Download as pdf or txt
Download as pdf or txt
You are on page 1of 16

Innovations from Venti Oelde

Computational Fluid Dynamics for fans and plants


Computational Fluid Dynamics
for fans and plants

Fan optimization with CFD modelling

In the field of turbomachin- 1 Introduction 2 The mathematics


ery construction, CFD has, behind CFD
within the space of just a
few years, become an indis- Computational Fluid Dynam- tial volumes that depend on The set of equations describ-
pensable tool for the opti- ics (CFD) is a method that is the occurring flow gradients. ing the processes of momen-
mization and new design of used for the computerized cal- These partial volumes then tum, energy and mass transfer
turbomachines, as well as culation of technical flows with form the basis for the equa- is known as the Navier-Stokes
for the elimination of fluid- state-of-the-art accuracy. With tions. For complicated geome- equations. These partial differ-
mechanic problems in in- this method, it is possible to tries, such as fans, such cal- ential equations, which were
stalled systems. The follow- make far more precise state- culations are therefore only derived in the early 19th cen-
ing paper demonstrates the ments regarding the behav- feasible if they are performed tury and have no known gen-
basic knowledge as well as iour of a technical or natural by computer. This particularly eral analytical solution, can be
applied examples of the CFD flow than could be made on applies to simulation of tran- solved numerically. For the nu-
simulations of process fans. the basis of classical partly- sient flow conditions, i.e. con- merical solution of this system
empirical, test-based ap- ditions that vary over time, as of equations, the differential
proaches. CFD has been em- a number of related time transport equations are first
ployed on an economic scale steps have to be computed in transformed into algebraic
since around 1973, when it order to obtain sensible re- equations by so-called discret-
was first applied in the aircraft sults. ization. In the case described
construction sector. The un- here, the Finite Volume Meth-
derlying partial differential od is used for this purpose.
equations according to Na- This method is generally ap-
vier-Stokes, which precisely plied as the basis for simula-
describe the flow characteri- tion programs in the Computa-
stics of a Newton fluid, have tional Fluid Dynamics sector.
been known since the first half In the FM method, a three-
of the 19th century. The cal- dimensional computational
culation work required for solv- mesh is laid over the flow vol-
ing these differential equa- ume to be investigated. Deter-
tions is very complex: the flow mination of the distribution of a
volume has to be subdivided flow value Φ thus takes place
into a number of smaller par- via discrete calculation points
2
(nodes). In this way, the con- vantages of CFD vis à vis cyclones or the air handling to make further improvements
tinuous distribution of Φ is re- classical test bench methods systems of recycling plants. In without the aid of numeric
presented by the Φ values at is the ability to examine in de- addition to such tasks as aero- methods, because advance-
discrete points. This creates tail any desired area within dynamic optimization, other ment of machine designs is
an algebraic equation (finite the simulated boundaries and fields of application for the generally accompanied by an
difference equation) for every thus to clearly identify and software used at Venti Oelde exponential rise in costs com-
calculation point representing subsequently optimize the are the simulation of mixing pared to benefit.
a single computation volume zones of rough flow. The field processes in stacks and heat
(cell). This algebraic equation of application of the software transfers during drying pro-
is then used for determining thus stretches from the devel- cesses and to determine the
the flow values. The differen- opment of new products forces imposed on different
tial equations from all the cal- through the optimization of structural geometries by the
culation points form a system existing machine series and flow passing through them.
of linked algebraic equations. up to the elimination of fluid- These can then be employed
This system of equations has mechanic problems in in- as boundary conditions for
to be solved with a numeric al- stalled systems. The CFD structure-mechanical simula-
gorithm. A number of direct or user gains an understanding tions in order to determine the
iterative calculation methods of flow phenomena that he maximum occurring stresses
are available for the solution can hardly hope to achieve or for modal analysis to deter-
of algebraic equation systems. through practical trials, or that mine possible resonance ef-
can only be obtained at the fects caused by transient
As the processing power of highest academic level where forces. In contrast to classical
available computer technology the limiting factors are then trials, a simulation entails no
increases, the application of the cost and time involved as restrictions with regard to con-
CFD is becoming more and well as know-how transfer. struction size, power con-
more interesting for even me- sumption, possibly harmful
dium-sized and small compa- At Venti Oelde the method is conveying media or high tem-
nies, in spite of the high not only used for fan con- peratures. To summarize, it
purchase and maintenance struction, but also for system can be stated that given the
costs of the software and the engineering, for instance in current high technological lev-
present lack of specialist the optimization of duct el of turbomachinery construc-
users. One of the biggest ad- routes, stacks, filter plants, tion, it is practically impossible
3
Figure 1 Parameterized layout with modification

3 Performance of a CFD a) Creation


simulation

To allow performance of a The first step is to create a any modifications in the ge- ition, one or more two-dimen-
CFD simulation a number of three-dimensional model of ometry can be quickly incor- sional sketches are produced.
work steps have to be carried the geometry using CAD soft- porated into the model by sim- These can be used for cre-
out in the sequence described ware with 3D capability. It ply entering the new dimen- ating three-dimensional mod-
below. It is presumed for this must be remembered at this sion, whereupon the entire els, for example by extrusion,
purpose that the geometry stage that in the case of flow model and any adjacent rotation or by means of Bool-
data and the boundary condi- simulations the 3D geometry zones adapt themselves to ean operations like cutting or
tions are known: of the flow zone is needed the new dimension. merging.
and not that of the structure,
as is usual in other types of This brings a significant time
analysis. Increasingly, the saving when optimizing the
created 3D models are com- model compared to the other-
pletely parameterized (Figs. 1 wise necessary new creation
and 2) in order to ensure that of a different model. In add-

Figure 2 Adapted 3D models

4
Laminar boundary layer
Turbulent flow
k δ1

W max
d r

Figure 3 Unstructured Figure 4 Block-structured Figure 5 Flow profile


computational mesh hexahedral mesh

b) Discretization

After the required 3D models by the boundary layer, as the


have been designed, a three- flow in this zone has to over-
dimensional computational come not only the pressure in-
mesh has to be created. The crease but also the wall shear
elements of this mesh form the rate. If there is turbulent flow
basis for the control volumes (and nearly all technical flows
that are needed, as already are turbulent) the boundary
mentioned, for the partial dif- layer is only a few tenths of a
ferential equations, so that millimetre thick and has a para-
these can be iteratively solved bolic curve with respect to the
in matrix form in accordance velocity distribution profile (Fig.
with the number of elements. 5). The computational mesh
Depending on the employed for this zone must be given
software, the user is generally such a fine resolution that this
able to create either a simple- parabolic curve is represented
to-generate, unstructured com- with sufficient accuracy. In a
putational mesh (Fig. 3) or a normal view of the wall, this re-
considerably more difficult-to- quires between 8 and 14 ele-
generate, block-structured ments in the boundary layer
mesh (Fig. 4). As a rule, a zone (Fig. 6).
block-structured mesh with a
comparable structure resolu-
tion produces better results in
significantly shorter processing
time.
Figure 6 Boundary layer discretization with vector arrows
For this reason CFD studies at
Venti Oelde are, with very few
exceptions made for time rea-
sons, all based on block-struc-
tured computational meshes.
Care has to be taken that the
flow dynamic boundary layer is
given an adequate resolution,
because – for instance – flow
disruptions are always caused
5
c) Simulation

After the 3D model and sub-


sequently the computational
mesh have been created, the
simulation software (ANSYS
CFX) is brought into use.

ANSYS CFX consists of


three modules: ANSYS CFX-
Pre (Fig. 7) for definition of
the boundary conditions of
the simulation, ANSYS CFX
Solver Manager (Fig. 8) for
iteratively solving the partial
differential equations until a
predefined convergence cri-
terion is reached, and ANSYS
CFX Post (Fig. 9) for visual Figure 7 ANSYS CFX-Pre

and numeric evaluation of the


simulation results.

Figure 8 ANSYS CFX Solver

Figure 9 ANSYS CFX Post

6
4 Applications

4.1 Optimization of the in-


take flow of a stack

In this case, after modification ent coming from the left was
of a pressure-side duct lead- considerably higher than
ing to a stack and containing that entering the collecting
several branches (Fig. 10), vi- duct from below. As can be
brations occurred on one of clearly seen from Fig. 11, this
the fans, causing an automat- causes a severe constriction
ic safety shutdown. These vi- of the air stream entering the
brations had not been experi- collecting duct from the bot-
enced prior to the modifica- tom left. As a consequence,
tion. As it could be presumed there were pressure fluctua-
that the vibration problems tions in excess of 250 Pa in
were a consequence of non- this supply duct (Fig. 12), and
optimum design of the supply these throttled the fan so se-
duct, this was analysed in verely that it reacted with sig- Figure 10 Stack geometry with marked problem zone
detail by CFD. The stack is nificantly increased vibration
around 70 m high while the velocities.
horizontal duct is approx. 3 m
high, 2 m wide and has a total
length of approx. 25 m. The
mentioned vibration problems
affected the fan upstream of
the marked duct. It was estab-
lished that the fan vibrations
were caused by poorly de-
signed cross-sections. The Pressure gradient at measuring points
measured flow velocity in the
horizontal section of the duct
Pressure in Pa

was 22 m/s, while that in the


vertical section marked in Fig.
10 was only approx. 2-3 m/s.
With approximately compar-
able duct cross-sections, the Outlet F3B 1
Outlet F3B 2
flow momentum (mass flow x Outlet F3B 3
velocity) of the flow compon-

Time in s

Figure 12 Pressure pattern at measuring points

Figure 11 Velocity distribution in stack

7
Figure 13 Original geometry Figure 14 Duct with baffle plates

As the described unsatisfac- original situation (Fig. 11), so


tory conditions had to be re- that the gas streams meet
medied without significantly with identical velocities and
interrupting operation or al- no longer negatively affect
tering the design, it was ne- each other. Also, the pres-
cessary to eliminate the prob- sure fluctuations measured at
lem by the simplest means the points shown in Figures
possible. Without altering the 21 and 25 were reduced from
duct geometry (Fig. 13), this 250 Pa to below 10 Pa. Sub-
was achieved by installing sequent to the conversion
two baffle plates (Fig. 14), work, the upstream fan could
which accelerated the flow be run up to its rated speed
from the two lower supply without any problem. The vi-
ducts to the extent that the bration characteristics were
velocities of all three gas so greatly improved by the
streams are identical at the conversion work that meas-
point where they meet. This urements taken at the fan de-
measure succeeded in opti- tected no further effects on Figure 15 Velocity with optimized geometry
mizing the velocity distribu- fan operation.
tion (Fig. 15) compared to the

Pressure gradient at measuring points


Outlet F3B 1
Outlet F3B 2
Pressure in Pa

Outlet F3B 3

Time in s
Figure 16 Pressure gradient with baffle plates

8
4.2 Optimization of an
induced draught fan

In addition to optimization of the first branch the pressure


the fan itself, the upstream drop is approx. 260 Pa. From
and downstream flow situa- the inlet until the rear branch
tions are of great importance the drop even reaches 430 Pa
for the troublefree operation because of the flow disruption
of the fan and achievement at the rear of the horizontal
of the maximum possible effi- duct, as shown in Figure 19.
ciency. If, for instance, the in-
coming flow is turbulent or In Figure 20 the points of par-
swirling because of poorly ticularly high pressure drop
designed bends or changes are highlighted. The black
in cross-section, this inevit- circles indicate unfavourable
ably results in a worsening of situations because this design
the fan’s operating character- involves transitions that are Figure 18 3D model of induced draught

istics. Disturbances in the in- too sharp-edged and thus


flow and outflow zones have prevent the flow from follow-
particularly serious effects in ing the contours, resulting in
the case of fans with espe- turbulence. This causes pres-
cially high efficiency levels sure drops and thus leads to
> 80 %, as such fans depend additional expenses for elec-
on a non-swirling inflow in or- trical energy. By contrast, the
der to achieve their efficiency white circle indicates a duct
figures. As well as aiming at end producing an inefficient
optimizing the fan impeller fluid flow. At the marked rear
and casing it therefore makes end of the duct the flow forms
absolute sense to analyse a vortex (Fig. 19), which has
the inflow and outflow situa- several negative conse-
tions. This application is quences. Firstly, the flow is
therefore concerned with the subjected to pressure fluctu-
optimization of the induced ations because the vortex, as
draught flow of a double-inlet already described, wanders
fan. Figure 17 shows the sys- between the left-hand and
tem arrangement drawing. right-hand duct walls. Sec-
The 3D model derived from ondly, a portion of the vortex
this drawing is shown in Fig- is sucked through into the fan
Figure 17 System arrangement drawing with induced draught fan
ure 18. The very conspicuous (Fig. 21, view from below into
sharp-edged transitions the duct), so that the fan has
are very detrimental to the to cope with strongly swirling
achievement of smooth flow inflow air and therefore loses
guidance. This becomes efficiency and may suffer from
plain in the subsequent evalu- mechanical vibrations (reso-
ation of the flow simulation nance).
(Fig. 19), which reveals ex-
tensive flow disruption in the
rear area that is also unstable,
i.e. it oscillates between the
left-hand and right-hand
sidewalls and thereby ob-
structs a portion of the
branching-off rear induced
draught flow. The pressure Pressure loss outlet 1 = 261.365 [Pa] Pressure loss outlet 2 = 431.966 [Pa]
drop of this system is corres-
pondingly high, as is illus-
trated by the total pressure
profile shown in Figure 20. Figure 19 Flow lines
From the inlet until the end of

9
Pressure loss outlet 1 = 261.365 [Pa] Pressure loss outlet 2 = 431.966 [Pa]
0 10 20 30 40 50 60 70

Pressure loss outlet 1 = 261.365 [Pa] Pressure loss outlet 2 = 431.966 [Pa] Velocity [m s^-1]
0 1500 3000 [m]
1750 2250

Figure 20 Total pressure profile with original design Figure 21 Strong swirl with original design

Optimization of the geometry swirling of the fan intake air in a lower pressure drop. The
involved redesigning the criti- (Fig. 21, red circles) was also mean pressure drop saving of
cal points highlighted in Fig. significantly reduced (Fig. 23), 275 Pa at a volume flow of
20 so that the cross-section enabling the fan to achieve its 500000 m³/h significantly cuts
transitions were smooth. This rated performance figures. the power consumption. The
redesigning succeeded in re- resultant power saving of
ducing the pressure drop at When the flow lines of the 49 kW, assuming 24-hour
the front inflow duct to the fan original duct design and the operation at an electricity price
by a factor of four, from 261 Pa optimized duct design are of 10 cent per kWh, leads to an
to 66 Pa (Fig. 22), while the compared (Figs. 24 and 25) it annual electricity cost saving of
pressure drop at the rear in- is obvious that the flow turbu- 43000 €.
flow duct that had been caused lence has been completely
by the poorly-designed hori- eliminated and the flow velocity Related purely to the flow
zontal duct end (Fig. 20, white of max. 70 m/s of the original through the duct, the pressure
circle) was even reduced to design has been reduced to drop coefficient ζ, which is de-
little more than one sixth the approx. 55 m/s, resulting in cisive for the system pressure
original figure. The severe decreased swirling and also drop, has been halved.

Pressure loss outlet 1 = 65.8793 [Pa] Pressure loss outlet 2 = 76.8992 [Pa]
0 10 20 30 40 50 60 70

Pressure loss outlet 1 = 65.8793 [Pa] Pressure loss outlet 2 = 76.8992 [Pa] Velocity [m s^-1]
0 1500 3000 [m]
1750 2250

Figure 22 Total pressure profile after optimizing design Figure 23 Swirl after optimizing design

10
Pressure loss outlet 1 = 261.365 [Pa] Pressure loss outlet 2 = 431.966 [Pa]

Figure 24 Flow lines of original design

Pressure loss outlet 1 = 65.8793 [ Pa ] Pressure loss outlet 2 = 76.8992 [Pa]

Figure 25 Flow lines of optimized design

11
4.3 Optimization of a large
double-inlet fan DHRV 50

During the introduction phase max. 1.3. If these limits are ex-
of the software, its computa- ceeded, which takes place
tion accuracy was tested by more and more frequently, it is
comparing simulation and no longer possible to achieve
measurement using a test- maximum efficiency ratings by
bench fan of type HRV 63S means of the classical empiri-
and it was found that the simu- cal design methods for fans
lation deviated by less than (acc. to Prof. L. Bommes or
1% from the measured Dr.-Ing. Bruno Eck). However,
values. due to the substantial increase
in energy costs and also be-
The first results of an ongoing cause of the energy efficiency
optimisation of a large double- directives aiming at a reduction
inlet fan type DHRV 50B-2000 in CO2 emissions, it is impera-
are presented in the following. tive to achieve the highest pos-
This is an ongoing project sible fan efficiency rates. To Figure 26 Geometry DHRV 50B-2000

which has the aim of deter- enable this, it is either neces-


mining the optimization poten- sary to plan and carry out ex-
tial of the DHRV 50, particu- tremely expensive, time-con-
larly at high peripheral speeds suming and labour-intensive work and expenditure. For ex- ally describe the behaviour of
> 150 m/s. The background to test series or to employ CFD ample, measuring devices fit- a Newton fluid in a physically
this is the fact that for some tools, which are the state of ted onto the rotating impeller correct manner, CFD is cur-
years now, clients have been the art in turbomachinery con- would influence the flow in a rently the most effective tool
requesting higher and higher struction. One great advantage way that would falsify the for achieving further optimiza-
pressure differences together when using CFD for designing measurement results or are so tion of turbomachines. Particu-
with a high volume flow. These and optimizing fans is that it complex that they can only be larly in the field of compressor
requirements can only be ful- provides a detailed analysis of applied in academic studies construction, which is closely
filled if the speed and thus the every flow zone. Many of the (light sectioning principle). related to that of fan construc-
peripheral speed of the fan im- flow zones that are of interest However, in order to improve tion, CFD has become an in-
peller are increased. This to the design engineer, for in- the efficiency rates beyond the dispensable tool for ana-
means that the speeds and stance the leading edges of current status of > 80 %, it is lysing, for example, impellers
pressures occurring in the fan impeller blades, either cannot essential to obtain a well- with twisted blades, twisted
reach and exceed the design be represented in sufficient de- founded knowledge of flow leading edges and raked
limits normally defined in fan tail by conventional measuring dynamic processes in the im- trailing edges. The project
specifications of Mach 0.3 or techniques or would demand a peller and fan casing. Due to described below was the first
100 m/s at a pressure ratio of disproportionate amount of the ability of CFD to numeric- step in the optimization of a

Figure 27 Mesh of DHRV 50B-2000 with detail of leading edges of impeller blades

12
quency of 893 min-1, a pres-
sure difference of 8500 Pascal
and a mass flow of 145.13 kg
s-1 at a process temperature
of 300 °C, the transient cal-
culation of the case can be
started (Fig. 29).

After the specified conver-


gence criterion has been at-
Figure 28 Specification of the boundary conditions tained and the simulation has
been concluded, the results
are evaluated. Figure 30
depicts the flow lines in the
large double-inlet fan of type Fig. 27), in order to represent tional mesh of the impeller ro- impeller and the casing.
DHRV 50B-2000. the earlier described boundary tates during a transient simula- Every flow line represents the
layer with adequate precision. tion through a defined angle track of a single fluid particle
First of all, a 3D model of the In the case of the described for each of the individual com- through the entire calculation
inlet box, the impeller and the large double-flow fan of type puting steps (usually by 1-5° zone. The colour of the flow
volute casing (Fig. 26) has to DHRV 50B-2000, which has depending on the number of line indicates the local vel-
be created and then provided an impeller with an outer dia- blades), while the inlet boxes ocity of the fluid particle. With
with a computational mesh. meter of 3.2 m, the total num- and the volute casing remain the aid of this flow line dia-
During the discretization proc- ber of computational mesh stationary. When the boundary gram the design engineer
ess, special attention is paid to cells in all zones is approx. conditions are being entered, can identify points with par-
the most uniform possible dis- 8 million elements. If, as in the these zones then have to be ticularly high flow disturb-
tribution of the mesh elements, described case, there are adja- connected by interface (Tran- ances, excessively high vel-
which has to be oriented to the cent stationary and rotating sient Rotor Stator Interface). ocities or excessively sharp
occurring flow gradients. For zones, these have to be sepa- deflections. The diagram is
instance, the computational rately represented in the model After the boundary conditions also used for checking
mesh has to assure great de- and separately discretized. have been specified (Fig. 28), whether, for example, the air
tail on all the walls (detail in This is because the computa- including a rotational fre- flows to the impeller blades or

Figure 29 Solution of the differential equations

13
Figure 30 Flow line diagram Figure 33 Pressure distribution
in impeller and casing

the casing cut-off (the point at the casing, the casing cut-off, the impeller while approx. one basis of Figure 30, these
which the flow is divided) as where the flow is divided. It is third takes place in the cas- were projected as 2D flow
planned. extremely important that the ing, which thus acts as a lines onto a plane inserted in
flow divides at the centre of collecting plenum and diffuser. the centre of the casing. This
In addition to the flow lines, a the cut-off, as shown in Fig. One of the aims of the on- clearly shows that the vel-
diagram of the flow vectors is 32. If this does not happen, going optimisation of fan type ocities in the area of the cut-
a suitable way to visualize the the result is loss of fan effi- DHRV 50B is to achieve com- off (red circle in Fig. 34) are
characteristics of a technical ciency, vibrations and a sig- plete uniformity of the cur- slightly too high and that the
flow. The flow vectors indicate nificant increase in sound rently not entirely homoge- fluid does not flow away in
both the magnitude of the pressure level. neous pressure distribution at fully logarithmic manner
velocity and its direction. This the periphery of the casing. (black circle). In order to de-
is of great importance, espe- Figure 33 shows the pressure termine the optimization po-
cially in the case of flow turbu- distribution in the impeller and The optimization potential still tential at these points, differ-
lences. Figure 31 shows the casing. The right-hand image existing in the casing can also ent designs are currently
vectors at the leading edge of shows the inlet box. The pres- be seen from the flow lines being simulated and compa-
the impeller. Figure 32 shows sure distribution reveals that shown in Figure 34. To enable red. For instance, the casing
the flow vectors at one of the approx. two thirds of the pres- easier interpretation of the flow width is being varied in a
most critical fluidic zones of sure increase take place in lines than is possible on the number of increments in order

Figure 31 Vectors at leading edge of impeller Figure 32 Vector arrows on casing cut-off

14
Figure 34 Flow lines in 2D

to improve the flow and to are selectively optimized in


achieve a saving in material order to achieve a more uni-
and thus in machine weight. form total pressure distribu-
After this test series has been tion, the amount of incrusta-
concluded, the logarithmic tion will certainly be reduced.
function used for designing
the casing periphery will be
subjected to a series of
checks in order to achieve
improved flow guidance. In
addition to the casing, which
offers the greatest optimiza-
tion potential, the impeller is
also being analysed. One
possibility that has only re-
cently been added to the soft-
ware is to represent variables
as volumes in specified ranges.
Figure 35 Total pressure volumes
Using this software capability,
Figure 35 was generated. This
depicts the total pressure in
the impeller within a range of
-4000 to -6000 Pascal. The
shape assumed by the indi-
vidual volumes corresponds
very closely to photos of ma-
terial incrustations on impel-
lers conveying dust-laden me-
dia (Fig. 36). If these areas

Figure 36 Raw meal caking

15
A 1/ 09/ 10/ 1. Stu
5 Conclusions and
prospects

Subject to alterations
Within a period of one year important role in system en- In the field of turbomachinery
after the test phase, the CFD gineering and construction construction, CFD has, within
method had completely estab- because, besides enabling the space of just a few years,
lished itself at Venti Oelde. In purely aerodynamic assess- become an indispensable
addition to its primary function ment of a flow, it also allows tool for the optimization and
as a tool for optimizing and calculation of mixing proced- new design of turbomachines,
advancing the design of fans ures or heat transfers and as well as for the elimination
and their inflow and outflow distributions. of fluid-mechanic problems in
zones, the software plays an installed systems.

Industrial fans

Dust collection and process air cleaning plants

Exhaust air treatment plants


Ventilatorenfabrik Oelde GmbH
P.O. Box 37 09
Ventilating, heating and air conditioning plants
D-59286 Oelde
Phone: +49 25 22 75 - 0
Fax: +49 25 22 75 - 2 50 Recycling and waste processing plants
info@venti-oelde.de
www.venti-oelde.de Surface technology

You might also like