ple, where differences between the Moody chart results and the 2  2   !
∂v ∂w 2

∂u ∂v ∂u ∂w
highly accurate new measurements have been found. + + + + + + (5)
∂y ∂x ∂z ∂x ∂z ∂y
The pressure drop ∆p in (1) does not correspond to the ex-
ergy losses in general. Indeed it can be used only in those cases
in which it corresponds to the head loss in the flow. Here head The overall specific dissipation ϕ then follows from the in-
loss means loss of mechanical energy (in favour of internal or tegration of (5) referred to the mass flow rate ṁ through the com-
thermal energy), i.e. loss of available work (=exergy, see above). ponent. When all dissipation occurs in the volume Vc of the com-
Head loss in a flow field occurs due to dissipation of mechanical ponent the integration is straight forward in Vc . When the com-
energy. Therefore a thermodynamically motivated definition of ponent affects the upstream and downstream flow, however, the
a head loss coefficient should be additional dissipation outside of the component must be deter-
mined and added to the value found in Vc , see [18].
The local dissipation rate Φ000 is closely linked to the local
2ϕ entropy generation rate Ṡ000 by
K≡ (3)
u2m

with ϕ being the specific dissipation associated with the conduit Φ000
Ṡ000 = (6)
component for which K holds. As will be discussed afterwards Tm
in more detail, the dissipation does not necessarily occur within
the conduit component but may in parts be located upstream and and to the local exergy loss rate ĖL000 by
downstream of the component. This fact is well known and has
often been reported. If, however, the exact distribution of losses
should be determined the exact flow field upstream and down- T0 000
ĖL000 = T0 Ṡ000 = Φ (7)
stream of the component must be known. This is a major concern Tm
of our study.
When the dissipation occurs between two cross sections 1 Here Tm is the mean temperature in a cross section of the
−1
and the specific dissipation ϕ (in W/(kg/s) = J kg ) can be
2 conduit component, whereas T0 is the environmental tempera-
approximately found from the one-dimensional mechanical en- ture.4
ergy equation, see [3], to be Only for the isothermal case with Tm = T0 the local dissi-
pation rate Φ000 directly corresponds to the local exergy loss rate
ĖL000 . When the flow occurs on a temperature level Tm 6= T0 the
p1 − p2 α1 u2m1 − α2 u2m2 amount of lost exergy depends on the ratio T0 /Tm .
ϕ12 = + + g(y1 − y2 ) (4)
ρ 2 Based on these considerations, from a thermodynamics
point of view two parameters should be introduced in order to
Obviously the frequently used definition (1) for the head characterize flow losses of internal flows:
loss coefficient K b with ∆p = p1 − p2 is one for the special sit-
uation in which α1 u2m1 = α2 u2m2 and y1 = y2 , i.e. for a flow with
no change in the kinetic and potential energies between the two 2ϕ 2Tm
Z
K≡ = Ṡ000 dV (head loss coefficient) (8)
cross sections. This is the case for a fully developed horizon- 2
um ṁ u2m
tal pipe flow, but not for a vertical one (y1 6= y2 ), for a devel- Ṽc
oping one (α1 6= α2 ) or for a pipe with varying cross sections 2ĖL T0
(α1 u2m1 6= α2 u2m2 ), for example. KE ≡ 2
= K (exergy loss coefficient) (9)
ṁ um Tm
The exact amount of the local volumetric dissipation rate3

4 Taking the mean temperature in (6) and (7) neglects the small effect of the
3 The influence of the so-called second or bulk viscosity is neglected here, temperature distribution that may exist in the cross section.
since a completely incompressible (model) fluid is assumed.
3 Copyright
c 2012 by ASME
Here the symbol Ṽc means that the integration comprises the TABLE 1. REYNOLDS NUMBER DEPENDENCE OF THE K-
entropy generation in the component as well as the additional en- VALUE FOR LAMINAR FLOW DEDUCED FROM A GENERAL
tropy generation due to the component upstream and downstream BALANCE OF FORCES; S: SMOOTH, R: ROUGH
of it.
With the coefficients K and KE together, a more comprehen- Re: low moderate high
sive measure of flow losses exists than with the special definition
cond. comp. Re−1 Ren−2 const
of K b according to (1) alone.
Formally K might also be determined through an integration
S Re−1 Re−1 Re−1
of Φ000 instead of Ṡ000 , see (6). We prefer Ṡ000 , however, since then channel
Ṡ000 can be supplemented by the corresponding entropy genera- R Re−1 Re−1 Ren−2
tion in a temperature field when convective heat transfer situa-
tions are analyzed, see [7] for further details.
All these considerations hold for conduit components of For a general discussion of conduit components and straight
macro and micro size alike. The only difference is that macro channels let us assume
sized conduit components will often occur in flows with rather
high Reynolds numbers whereas the micro sized components
will be operated at low Reynolds numbers. Therefore, in the mi-
F ∝ unm with 1≤n≤2 (12)
cro range laminar flows will prevail whereas turbulent flow will
be the standard situation in macro sized flow systems.
In the following we only deal with small or moderately large so that with Re according to (10) and Dh = 4A/P, A being the
Reynolds numbers which are typical for flows in micro sized sys- cross section with perimeter P
tems and thus restrict ourselves to laminar flows. The extension
to higher Reynolds numbers and thus to turbulent flows can be
found in found in [18]. K ∝ Ren−2 . (13)
With the Reynolds number Re defined as

In the common notation of the NAVIER -S TOKES equations


ρum Dh inertia forces appear on the left hand side, where the only non-
Re = (10) linear terms are (assuming constant properties for the moment)
µ
so that inertia forces are non-linear forces (∝ u2m ). The force bal-
ance will thus lead to a friction force F ∝ unm with n 6= 1 whenever
(um : mean velocity in a cross section with the hydraulic diameter forces appear in the balance that come from the left hand side of
Dh ) a priori considerations can be made about the dependence of the NAVIER -S TOKES equations.
K on the Reynolds number Re by considering the Navier-Stokes With these considerations in mind we can distinguish three
equations. options for n, which are all found in table 1. Here, Reynolds
From a continuum point of view, they basically are N EW- numbers are characterized as low, moderate, and high. Since
TON ’s second law applied to an infinitesimal mass dm = ρdV and straight channels may also be components in a system they are
thus constitute the balance of forces with respect to dm. Forces included. They need, however, an extra treatment since in chan-
involved are inertia forces, pressure forces, buoyancy forces, and nels wall roughness often has a strong influence (see [8] for fur-
“friction forces” F, related to dissipation. For internal flows ther details). The influence of roughness in conduit components
these friction forces F correspond to the nominator in the def- may also be accounted for. Here, we neglect it, however, since in
inition of the head loss coefficient, see (1) and (8), respectively. most cases dissipation in the bulk of the flow dominates.
Thus K is The three options are:
1. n = 1, i.e. K ∝ Re−1 :
F The left hand side of the NAVIER -S TOKES equations van-
K = const (11)
u2m ishes completely. This is the case
• for conduit components when the flow is laminar with
According to this equation, K is a constant only when F ∝ um → 0, which then is a creeping flow (low Reynolds num-
u2m . In all other cases K depends on um , which in a nondimen- ber)
sional theory corresponds to a Reynolds number dependence of • for straight channels with smooth walls when the flow is
K. laminar with gradu = 0, which then is a fully developed

4 Copyright
c 2012 by ASME
flow, or with rough walls but only for low and moderate

upstream channel part Vu


Reynolds numbers.
2. n = 2, i.e. K ∝ Re0 = const :
The inertia forces dominate and thus are effectively the only

95% of ∆ϕu
forces in balance with F. This is the case for conduit compo-
nents when the flow is laminar with high Reynolds numbers.

Lu :
3. 1 < n < 2, i.e. K ∝ Ren−2 :
Inertia forces are present together with other forces. This is downstream channel part Vd
the case Vc Ld :
90 deg 95% of ∆ϕd
• for conduit components with moderate Reynolds numbers bend
• for straight channels with rough walls when the flow is
Z
laminar at high Reynolds numbers, see [9].
Z Z
Ṡ = (Ṡ000 − Ṡ◦000 )dV + Ṡ000 dV + (Ṡ000 − Ṡ◦000 )dV
Vu Vc Vd
K ∝ Re−1 in laminar flow at low Reynolds numbers and K =
|{z} | {z } | {z } | {z }
ϕ ṁ/Tm ∆ϕu ṁ/Tm ϕc ṁ/Tm ∆ϕd ṁ/Tm
const at higher values of the Reynolds number (1000 and above)
has been found experimentally in [19] for a selection of bends FIGURE 1. DETERMINATION OF THE OVERALL ENTROPY
and elbows. For conduit components with a large wake (e.g. a GENERATION RATE DUE TO A 90 DEG BEND
composition disk global valve) K ∝ Re−1/2 according to [19] for ∆ϕu : ADDITIONAL SPECIFIC DISSIPATION UPSTREAM OF THE
Reynolds numbers as low as 50. However, this Reynolds number COMPONENT
dependence can be expected to change to K ∝ Re−1 when the ϕc : SPECIFIC DISSIPATION IN THE COMPONENT
flow becomes a creeping flow. For smooth components this K ∝ ∆ϕd : ADDITIONAL SPECIFIC DISSIPATION DOWNSTREAM OF
Re−1 behaviour is found up to rather high Reynolds numbers. THE COMPONENT
Table 1 shows that K may be a constant only for high
Reynolds numbers in a conduit component.
the light shaded areas represent the additional losses outside the
bend. Thus, these light shaded areas are located at values of
THE SLA BENCHMARK CASE: 90 DEG BEND Ṡ0 /Ṡ◦0 > 1 since the entropy generation of fully developed flow
With the head loss coefficient K according to (8) associated far upstream and downstream of the bend and the fictitious devel-
with a 90 deg bend, the overall entropy generation caused by this oped flow near the bend corresponds to Ṡ0 /Ṡ◦0 = 1. The situation
component is accounted for. This comprises the entropy genera- of fully developed flow is illustrated by a black line at Ṡ0 /Ṡ◦0 = 1
tion within the component itself as well as the two contributions in the figure. Since at the end of the computational domain val-
upstream and downstream of the component. Schematically this ues of Ṡ0 are very close to the values Ṡ◦0 for undisturbed flow, the
is sketched in figure 1. Here Ṡ◦000 is the local entropy genera- size of the domain is regarded to be sufficient for Re = 512 and
tion rate in the fictitious undisturbed adjacent channels so that for the smaller Reynolds numbers.
(Ṡ000 − Ṡ◦000 ) is the local impact of the 90 deg bend in terms of the Once Ṡ according to the integration of Ṡ000 in figure 1 is
additional entropy generation rates. We quite generally define the known, the K-value follows immediately, c.f. (8). The local en-
lengths Lu and Ld as those lengths of influence within which 95% tropy generation rate in a three dimensional flow field according
of the additional entropy generation occurs. These lengths are in- to (5) and (6) reads:
troduced to characterize how far the component impact goes, but
do not fix Vu and Vd , the volumes in which the additional en-
tropy generation is determined. Due to the asymptotic decay of
"      #
000 µ ∂u 2 ∂v 2 ∂w 2
the additional entropy generation, Vu and Vd would be infinitely Ṡ = 2 + +
Tm ∂x ∂y ∂z
large. They are, however, kept finite but large enough to account
2  2   !
for all losses within the numerical accuracy of the solution. In ∂v ∂w 2

∂u ∂v ∂u ∂w
order to show, that this criterion is fulfilled, it is necessary to + + + + + + (14)
∂y ∂x ∂z ∂x ∂z ∂y
visualize the distribution of the cross section integrated entropy
generation rate Ṡ0 along the centerline of the bend. This is done
in figure 2 for Re = 512 where sc is the distance along the cen- Again an isothermal flow at a temperature level Tm is assumed,
terline starting at the inlet of the numerical domain, see [20] for neglecting minor effects when a (moderate) temperature distri-
details. The dark area represents the losses inside the bend while bution occurs in the component.

5 Copyright
c 2012 by ASME
Dh
7

LVu = 5Dh
5

4
Ṡ ′ /Ṡ◦′

3
r = Dh
2

1
LVd = 35Dh
0
0 10 20 30 40 FIGURE 3. GEOMETRICAL DETAILS OF THE 90 DEG BEND
sc /Dh LVu , LVd : LENGTHS OF INTEGRATION

FIGURE 2. VISUALIZATION OF THE LOSS DISTRIBUTION


ALONG THE BEND’S CENTERLINE AT Re=512; DARK SHADED 2
10
AREA: LOSSES INSIDE THE BEND; LIGHT SHADED AREAS:
95% OF THE ADDITIONAL LOSSES UP- AND DOWNSTREAM,
RESPECTIVELY

TABLE 2. DETAILED INFORMATION ABOUT LOSSES DUE TO


A 90 DEG BEND; CROSS SECTION: A = D2h , CURVATURE RA-
DIUS: r = Dh 1
K

10
Re ∆ϕu /ϕ ϕc /ϕ ∆ϕd /ϕ Lu /Dh Ld /Dh K

4 (0.0045) 0.9954 (0.0001) (0.3320) (0.0779) 22.19


8 (0.0066) 0.9913 (0.0022) (0.4048) (0.4347) 11.25
16 0.0097 0.9727 0.0176 0.4505 0.9091 5.91
0
32 0.0127 0.8985 0.0888 0.5183 1.3720 3.46 10 0 1 2 3
10 10 10 10
64 0.0130 0.7262 0.2609 0.5724 2.1634 2.53 Re
128 0.0104 0.5367 0.4529 0.6147 3.4676 2.26
FIGURE 4. HEAD LOSS COEFFICIENT K FOR THE 90 DEG
256 0.0077 0.4029 0.5894 1.0797 8.3494 2.17 BEND; : RESULTS FROM TABLE 2; —: ANSATZ (16)
512 0.0040 0.2859 0.7101 0.3791 15.1179 2.27

Details of the numerical procedure by which Ṡ according to least squares problem, which here minimizes the relative er-
figure 1 is determined can be found in [20]. From [20] the result rors ∑ (K(15) (Re)/Ktab 2 (Re) − 1)2 is solved. The best fit gives
in terms of table 2 is taken over to this study with geometrical Ĉ1 = 2.20, Ĉ2 = 88.98, and m = 2.19, so that
details shown in figure 3.
In order to cast the K = K(Re)-results into a simple mathe-
matical best-fit curve the asymptotic behaviour for Re → 0 and 1/2.19
K = 2.202.19 + (88.98/Re)2.19

Re → ∞ according to table 1 is accounted for in the formula (16)

1/m
K = Ĉ1m + (Ĉ2 /Re)m

(15) represents the 90 deg bend head loss coefficient. Figure 4 shows
the quality of the best-fit curve (16). These results are the basis
with three unknown constants Ĉ1 , Ĉ2 , and m. With the for a validation of the SLA-approach which will be performed
K(Re) results from table 2 as Ktab 2 and from (15) as K(15) a with experimental results hereafter.

6 Copyright
c 2012 by ASME
THE SLA VALIDATION CASE: 90 DEG BEND losses up- and downstream) can be determined. For the latter
So far the SLA-approach is introduced as a physi- case both pressure ports have to be upstream of the bend where
cal/mathematical model that still has to be validated with respect the flow is fully developed and unaffected by the bend.
to its application for conduit component assessment. This vali- In all cases two pressure tappings at a time were connected
dation is performed in accordance with the ASME rules, see [21] to the differential pressure transducer while the remaining ones
for details. For the case of plain channel flow with a wall rough- were closed.
ness a validation of the SLA-approach has been accomplished As fluids we used two different silicone oils (polydimethyl-
already, see [22] for details. siloxane, Obermeier GmbH & Co., see [24]) which have a New-
For the validation process the model results in figure 4 and tonian rheological behaviour, since the molecule size of the poly-
equation (16), respectively, have to be compared to experimen- mer oils is small at this rather low viscosities. Their properties
tal results gained for the 90 deg bend under consideration. For are
that purpose we used the experimental set up shown in figure
5. We decided to perform our investigation in macro- rather • Korasilon
R
M20: ν = 20 × 10−6 m2 s−1 , ρ = 950 kg m−3
then micro-scale since the physics for both cases are the same as • Korasilon M100: ν = 100 × 10−6 m2 s−1 , ρ = 970 kg m−3

R

long as continuum flow prevails and Reynolds numbers are the


The kinematic viscosity of the oil is 20 and 100 times that of wa-
same. Measurements in microscales performed e.g. by Haller
ter. All property values are provided at a temperature T = 25 ◦C,
et al, see [23], show all the problems of microscale measure-
whereas temperature dependent values were used for the evalua-
ments and make a correlation of the results very difficult. Due to
tion of the measurements. Therefore a temperature measurement
the macro scale we could have laminar flow at rather high mass
was performed in the storage tank with a calibrated thermocou-
flow rates, finish the channel with high precision and connect the
ple.
pressure taps directly to the channel. The pressure is measured at
A constant flow rate was ensured by a brushless DC-motor
locations of developed flow rather than taking a global pressure
gear pump (Diener Extreme Series 4000 ml/min), see [25], which
drop between two points outside the component tangents where
was controlled using a PID-controller in the LabVIEW measure-
the flow situation is less controlled. This pressure difference di-
ment software. A digital counter from the measurement interface
rectly corresponds to the dissipation and thus the entropy gener-
card (NI PCI 6225) was used to compute the rotational velocity
ation in isothermal flow. The low pressure drop of several Pascal
from discrete TTL pulses generated by a tachometer inside the
in macro-scale only can be measured by a differential pressure
pump. The difference between the target value and the actual
transducer with a large membrane and thus high sensitivity.
value was used as an input to the controller to ajust the speed.
The bend is formed by a machined channel in an L-shaped
The fluids were stored in a 12 L tank upstream of the bend,
block of Fortal R
STS, a special alluminium alloy optimized to
while the pump was located downstream of the bend in order to
be machined with high surface quality. The surface was pol-
minimize disturbances. The tank was placed 30 cm above the
ished and the channel was closed with a transparent PMMA
bend to relieve the pump though it should be self-priming. Af-
(Plexiglas R
) cover plate. To prevent leakage the contact area was
terwards the fluid was collected in a storage tank and weighed
covered with Hylomar R
M sealant before combining both parts.
continuosly with a Sartorius TE 12000 balance connected to Lab-
The square cross section a2 has a hydraulic diameter a = 10 mm.
VIEW via an RS-232 interface. Thus the mass flow rate at con-
Both bend tangents have a length of 600 mm corresponding to
stant speed could easily be determined via a least squares fit of
60 Dh . For the pressure measurements, holes with a diameter of
the linearly increasing weight signal.
1 mm were drilled at certain positions along the centerline. Each
of the holes is attached to an 8 mm thread which serves as an
adaptor for a tube connector. Thus, each of the pressure tappings Experiment and Results
can be connected with one of the two pressure ports of a GE Our measurements consisted of two parts: First the pressure
Druck LPM 9381 differential pressure transducer with a range difference ∆piJ between an upstream pressure tapping i and a
of 0 − 500 Pa. A cut through the channel at a pressure tapping downstream tapping J was determined. This was done repeat-
position is shown in figure 6 together with the tolerances of the edly for several flow rates each at a prescribed pump speed in
channel. a measurement series. Afterwards the pressure differences ∆p jk
In the upstream section four pressure holes are located at beween two upstream tappings were determined in the same way.
a distance L̂u = [15, 10, 5, 2.5]Dh upstream of the bend’s entry Before each measurement a pressure difference with the
cross section. In the downstream section seven holes are located pump deactivated was recorded and its mean value, which rep-
L̂d = [2.5, 5, 10, 15, 20, 30, 40]Dh downstream of the bend’s out- resents the hydrostatic part of the pressure signal, was used to
let section, see figure 5. With these different locations the upper modify all pressure differences. All pressure differences were
limits for Lu and Ld as well as the losses in an undisturbed chan- stored as time dependent values, with a first order time depen-
nel flow (which is needed for the determination of the additional dency of the general form

7 Copyright
c 2012 by ASME
NI PCI 6225

RS 232

TC amplifier

balance

1
2
3

FIGURE 5. EXPERIMENTAL SET UP; 90 DEG BEND SHOWN FROM ABOVE


BOTTOM: 90 DEG BEND WITH PRESSURE METERING POINTS
TOP: SCHEME OF THE WHOLE SET UP
1: DIFFERENTIAL PRESSURE MEASUREMENT; 2: DIFFERENTIAL PRESSURE MEASUREMENT FOR UNDISTURBED FLOW; 3: WEIGH-
ING TANK WITH BALL TAP FOR DRAINING; 4: STORAGE TANK WITH THERMOCOUPLE

cially important in our case, since the time constants were very
large: τ ≈ 85 s for M20 and τ ≈ 400 s for M100. These time con-
t − t0 stants were caused by the large sensor membrane which needed
∆p(t) = (∆p∞ − ∆pt0 )(1 − exp(− )) + ∆pt0 (17)
τ a comparably large amount of fluid to be transported through the
measurement tubes. These tubes, however, had a small diameter
Here ∆p∞ was the steady signal, ∆pt0 was the pressure difference of 0.063 In ≈ 1.6 mm and thus a high damping effect. In all cases
at a time t0 where the fit was started, and τ was the time constant ∆p∞ was crosschecked with the last recorded value of ∆p(t) with
of the transient part of the signal. All parameters had to be de- differences between ∆p(tmax ) and ∆p∞ of about 0.1 Pa, which
termined by a curve fitting approach to get ∆p∞ . This was espe-

8 Copyright
c 2012 by ASME
2
10

channel

K
10

FIGURE 6. CUT THROUGH THE CHANNEL (WITHOUT PMMA


COVER) AT A PRESSURE TAPPING POSITION

was small compared to the accuracy of the sensor.


During the pressure measurements, the weight signal was 0
recorded and evaluated afterwards. With a temperature depen- 10
0 50 100 150 200
dent function for the density, see [24], the mass flow rate could Re
easily be rescaled into the velocity needed for nondimensional-
FIGURE 7. EXPERIMENTAL RESULTS FOR K(Re). MEASURE-
ization. Also the temperature dependent viscosity needed for the
MENTS AT Re < 20 WITH M100, ALL OTHER CASES WITH M20;
determination of Re was obtained from an interpolation of the
—: THEORETICAL RESULT FROM ANSATZ (16)
temperature dependend values provided in [24]. These values,
however, represent averaged values based on various samples.
Since the viscositiy depends on the average molecule size of the 4
polymer oils, which has to be determined for the actual batch, L̂d = 2.5Dh
the temperature dependent values in [24] were scaled with the L̂d = 5Dh
viscosity from the actual batch at T = 25 ◦C. These values were 3.5 L̂d = 10Dh
provided by the manufacturer together with the uncertainty of L̂d = 15Dh
the used rheometer. They are ν = 18.3 mm2 s−1 for M20 and
ν = 95.4 mm2 s−1 for M100. Alltogether, this leads to an expres-
3
sion for K in terms of the measured quantities:
K

2.5
L̂ui + L̂dJ
∆piJ − ∆pjk
L̂u j − L̂uk
K=2 (18)
ρ(T )u2m 2

with um = ṁ/(ρ(T )D2h ) and the corresponding Reynolds number 40 60 80 100 120 140
Re Re

FIGURE 8. EXPERIMENTAL RESULTS FOR K(Re, L̂d ) WITH


M20 AS WORKING FLUID
um Dh
Re = (19)
ν(T )
number of tappings it is not possible to determine Ld with the res-
Results of our measurements are shown in figure 7 with good olution presented in table 2. Yet, the measurements give a good
agreement with respect to the theoretical results from equation agreement with the simulation, when the distance to the tapping
(16). from the bend outlet section becomes large enough. Within the
To check whether our numerical model gives a valid pre- measurement uncertainty the K-value then becomes independent
diction also for the downstream length, measurements have been of the downstream position L̂d . This happens at lengths larger
performed using four different pressure tappings in the down- than Ld , since Ld is defined to cover only 95% of the additional
stream section. The results are shown in figure 8. Due to the low losses downstream.

9 Copyright
c 2012 by ASME
Validation of the Model TABLE 3. ESTIMATED UNCERTAINTIES OF PARAMETERS IN-
CLUDED IN uval
In figures 7 and 8 the data uncertainty uD due to the mea-
surement is shown by errorbars. These uncertainties are needed, uncertainty value comment
to show, whether the comparison error E = KS − KD between
numerical (KS ) and experimental (KD ) values of K can be jus- u∆p 0.1%FS from manufacturer
tified within the accuracy of the experiment. According to [21] = 0.5 Pa
the data uncertainty uD , however, is not the only source of un-
certainty. Together with the numerical uncertainty unum and the uDh 0.01 mm maximum uncertainty, see fig. 6
uncertainty of model inputs uinp the so-called validation uncer- uL̂ 0.1 mm see fig. 5
tainty uval is
uṁ /ṁ 2% gained from validation measure-
ments with steady state mass
flow determination
uν /ν 5% for M100 inquired from manufacturer
q
uval = u2D + u2num + u2inp (20)
10% for M20

with They are

∂K D4
uval : uncertainty of validation procedure = 2ρ h2
uD : uncertainty of experimental data ∂ ∆piJ ṁ
unum : uncertainty of numerical calculation ∂K D4 L̂ui + L̂dJ
= 2ρ h2
uinp : uncertainty of numerical results due to uncertain model ∂ ∆pik ṁ L̂ui − L̂uk
inputs ∂K D3h

L̂ui + L̂dJ

= 8ρ 2 ∆piJ − ∆pjk
∂ Dh ṁ L̂ui − L̂uk
The data uncertainty uD can be computed from a propagation ∂K D4 ∆pjk
equation, which links the partial derivatives of K according to = 2ρ h2
∂ L̂(ui,dJ) ṁ L̂ui − L̂uk
(18) with respect to potential erroneous quantities with expected
uncertainties of these quantities shown in table 3. The propaga- ∂K D4 L̂ui + L̂dJ
= 2ρ h2 ∆pjk
tion equation includes estimates of the errors in the pressure mea- ∂ L̂u(i,k) ṁ (L̂ui − L̂uk )2
surements of the component and the channel (u∆piJ , u∆pik ), the ∂K D4h

L̂ui + L̂dJ

systematic errors in the geometry (uDh , uL̂(ui,dJ) , uL̂u(i,k) ), and the = 4ρ 3 ∆piJ − ∆pjk
∂ ṁ ṁ L̂ui − L̂uk
uncertainties of the flow rate measurement (uṁ ). The influence
of a temperature uncertainty affecting the density is neglected For small values of the Reynolds number uD is dominated by the
since the volumetric thermal expansion coefficients of both oils uncertainty of the pressure measurement, since then u∆p reaches
are small (M20: 9.7 × 10−4 K−1 , M100: 9.4 × 10−4 K−1 ). the order of the pressure differences. In order to increase the
Thus uD reads: accuracy at small Reynolds numbers M100 is used for Re < 20
for which higher velocities lead to enlarged pressure differences.
For Re > 20, however, the pressure inside the bend becomes so
s low5 , that bubbles of air (solved in the oil) appear so that the
∂K 2 2 ∂K 2 2 ∂K 2 2 assumption of single phase flow is no longer valid and M20 has
    
uD = u∆piJ + u∆pik + uDh to be used.
∂ ∆piJ ∂ ∆pik ∂ Dh
!2 !2 The determination of unum (numerical uncertainty) is rather
∂K 2 2 difficult, since adaptive grid refinement has been applied during
 
∂K 2 ∂K
+2 uL̂ + 2 u2L̂ + uṁ (21) the CFD-calculation in OpenFOAM to keep the grid as efficient
∂ L̂(ui,dJ) ∂ L̂u(i,k) ∂ ṁ
as possible, see [20] for details. A previous study, however, was

The partial derivatives have to be determined for every validation 5 With the pump located downstream of the bend the pressure inside the bend
case, i.e. for each Reynolds number applied in the experiment. is always lower than the ambient pressure.

10 Copyright
c 2012 by ASME
TABLE 4. UNCERTAINTY ESTIMATION FOR unum BASED ON
GRID REFINEMENT AND EXTRAPOLATION
Kext −K3
1 2 3 Kext Kext ∂K uν
uinp = (−Re) (23)
∂ Re ν
number of cells 168432 333984 663950

grid size ratio — 1.254 1.238 With (16) the derivative of K is


see [21]
 m ! m1 −1  m
∂K Ĉ2 Ĉ2
K(Re), Re = 4 22.01 22.09 22.14 22.22 0.39% m
= − Ĉ1 + Re−1
∂ Re Re Re
8 11.15 11.20 11.22 11.26 0.35%
16 5.86 5.88 5.90 5.93 0.49%
so that (23) now reads
32 3.43 3.45 3.46 3.48 0.57%
64 2.51 2.52 2.53 2.55 0.62%
 m ! m1 −1  m
128 2.26 2.27 2.28 2.30 0.73% Ĉ2 Ĉ2 uν
uinp = Ĉ1m + (24)
256 2.18 2.19 2.20 2.23 1.09% Re Re ν

An estimation for uνν is included in table 3. Note that uinp be-


performed with a fixed grid. Results of this study for three differ-
comes small for Re → ∞ in (24), since then K approaches a con-
ent grids are shown in table 4 for Reynolds numbers in a range
stant value. The sensitivity of K with respect to changes in ν will
that has also been covered by the measurements. In the last but
be large for Re → 0 only.
one column, an extrapolated value Kext (according to eqns. (2-
Since the temperature dependence of the viscosities is small
4-3)-(2-4-8) in [21]) based on the results for the different grids
compared to the viscosity uncertainties, see table 3, temperature
is shown. These Kext -values are very close to those for the lo-
measurements were assumed to be exact. All measurements were
cally refined grid, see table 2, and very close to the K3 -values for
performed at temperatures between 22 ◦C and 27 ◦C, being al-
the finest grid, see table 4, last column. Obviously the numeri-
most constant during one measurement.
cal uncertainty is small compared to the other errors so that unum
Finally, the comparison error E = KS − KD between simu-
will be neglected. The authors are aware, however, that a CFD-
lation results (16) and experimental data from figure 7 is of in-
calculation is generally subject to uncertainty. A verficication of
terest. This error, however, has to be assessed with respect to the
the solver was not performed, since it uses the well known and
validation uncertainty uval . Assuming an acceptable validation
already verified SIMPLE algorithm.
uncertainty, |E| should be smaller than |uval | for an appropriate
So far no estimation about the uncertainty influence of the
model. Then the model can be called validated. Values of |E| and
viscosity has been made. Since only two fluids with two viscosi-
±uval around that value according to [21], for the measurements
ties are available in the experiment, it is not possible to determine
from figure 7, are displayed in figure 9. For almost every Re the
the sensitivity of K to changes in ν using finite differencing. Un-
validation condition is fulfilled, i.e. |E|/KS is smaller than the
certainties of viscosity can either be interpreted as an uncertainty
validation uncertainty. We take this as a successful validation
of Re in the experiment or as an uncertainty in the model input
though three out of the twelve cases slightly miss the validation
for ν. In both cases the Reynolds number between experiment
criterion.
and simulation would differ. Here, the uncertainty of the viscos-
ity is interpreted as an uncertainty of the model input. Thus uinp
becomes Systematic Errors
Figure 7 shows that experimentally determined values with
M100 are systematically below the numerical values while the
∂K ∂ K ∂ Re corresponding values for M20 are systematically above the nu-
uinp = uν = uν (22)
∂ν ∂ Re ∂ ν merical values. This is probably due to a systematic error in the
determination of viscosity values.
The chain rule is applied to determine the derivative of K with In the numerical model, incompressible flow was assumed
respect to Re. This is rather convenient since then the approx- and viscous heating has been neglected. Thus the viscosity and
imation (16) can be used to determine the derivative. With the density (though it does not exist in the incompressible code)
∂ Re 2
∂ ν = −uDh /ν = −Re/ν, (22) becomes are assumed to be constant and velocities should be small. In

11 Copyright
c 2012 by ASME
0.4
6

0.3
5

0.2 4

Ṡ ′ /Ṡ◦′
3
0.1

2
0
1
−0.1
0
0 20 40 60 80
sc /Dh
−0.2
0 50 100 150 200
Re FIGURE 10. INLET EFFECT UPSTREAM OF THE BEND FOR
Re=512; INLET LOCATED AT sc = 0; DARK SHADED AREA: IN-
FIGURE 9. COMPARISON ERROR AND VALIDATION UNCER-
LET EFFCT; LIGHT SHADED AREA: LOSSES DUE TO THE BEND
TAINTY; ∗: RELATIVE COMPARISON ERROR E/KS ; BARS INDI-
CATE ±uval /KS
to cover all effects induced by the bend but they are small
enough to minimize data uncertainty of the measurement.
the experiment, mean values of the velocity do not exceed um = 3. The fluids were chosen carefully to provide Newtonian be-
0.4 m s−1 . This would be exactly the velocity for M20 at Re = haviour with a pressure drop large enough for a differential
200. For M100 at Re = 20 the velocity is less than 0.2 m s−1 . pressure sensor.
With a heat capacity c ≈ 1500 J/(kgK) the Eckert number Ec =
u2m After this validation we suggest to use the numerical method de-
cT0 is small and viscous heating is indeed negligible.
Another source of error could be the inlet section at the up- scribed here together with our definition of K for the determi-
stream tangent with an effect on the flow in the bend and the nation of loss coefficients for arbitrary conduit components in
undisturbed channel upstream of the bend. However, it has been laminar flow.
shown in a numerical simulation including the inlet section dis-
played in figure 5, that the influence of the inlet is small, when
the length of the upstream tangent is larger than 50 Dh , even at a ACKNOWLEDGMENT
Reynolds number as high as Re = 512. The decay of the addi- The authors gratefully acknowledge the support of the DFG
tional entropy generation induced by the inlet is shown in figure (Deutsche Forschungsgemeinschaft).
10.

REFERENCES
CONCLUSION [1] Moran, M., and Shapiro, H., 1996. Fundamentals of Engi-
A model for the prediction of losses in conduit compenents neering Thermodynamics, 3. ed. John Wiley & Sons, Inc.,
has been presented which consistently determines loss coeffi- New York.
cients as well as upstream and downstream lengths of impact [2] Baehr, H., and Kabelac, S., 2009. Thermodynamik, 14. ed.
with a single computation. The SLA-model has been success- Springer-Verlag.
fully validated within the validation accuracy. A test facility has [3] Herwig, H., and Kautz, C., 2007. Technische Thermody-
been built which meets several requirements: namik. Pearson Studium, München.
[4] Bejan, A., 1977. “The concept of irreversibility in heat ex-
1. The scale was a compromise between low differential pres- changer design: counter-flow heat exchangers for gas-to-
sures in macro-scales and geometry uncertainties in micro- gas applications”. Journal of Heat Transfer, 99, pp. 274–
scales. 380.
2. The distances of the pressure tappings from the bend’s in- [5] Bejan, A., 1982. Entropy generation through heat and fluid
let and outlet, respectively, have been chosen large enough flow. John Wiley & Sons, New York.

12 Copyright
c 2012 by ASME
[6] Bejan, A., 1996. Entropy generation minimization. CRC and t-junctions”. International Journal of Heat and Mass
Press, Boca Raton, New York. Transfer, 52(11-12), pp. 2678 – 2689.
[7] Herwig, H., 2012. “The role of entropy generation in [24] Obermeier GmbH & Co. Kg, 2009. Korasilon Öle der M
momentum and heat transfer”. Journal of Heat Transfer, Reihe. product brochure (web). URL www.obermeier.
134(3), p. 031003. de/fileadmin/contents/pdf/silikon-allg/
[8] Herwig, H., and Wenterodt, T., 2009. “Wall roughness ef- KORASILON_Oel_M_DE.pdf.
fects: A second law analysis (SLA)”. In Proceedings of the [25] Diener Precision Pumps Ltd, 2008. Diener 4000
IUTAM symposium on the physics of wall-bounded flows ml/min Extreme Series. product brochure (web). URL
on rough walls, Cambridge, GB. http://www.dienerprecisionpumps.com/de/
[9] Herwig, H., Gloss, D., and Wenterodt, T., 2008. “A new zahnradpumpe_4000.html.
approach to understand and model the influence of wall
roughness on friction factors for pipe and channel flows”.
Journal of Fluid Mechanics, 613, pp. 35–53.
[10] Munson, B., Young, D., and Okiishi, T., 2005. Fundamen-
tals of Fluid Mechanics, 5th ed. John Wiley & Sons, Inc.,
New York.
[11] Idelchik, I., 2007. Handbook of Hydraulic Resistance.
Begell House, Inc., Redding.
[12] Ward-Smith, A. J., 1980. Internal Fluid Flow. Clarendon
Press, Oxford.
[13] VDI, 2006. VDI-Wärmeatlas, 10 ed. Springer-Verlag,
Berlin, Heidelberg.
[14] Moody, L., 1944. “Friction factors for pipe flow”. Trans.
ASME, 66, p. 671.
[15] Nikuradse, J., 1933. “Strömungsgesetze in rauhen Rohren”.
Forschung auf dem Gebiet des Ingenieurwesens, 361,
pp. 1–22.
[16] Schiller, L., 1923. “Über den Strömungswiderstand
von Rohren verschiedenen Querschnitts- und Rauhigkeits-
grades”. Z. Angew. Math. Mech., 3, pp. 2–13.
[17] Langelandsvik, L.I.; Kunkel, G. S. A., 2008. “Flow in a
commercial steel pipe”. Journal of Fluid Mechanics, 595,
pp. 323–339.
[18] Schmandt, B., and Herwig, H., 2011. “Internal flow losses:
A fresh look at old concepts”. Journal of Fluids Engineer-
ing, 133(5), p. 051201.
[19] Kittredge, C.P.; Rowley, D., 1957. “Resistance coefficients
for laminar and turbulent flow through one-half-inch valves
and fittings”. Trans. ASME, pp. 1759–1766.
[20] Herwig, H., Schmandt, B., and Uth, M.-F., 2010. “Loss
coefficients in laminar flows: Indispensable for the design
of micro flow systems”. ASME Conference Proceedings,
2010(54501), pp. 1517–1528.
[21] The American Society of Mechanical Engineers, 2009.
Standard for Verification and Validation in Computational
Fluid Dynamics and Heat Transfer. ASME, New York.
[22] Gloss, D., and Herwig, H., 2010. “Wall roughness effects in
laminar flows: An often ignored though significant issue”.
Exp. in Fluids, 49, pp. 461–470.
[23] Haller, D., Woias, P., and Kockmann, N., 2009. “Simu-
lation and experimental investigation of pressure loss and
heat transfer in microchannel networks containing bends

13 Copyright
c 2012 by ASME

View publication stats

S-ar putea să vă placă și