Sei sulla pagina 1di 4

Verifying the dynamic model of a heat exchanger configuration

Pierre Kauhanen,
Department of Chemical Engineering
University of Lund, Sweden.
Abstract
With the development in the field of computer technology, modeling and simulation is getting more and
more interesting alternative for pilot plants for companies. Tetra Pak Processing Systems have developed a
model over a heat exchanger in the software Dymola. Dymola is a modeling and simulation tool for multi-
engineering purposes developed by Dynasim. In order to be able to make any valid conclusions the model
has to be verified, this study tries to verify the dynamics over a heat exchanger model. The shell- and tube
heat exchanger used in this study is a Tetra Pak Spiraflo. There exists a static error of about +10% in the
heat transfer coefficient. The analysis shows that it is recommended to use the Quantum database when
simulating a dynamic behavior with large perturbations. Required discretization depends on what type of
perturbation or heat exchanger configuration. Step perturbation requires lower discretization (N=2) than a
pulse discretization (up to N=10). Counter-current configuration demands a lower discretization, especially
if looking on the channel opposite of the perturbation channel. The transport delay parameter gives a better
correlation at lower discretization for the tube side in the co-current configuration; in the counter-current
configuration there is no significant difference between transport delay and no delay. The shell side shows
a different pattern, simulation without transport delay gives a better result.

Introduction 1. Is it necessary to use the Quantum


The traditional way of making studies is by using database when simulating?
real experiments. Scientists, engineers and 2. How high discretization is necessary to
companies spend a lot of money and resources the model to get good simulation results?
each year by building pilot plants to conduct 3. Is the transport delay parameter
these real experiments. The most cost efficient contributing to the model?
way, up until now, to evaluate and do research
has been by building a pilot plant. Along with The Model
the development of computer technology, The basic model for heat transfer is by
computer modeling and simulation has become convection and conduction through a wall. The
more and more interesting as a new research wall is split in two layers with a homogenous
tool. Companies and the academic world have temperature in each layer. This enables the wall
found that it is more cost efficient to do research to have different surface temperatures. The wall
inside a computer instead of building expensive will function either as a part of the overall
pilot plants. Tetra Pak Processing Systems have system or as a system boundary (insulation).
developed a model over a heat exchanger in the
software Dymola. Dymola is a modeling and Energy balance
simulation tool for multi-engineering purposes
developed by Dynasim. However, by Channel 1:
substituting real experiments for computer d
modeling and simulation a certainty, that the ρ1mean ⋅ Cp1mean ⋅ V1 ⋅ TOut 2 = Q11 − Q12 + Q1
computer model correctly describes the real dt
system, has to exist. If a scientist is to draw valid Channel 2:
and reliable conclusions from his research on the d
computer model, he has to be certain that the
ρ 2 mean ⋅ Cp 2 mean ⋅ V2 ⋅ TOut 2 = Q21 − Q22 + Q2
dt
model give correct data, i.e. he has to be sure Where Qik is the heat flux and QF is the
about the validity of his model. Tetra Pak has volumetric flow in the channels:
constructed a model over a shell- and tube heat
exchanger. This study tries to answer the Qi1 = ρ imean ⋅ Cp imean ⋅ QF ⋅ TIn
question whether this heat exchanger model is a Qi 2 = ρ imean ⋅ Cpimean ⋅ QF ⋅ TOut
good description of the real heat exchanger. To
One problem when discretizing a volume is that
simplify this rather large question the purpose
a temperature perturbation can get an instant
boils down to the following three questions:
impact when discretizing is low. The technique
normally used is a mean value of inlet and outlet
temperature is used as a temperature of the water level between the two tanks, which will
volume element. manifest in a different pressure profile in the
Tx = (T1 + T2 ) / 2 system. These involuntary flow perturbations
cannot be avoided when opening and shutting
The result is that perturbations when entering a the valves but are taken into account in the
volume element will get an instant impact of the simulations as the simulations have the same
temperature used for the driving force. Now flow as the experiments. Flow perturbation is
when we want to portray a transport delay the done by opening or shutting a ball valve that are
following equation is used: situated after the flow transmitter. The flow
perturbation in itself is almost a perfect step,
Tx = (T f + T2 ) / 2 where Tf : which means that the flow change is almost
immediate.
dT f r
V⋅ = Q ⋅ (T1 − T1 ) . Simulation
dt The purpose with the simulations is to make a
comparison with the real experiments. This
Medium 2
comparison can only be made if the model and
the real experiments experience the same
Q22 Q2 T2
Q21 perturbation. By using the same input data in the
model that was sampled from the experiments it
T2W
is somehow secured that the model experience
QW
Wall th
the same disturbance as the real experiments. By
T1W pasting the sampled data into tables that controls
Q1
T1
the temperatures and the flows in the model,
Q11 Q12 almost the same disturbance can be performed.
Medium 1 See figure 2 and 3 below for a comparison
between the simulated perturbation and the
Figure 1 Model of two channels and a wall. experimental perturbation. All simulated
perturbations have the same pattern that is shown
figure 2 and figure 3. All simulated temperatures
When introducing Tf into equation we get:
and flows are the simulated outputs from the
dT2 (T1 − T2 ) NTU ⎡ (T f + T2 ) ⎤ model transmitters.
= + ⎢T3 − ⎥
dt τ τ ⎣ 2 ⎦
r Temperature Pulse Perturbation
Co-Current

(T − T )
Temperature

dT f 25

= 1 1
dt τ
r
20

Tf will increase until T1= T1 .


15

where τ = V/QF and NTU = kA/ρCpQF. Simulations


Experiments

10

The transport delay will get a bigger impact with


lower discretization. The larger discretization, 5

the smaller volume element where the fluid will


pass and thus there will be a smaller transport 0
0 20 40 60 Time 80 100 120 140

delay.
Figure 2 Temperature perturbation, simulation
Experiments and experiment
By opening the valve to the hot-water tank and
closing the valve to the cold-water tank a
temperature perturbation is achieved. It is
important that the opening and closing of the
valves are made in the right order; otherwise the
flow can be shut and thus causing an unwanted
flow perturbation. There is also another flow
perturbation that comes from the difference in
Flow Pulse Perturbation
Counter-Current
Table 2 Pulse Perturbation linear correlation
Flow (m3/h)
2,5

Configuration Flow Temperature


2 Tube Shell Tube Shell
CoCurrent N=6 N=10
1,5 r=0.994 0.983
CounterCurrent N=4 N=2
Simulation
Experiment N=10 N=10
1 r=0.960 r=0.990 r=0.998 r=0.9693
An acceptable correlation of 0.99 between the
simulations and the experiments is achieved for
0,5

0
temperature perturbation on the tube side and on flow
0 20 40 60 Time 80 100 120
perturbation on the shell side in a counter-current
Figure 3 Flow perturbation, simulation and heat exchanger.
experiment Table 3 Step Perturbation
Result and analysis Configuration Flow Temperature
Table 4 display the difference between the Tube Shell Tube Shell
Quantum database and the real experiments, CoCurrent N=4 N=2 N=2 N=2
Quantum give different pressure drop and ca r=0.997 r=0.998 r=0.9997 r=0.998
10% larger k-values than what the real CounterCurrent N=10 N=4 N=2 N=4
experiments does, see figure 4 for comparison. r=0.9881 0.990 r=0.9997 r=0.994
Table 1 Experimental data for constant k-value
simulations. All simulations show a good correlation with
Simulation Type K DpTube DpShell experiments. This is explained by the step
W/m2/C Pa Pa perturbation do not involve much dynamics;
Co-Current 2850 79000 37500 there is only dynamic at the beginning and at the
Temp1 (3253) (42000) (128000) end of the step perturbation.
Counter-Current 2725 60000 41000 Temperature Pulse Perturbation

Temp2 (3018) (37000) (93000)


Correlation
Co-Current
1
Counter-Current 3000 40500 89000
Flow3 (3319) (43000) (143000) 0,995

(xx) Data from the Quantum database with same 0,99

input as the experiments used to calculate the k- 0,985

value. 0,98

No Delay Tube
0,975 No Delay Shell
Heat transfer Coefficien vs Reynolds number Delay Tube
K (W/m2 C) Co-Current Delay Shell
0,97
6000

0,965

5000

0,96
1 2 3 4 Disctretization 5 6 7

4000

Figure 5 Transport delay Co-current


3000 Experiment Tube
Quantum Tube
Experiment Shell
configuration
Quantum Shell
2000

1000

0
0 20000 40000 60000 Reynolds 80000 100000 120000

Figure 4 K-value from experiments and


Quantum database
Temperature Pulse Perturbation
Correlation Counter-Current
1
The necessary discretization to make a good
simulation differs very much from what type of
perturbation, i.e. step or pulse perturbation, is
0,99

0,98 introduced to the model. Also important is what


kind of heat exchanger, i.e. co-current or
0,97
No Delay Tube
No Delay Shell
counter-current, and what kind of perturbation is
0,96
Delay Tube
Delay Shell simulated, i.e. flow or temperature perturbation.

0,95
Step perturbation does not demand very high
0,94
discretization, discretization of two or four times
1 2 3 Discretization
4 5 6 7
is usually enough. This is explained by the fact
Figure 6 Transport delay, Counter-current that step perturbation does not involve much
configuration. dynamics.

By introducing transport delay into the model, Pulse perturbation on the other hand demands a
better correlation is achieved at lower greater discretization, sometimes up too ten
discretization for the tube flow in a co-current times and that is not always good enough. The
heat exchanger. However, the difference between high grade of discretization is because pulse
the simulation with transport delay and without perturbation involves more dynamics than step
transport delay decrease with higher perturbation but there also seems to be a
discretization. No significant difference between connection between on what side the
transport delay and no transport delay can be perturbation is on and necessary discretization.
seen when simulating a tube side in a counter- This could be explained by the fact that the
current HEX. See figure 6. Transport delays on transmitters in the model do not show exactly the
the shell side show a different pattern from same dynamics as the real transmitters.
transport delays on the tube side. Both co-current
and counter-current configuration show the same The counter-current configuration does not
pattern; simulations without transport delay demand a high discretization, if one is looking
show better correlation than simulations with on the opposite side of the perturbation. The
transport delay. See figure 5. reason for this behavior is that the opposite side
of perturbation in the counter-current
Discussion configuration meets the perturbation
The static analysis shows that there is a immediately as it enter the heat exchanger and
stationary error between the Quantum database can thus react instantly on any change. The co-
and the real experiments. This is explained by current configuration do not show this
that the heat transfer correlation for the MT phenomenon, the perturbation will propagate
25/16S-6 Pilot is interpolated from correlations through the whole heat exchanger until it will be
of other heat exchangers that are empirically able to be read at the end of the heat exchanger.
derived. The heat transfer coefficient is about 10 This is the reason why opposite side of the
% bigger in Quantum than it is when calculated perturbation in the counter-current configuration
from the real experiments. have a better linear correlation.

Special precaution should to be taken when The transport delay parameter gives a better
simulating large flow perturbation without the correlation at lower discretization for the tube
Quantum database, it is not recommended to do side in the co-current configuration; in the
such simulations. Temperature perturbation counter-current configuration there is no
without the Quantum database does not show the significant difference between transport delay
same deviation as the flow perturbation from the and no delay. The shell side shows a different
real experiments. Simulation without the pattern, simulation without transport delay gives
Quantum database is however developed to a better result. This may be explained by the fact
function with small perturbations around a that the temperature perturbation is on the shell
working point. The conclusion to be drawn from side and the perturbation side shows generally
the database analysis is that the Quantum not as good correlation as the non-perturbation
database is recommended to make dynamic side.
simulations with large perturbations.

Potrebbero piacerti anche