Академический Документы
Профессиональный Документы
Культура Документы
Fundamentals
SIEGFRIED RIPPERGER, Gonbach, Germany
WALTER GO€ SELE, Heidelberg, Germany
CHRISTIAN ALT, M€unchen, Germany
THOMAS LOEWE, Sartorius Stedim Biotech GmbH, G€ottingen, Germany
adheres to the separated solids (cake with Dynamic filtration and static (normal)
residual moisture) and the filtrate often contains filtration. In case of dynamic filtration are
some solids (solids content in the filtrate or during the filtration process mechanisms
turbidity). active which helps to reduce the build up of
The purpose of filtration may be clarifica- a filter cake. The most common dynamic
tion of the liquid or solids recovery or both. In filtration process is cross-flow-filtration
clarification the liquid is typically a valuable 4. Application. For example water filtration,
product and the solids are of minor quantity beer filtration
and are often discarded without further treat-
ment. If however, the solids are to be recov- Filtration is effected by application of a
ered, they very often have to be washed, pressure difference which can be produced by
deliquored and dried (see Fig. 1). In this a pressurized fluid, by a vacuum, by the gravity
article, washing means the cleaning of a prod- or by centrifugal force (see Fig. 2). Pressure
uct (filter cake) and it is distinguished from filtration typically requires a pump for deliver-
cleaning parts of the filter itself, which will be ing the suspension to the filter. Vacuum filtra-
called rinsing (e.g., rinsing a filter screen or a tion requires a vacuum pump. The pump
filter cloth by jets of water). A further distinc- evacuates the gas from a filtrate receiver, where
tion is to be made between washing and the filtrate is separated from the gas. The filtrate
extraction or leaching. Washing eliminates is drained either by a barometric leg of at least 8
liquid contaminants from the pores between to 10 m or by a pump that is able to run on snore
the particles of a filter cake. Extraction recov- (i.e. with a deficiency of feed liquid so that it
ers soluble matter out of the solid particles tends to draw in air). In some cases the liquid is
themselves (! Liquid – Solid Extraction). allowed to flow through the filter medium only
The term drying means thermal drying, while by gravity (gravity filtration). Centrifugal fil-
the elimination of liquid from the filter cake tration is done in perforated centrifuge rotors
by mechanical forces is called deliquoring or (! Centrifuges, Filtering).
dewatering, e.g., deliquoring by a pressurized In case of vacuum filters the cake is freely
gas or by expression. accessible. This facilitates automatic cake
Filtration processes can be classified in
accordance with different criteria:
handling. However, vacuum filters cannot han- deposited on the upstream side of the filter
dle hot liquids, or solvents with high vapor medium as a homogeneous porous layer with
pressure. The pressure difference across vac- a constant permeability. As soon as the first
uum filters is very limited, and the residual layer of cake is formed, the subsequent filtra-
moisture of the filter cake is higher than with tion takes place an the top of the cake and the
pressure filters. Pressure filters allow high pres- medium provides only a supporting function.
sure differences. They are preferred when the Thus, if the flow rate dV/dt is constant, the
product must be kept in a closed system for pressure drop will increase linearly, propor-
safety reasons, or if the residual moisture tional to the quantity of solid deposited. This
content is important. The handling of the filter model can be applied especially for all hard,
cake is obviously more difficult in a pressure particulate solids.
filter. Filtration by centrifugal force requires
more technical equipment, but as a general Blocking Filtration. The pressure drop is
rule it yields solids with lower residual moisture caused by solid particles blocking pores.
(! Centrifuges, Filtering). Soft, gelatinous particles retained by a sieve
During a dynamic filtration the collected exhibit such a behavior. If the flow rate dV/dt is
solids on the filter media are continuously constant, the pressure drop increases exponen-
removed, mostly with a tangential flow to the tially with the quantity filtered, the number of
filter medium (cross-flow filtration). Cross- open pores asymptotically approaching zero.
flow filtration is a standard operation with The pores may belong to a filter medium
membranes as a filter medium. The flow paral- (screen or filter layer) or it may be pores within
lel to the filter medium reduces the formation of a filter cake of coarse particles, which are
a filter cake or keeps it at a low level. So it is blocked by migrating fine particles.
possible to get a quasi-stationary filtrate flow
for a long time. Deep Bed or Depth Filtration. Solid parti-
Various models have been developed to cles are retained in a deep filter layer. This takes
describe the physical process of filtration. place for example in sand filters for clarifica-
This chapter concentrates on four idealized tion of drinking water, which retain even col-
filtration models depicted in Figure 3. loidal particles. The typical effect of deep bed
filtration is adhesion of solids to the grains of
Cake filtration is the most frequently used the filter layer. Only rather big particles are
model. Here it is assumed that the solids are retained by the screening effect. When the filter
The resistance to flow of a porous medium For practical reasons the viscosity h is very
(filter medium or filter cake) can be described often not measured separately. Then it is
by Darcy’s law [1] (see Fig. 4). Consider a legitimate to include it in a term aHh (unit
liquid flowing through a filter cake (or a stream mPa s/m2) or amh (unit mPa s m/kg),
of water percolating through soil as considered respectively.
by Darcy). The pressure drop Dp of this flow is Using this latter term aHh or amh filter
proportional to: resistances lie between 1011 mPa s/m2
6 Filtration, 1. Fundamentals
e is the porosity (void volume fraction) of the therefore to a large degree on particle sizes,
porous medium. Sv is the specific inner surface cake porosity and hence in case of fine particles
of the medium (filter medium or filter cake) on the surface forces producing this porosity.
related to the volume of solid mass. In case of a Filter media for surface and cake filtration
spherical particle system one can write: are selected according to their pore diameter
( hydraulic diameter). To reduce their hydrau-
6
Sv ¼ lic resistance they are produced with a high
dS
porosity.
dS is the Sauter diameter, an average diameter
of an particle size distribution, obtained by
dividing the total volume of a particle by the
total surface area. One can see that the specific
surface and the hydraulic pore diameter depend
strong on the particle sizes.
Based on a laminar flow inside a porous
system with pores of a hydraulic diameter dh
one get in combination with the Darcy equation
(1) an useful approximation for the specific
resistance of a porous medium:
ð1 eÞ2 36 ð1 eÞ2
aH 5 5 : ð3Þ
e3 d2S e3 S2v
V_ V_ ahK bh
Dp ¼ Dp1 þ Dp2 ¼ aH h H þ bh ð5Þ dt ¼ V dV þ dV
A A A2 Dp A Dp
or
aH hK H bh
t¼ V2 þ V ð10Þ
m V_ V_ 2A2 Dp A Dp
Dp ¼ Dp1 þ Dp2 ¼ am h þ bh
A2 A
or
If the suspension is homogeneously mixed,
the cake height H (or m/A) will be proportional am hK m bh
to the quantity of filtrate. The influence of the t¼ V2 þ V ð11Þ
2A2 Dp A Dp
concentration and cake porosity is described by
the factor KH:
2.2.2. Evaluation of Experiments with Linear
cv HA Diagrams
KH ¼ ¼ ð6Þ
ð1 eÞ V
represents the resistance at the very first If the pressure is constant during the experi-
moment of filtration, before a cake is formed, ment, a straight line should be obtained
hence the resistance of the filter medium
including the boundary layer to the cake. t
¼
am hK m
V þ
bh
The slope b contains the filter resistance a V 2A2 Dp A Dp
ð15Þ
according to aH hK H bh b
¼ V þ ¼ V þa
2A Dp
2 A Dp 2
D Dp dV
dt
ahK
b ¼ slope ¼ ¼ 2
DV A This kind of plot is still very popular in
practice because the evaluation is often easier
Inserting KH or Km (see Eqs. 6 and 7) than with Equation (12). Experimental values
respectively, leads to measured with a bucket and a stopwatch can be
inserted directly into Equation (15), while for a
dt dt
D Dp D Dp differential diagram they have to be converted
dV A2 dV AV
aH h ¼ ¼ into momentary flow rates dV/dt.
DV KH DV H ð13Þ
On the other hand Equation (15) is correct
dt A
¼ D Dp
dV e He
only for Dp ¼ const. Furthermore, the resulting
diagram shows less clearly the deviations from
and linearity than a diagram based on instantaneous
flow rates. If for example a cake stops growing
D Dp
dt
D Dp
dt and keeps a constant resistance this gives a well
dV A 2
dV A2 V discernible plateau in a differential diagram,
am h ¼ ¼
DV K DV m ð14Þ
m
but only a gradual hyperbolic bend in the inte-
dt A2
¼ D Dp grated diagram.
dV e me
In case of a short filter cycle, the evaluation
of the real beginning of filtration and the deter-
2.2.2.2. Linear Diagram in Integrated mination of the filter medium resistance could
Form be problematic. Tichy [8] showed, that the
Another approach starts from the integrated crossing point of the combined graphs of
filter Equations (10) or (11). The experimental Equation (12) and (15) gives a realistic starting
results are plotted according to point. He also showed, that the filter medium
resistance based on filter experiments is much
t higher than the filter medium resistance eval-
¼ f ðVÞ
V uated on the pressure drop of a clean liquid
Filtration, 1. Fundamentals 9
(water). The reason for that the increase is an The filter medium resistance can be calcu-
additional resistance based on a partial pore lated by
blocking of the filter medium during the first
layer formation. Dp A 105 20 104 1
b¼a ¼ 3:6 106 m
h 103
2.2.2.3. Example ¼ 7:2 1011 m1
The interpolation of the test results shown in
Figure 7 leads to the intercept a and the slope b If the porosity of the filter cake is e ¼ 0.4 and
the part of liquid inside the filter cake is neglec-
a ¼ 3:6 s=mL ¼ 3:6 106 s=m3 ted, the solid concentration (as a part of total
volume) in the suspension is:
and
A He ð1 eÞ
cv ¼ ¼ K H ð1 eÞ ¼ 0:289
28 s=mL Ve
b¼ ¼ 1:87 1011 s=m6
150 mL
From Equation (13) or (14) the cake filter
resistance can be calculated by
The following parameters are known or
measured:
A2 Dp 400 108 105 2
aH ¼ b ¼ 0:187 1011 m
Filtration pressure p ¼ 1 bar ¼ 105 N/m2 KHh 0:481 103
Filter area A ¼ 20 cm2 ¼ 20 ¼ 1:56 1014 m2
104 m2
Viscosity h ¼ 0.001 Pa s ¼
103 Ns/m2 A2 Dp 400 108 105 2
am ¼ b ¼ 0:187 1011
Cake height He ¼ 37 mm Kmh 158 103
m
Concentration factors:
2.2.2.4. Deviations from Linearity
4
A He 20 10 0:037
Experimental data very often differ from the
KH ¼ ¼ ¼ 0:481 linear characteristic shown in Figure 7. The
Ve 154 106
kind of deviation indicates the nature of sec-
or ondary effects (see Fig. 8):
dðDpÞ
¼ const ðDpÞq ð20Þ
dV
1. Particles larger than the pores are trapped here l0 is the initial filter coefficient of a clean
mechanically. This is typically true for par- filter medium. As soon as the medium is loaded
ticles 10 mm with solids, its efficiency will change and that is
2. Particles 1 – 10 mm in diameter hit the solid why the solution of the differential equation
surface mainly due to inertia effects and becomes rather difficult. Different models exist
14 Filtration, 1. Fundamentals
to describe the process, but they are rarely used be described by Boucher’s law (Eqs. 21
for practical purposes. To find a suitable filter and 22).
medium in a depth filter which shows good
retention efficiency over a long cycle time, The total pressure drop of a depth filter at
laboratory tests over a realistic cycle time constant flow rate can then be approximated by
have to be carried out with filter layers of an equation of the type
realistic depth. pðtÞ ¼ const1 t þ const2 eJt
The pressure drop in a depth filter can be
interpreted as an effect of two different phe- The pressure drop should be measured in a
nomena of blocking filtration: test filter with a vertical height close to that of
the full-scale unit (Fig. 12 A) [15]. For the case
1. The filter media exhibit a resistance to flow, that the driving force is gravity, the pressure
which is increased by the solids deposit in profile is shown in Figure 12 B. In the static
the pores. The quantity of deposit is gener- equilibrium, at flow zero, 1 m of pressure head
ally small compared to the pore volume, and is gained for every meter of depth. The down-
the additional pressure loss per unit depth stream consumers may then draw a constant
can be described as proportional to the local flow rate. When flow has started, the pressure
specific deposit. For constant solids concen- drop within the medium increases linearly with
tration at the inlet and constant retention the depth. As solids are deposited in the pores, the
increase in this pressure drop with time is local pressure loss will increase in the upper
therefore approximately constant. layers and the pressure line becomes distorted.
2. In addition there is often a pressure drop due When the pressure line touches the atmospheric
to deposits on the surface of the filter bed. pressure value, the required flow cannot be
This is a typical blocking filtration and can maintained and the filter should be cleaned.
2.4.2. Cleaning and Sizing of Deep Bed becomes clogged rapidly by captured solids.
Filters This disadvantage is diminished when a second
layer of coarse material with lower specific
Depth filters in form of deep bed filters (e.g., weight is added on top of the sand layer. During
sand filters) [16, 17] can be cleaned by back- backflow these low density grains collect at the
flushing. The flow is reversed to wash off bed surface and form a coarse layer able to
deposited solids and flush them, e.g., to a retain a great quantity of dirt (so-called multi-
wastewater station. Often air scour is used to media filters or dual media filters, e.g., with a
increase the movement of the particles and layer of anthracite particles 1–2 mm in size on
reduce the amount of water necessary for flu- top of a main layer of sand grains 0.5–1 mm in
idization. A typical set of data of a sand filter size.)
for cleaning of water could be [18]: Other designs of sand filters convert the
disadvantage of classification into an advantage
Medium depth 1m
by operating in the upflow mode with filtering
Medium size 0.5–1 mm (as uniform as possible) and back-flushing from below. This offers the
Flow rate during 5–10 m3 m2 h1 advantage that large quantities of incoming dirt
filtration are retained in the lower layer of sand, which
Flow rate air scour 60 m3 m2 h1 for 10 min
Flow rate water 15–30 m3 m2 h1 for 3–8 min after air
contains the coarser grains and larger pores.
backflow scour These upflow filters however need special safe-
guards against breakthrough of dirt when the
pressure drop becomes high.
During backflow the sand grains are classi- In order to increase the retention capacity in
fied, the smallest ones are entrained to the top, a given volume, it has also been proposed to use
the coarse ones sink to the bottom of the filter porous grains instead of sand grains. Small
bed. This effect is undesired, since it leads to pieces of polymeric foam have indeed this
premature blocking of the top layer. Therefore effect [19]. Cleaning by simple back-flush is,
it is important to use sand with grain sizes as however, not possible with this material, it must
uniform as possible. be cleaned by back-flush combined with com-
The sizing of a deep bed filter consists of the pression. Therefore this method has not yet
following steps: found practical application.
Quasi-continuous operation can be achieved
1. Preselection of the filter material by com- using two ore more conventional deep bed
parative laboratory tests with small layers of filters in parallel. Truly continuous deep bed
different materials under standard condi- filters operates in the normal manner expect a
tions and with the filtrate quality as a portion of the filter bed is continuous or inter-
criterion mittently removed from the lower layer,
2. Selection of the right grain size in pilot tests cleaned to removed the contaminant and then
with a filter column of 1 m height and a automatically returned to the top of the filter
realistic flow rate and pressure drop bed. The cleaning and transport of the particles
is mostly realized in the same time form of a
The minimum diameter of the cylindrical hydraulic transport from the bottom to the top
bed is 150 mm. If necessary the grain size must of the filter bed.
be adapted so that the bed is saturated (the
filtrate gets turbid) at the same time as the
maximum pressure drop is reached. If the max- 2.5. Cross-Flow Filtration
imum pressure drop is reached first, the grains
should be coarser and vice versa. The ability to Cross-flow filtration is a form of a dynamic
clean the filter bed by back-flushing should also filtration. In cross-flow filtration the build-up of
be tested. a filter cake on the surface of the filter media is
As mentioned above the sand grains are hindered by a strong flow tangentially (parallel)
classified during back-flush and the uppermost to the filter surface. Clear liquid passes through
layer is most efficient in collecting and the filter medium (mostly a membrane) and the
16 Filtration, 1. Fundamentals
influence retention [22–27]. Therefore, reten- to as “filter intergrity tests” are available. The
tion has to be tested as close as possible to the most common are the “bubble point”, the
individual sterilizing grade filtration process “diffusive forward flow”, and the “pressure
and retention testing is therefore an important decay test”.
part of the required filter validation in pharma-
ceutical processes. The method of testing reten- 2.6.5. Regulations for Medicinal and
tion is referred to as “bacterial challenge test”. Pharmaceutical Manufacturing
In general, a defined quantity of microorgan-
isms suspended in a fluid is filtered under The most important general regulations for
defined conditions through a sample of the medicinal and pharmaceutical manufacturing
sterilizing filter which could be: (i) membrane including the use of sterilizing grade filters
material in suitable holding devices, (ii) small are the US 21 Code of Federal Regulations
disposable filters (capsules), or (iii) in rare Part 211 [29] and EU EudraLex [30]. These
cases also large filters as filter cartridges. are assisted by more detailed guidelines as the
The filtrate is tested for presence of micro- US FDA Guidance for Industry [31]. Technical
organisms, most often by conducting the entire reports like “Sterilizing Filtration of Liquids”
filtrate through an analytical filter retaining [22] of the Parenteral Drug Association, PDA,
present microorganism. The analytical filter are focussed on detailed recommendations for
is placed on nutrition media and cultivated to sterilizing grade filtration for specific applica-
grow single organisms to colonies in order to tions and their validation.
quantify and identify those. The resulting num-
ber of colony forming units (CFU) which is
present in the filtrate should be zero as by 2.7. Virus Filtration
definition for a “sterilizing grade” filter. Since
retention is also dependent on the number of 2.7.1. Introduction
microorganisms in the fluid applied to the filter,
the log10 reduction value (LRV) is defined by Therapeutic products manufactured by bio-
log10 of CFU applied to the filter divided by technological processes often use raw materials
CFU in the filtrate. As well as the “centering of animal or human origin as, e.g., cells, com-
test method”, ASTM F838-05 [28], defines the ponents from transgenic animals, natural
challenge level of bacteria challenge tests by > extracts, and human or animal blood plasma.
1107 CFU per square centimeter effective These raw materials represent a risk of
filtration area and, therefore, many sterilizing contamination by known or unknown patho-
grade filters are described with LRV > 7. gens [32–34] and regulatory agencies require a
ASTM F838-05 defines also the microorganism demonstration of viral safety prior to clinical
Brevundimonas diminuta, (American Type Cul- use and/or marketing of biopharmaceuticals
ture Collection, ATCC19146), for the test. His- [34–36]. Therefore, biotechnological manufac-
torically, the denomination of 0.2 or 0.22 mm turers are asked to implement robust viral
pore size is often used for a sterilizing grade clearance techniques into biotherapeutic purifi-
filter qualified by the ASTM F838-05 test. cation processes in order to clear viruses with-
Although some microorganisms are known to out compromising the integrity of the products.
penetrate filters more easily than Brevundimo- Virus filters (often incorrectly termed
nas diminuta, no standard test which show a “nanofilters”) are specifically designed to
correlation between microorganisms and reten- remove viruses from the product (protein) solu-
tion exists. A certain common understanding tion through a size-exclusion mechanism. In
exists to use Acholeplasma laidlawii this way, they can provide a robust viral clear-
(ATCC19146), as representative mycoplasma ance mechanism that complements other virus
organism. clearance steps in the production of biothera-
The bacteria challenge test is a complex peutics [37, 39]. Virus filters devices are typi-
procedure and not applicable to large quantities cally constructed from multiple layers of
of filters, thus faster tests with a correlation stacked flat sheet membrane or from hollow-
between microorganisms and retention referred fiber membranes. Virus filtration is only one
20 Filtration, 1. Fundamentals
element of a virus safety strategy that each macroporous substructure before they reach
manufacturer has to install. Complementary the virus retentive skin layer. Furthermore,
to virus removal by filtration, i.e., by size the operation of a single-use NFF device
exclusion, viruses can be inactivated by expo- reduces system design, validation, and cleaning
sure to heat, UV-C irradiation, low pH values, cost, as well as capital investment [40].
solvents, and detergents.
2.7.3. Virus Removal
2.7.2. Virus Filter Membranes
Due to the fact that membranes retain viruses
Similar to sterilizing grade filters, virus filters based on their size, their action is referred to as
are cast polymeric membranes manufactured a robust virus removal step which is indepen-
from a wide range of polymer materials. Avail- dent to changes in virus type, solution and
able virus filters are made from regenerated operating conditions. Still, a reliable virus
cellulose (RC), PVDF and PES. Engineering removal process requires the knowledge of
polymers like PVDF or PES have the advantage significant parameters of the process that can
of forming mechanically stable high-perform- provide consistent virus removal. Those param-
ance membranes, however, typically exhibit eters include, e.g., solution conditions like pH
hydrophobic surface properties, leading or ionic strength, blocking propensity of the
to accelerated fouling and protein binding feed stream, presence of aggregates, or applied
[37, 38]. These membranes are thus hydrophi- transmembrane pressure.
lized through surface modification techniques. The probability that any individual virus will
Further discriminating features are the mem- pass through the filter is considered identical to
brane format (flat sheet vs. hollow fiber), the the fraction of a virus population that will pass
nominal pore size of the membrane, its pore through the filter. A spiking study using feed
size gradient, its virus retention capacity, as streams with high virus concentration (high
well as its ability to transmit proteins. virus titer) is used to project the expected virus
Despite the common misnomer removal of low-titer challenges representative
“nanofilters”, virus filter membranes may range for industrial operation. The likelihood for
from fine microporous membranes with pore retention can be expressed by the LRV value.
ratings < 0.1 mm to coarse ultrafiltration mem- Depending on the nature and potential for viral
branes with nominal molecular-mass cut off contamination of the starting material, regula-
(NMWCO) ratings > 100000 kDa. Virus filtra- tory agencies demand a respective total
tion membranes are occasionally (but required LRV for the manufacturing process.
incorrectly) referred to as “nanofiltration mem- Therapeutics produced from cell lines contain-
branes” based on their 20–70 nm pore size. ing retroviruses will typically require higher
When virus filters were first established in LRV. Typically virus filtration steps provide a
the biotechnology industry, they were operated 4-log virus removal. In order to claim this LRV,
in a tangential-flow filtration (TFF) mode, end users perform spiking studies using high-
where the feed flows parallel to the tight skin titer infectious viruses. A panel of enveloped
layer of the asymmetric membrane. The TFF and nonenveloped viruses of different size are
mode of operation is established in ultra- selected and the respective LRV of a scaled-
filtration processes, as it provides considerable down virus filter device is evaluated. Common
performance improvements by reducing con- model viruses include animal parvoviruses
centration polarization and consequently mem- (e.g., PPV or MVM), poliovirus, SV40, sindbis
brane fouling. The ease of use and lower capital virus, and retrovirus. Retrovirus filters are
cost of normal-flow filtration (NFF) led to the designed to remove only larger viruses (greater
displacement of TFF processes. Today, virus than approximately 50 nm) while parvovirus
filters are used like a typical sterilizing grade filters are designed to provide significant
filter in a NFF mode, operated with the more removal of viruses as small as 20 nm. Virus
open side of the membrane facing the feed filter manufacturers often use bacteriophages as
stream. This allows foulants like, e.g., protein practical and economical human virus surro-
aggregates to be captured within the gates filter development. Bacteriophages,
Filtration, 1. Fundamentals 21
which are obtained at much higher purity and process filter is properly installed and free of
titers, demand less safety measures during han- gross damage or defects that would diminish its
dling and are considerably easier and faster to virus retentive performance. Furthermore, the
assay. pre-use test mitigates the risk of having to
Ultimately, a virus filter can only be selected reprocess the feed stream should the post-use
based on its retention performance with mam- integrity test fail. The post-use integrity testing
malian viruses under actual drug product and is required to insure that the claimed virus
process conditions. Spiking studies are used to clearance was in fact achieved since it is not
demonstrate the ability of a virus filter to retain feasible to determine the virus content in the
a particular virus under specified conditions. In filtrate at the low levels required for human
these studies, scaled-down virus filter devices pharmaceuticals produced from mammalian
are used, and the feed stream is spiked with high cells (< 1 virus particle per million doses).
titer viruses. Virus filter ratings provided by Today, most filters are tested by a non-
filter manufacturers should be used only as destructive air diffusion test before and after
nominal guides, and users should not over use.
emphasize slight differences in rating values.
Figure 16. Concentration of contaminant in the effluent of a filter cake during washing
a) Idealized piston flow; b) Effect of axial dispersion; c) Effect of diffusion from dead-end pores; d) Effect of contaminant
extracted from the solid
volume is several times the pore volume. The this is impossible since the required residual
washratio can be expressed as: contents in the cake are much smaller than the
precision of such a mass balance. Nevertheless
Vw
>1
analyses of the filtrate are often used to judge
V pore the progress of washing. This is possible if an
empirical correlation between the two concen-
Curve c is still more realistic; taking into trations in the filtrate and in the cake exists for
account that a filter cake contains stagnant the particular process. The validity of such
zones or dead-end pores, which are not reached empirical correlation is, however, restricted
by the flow. They deliver their contaminant to the particular equipment and operating
content by diffusion to the effluent. Diffusion parameters.
is an asymptotic process and depends on the
time elapsed, not on the quantity of liquid. This
curve is typical for most washing processes. 3.2. Example of Experimental Results
The first part depends on the washratio, the end,
however, is an asymptotic process depending An illustrative example of a washing process is
on time. given in Figure 17 A and B [42]. A kaolinite
When the curves a, b and c were drawn, it suspension contaminated with NaCl was fil-
was supposed that the mass m of contaminant tered to varying cake thickness and washed
eliminated was contained in the pore volume with varying quantities of water. After every
Vpore. The integral below the curves is therefore test the residual salt content in the cake was
the same for all three curves: measured and the results are reported as a
function of washing time and of wash ratio.
Z1
m¼ c dV ¼ c0 V pore
(In contrast to Fig. 16 the wash-ratio is related
0
to the solids mass, not the pore volume. Also the
residual content of the cake is reported, not the
Curve d finally depicts an example where the concentration in the filtrate)
washing resembles an extraction (! Liquid – Figure 17 A shows how at the very beginning
Solid Extraction) because the contaminant is of the washing process the residual concentra-
not contained in the pores only, but also in the tion depends on the wash ratio, disregarding the
solid (liquid inclusions or soluble solid matter). cake thickness. In this example this is true up to
The integral under curve d is therefore not the dash-dotted line which corresponds to less
related to the pore volume Vpore but to a higher than a one-fold displacement of the pore vol-
total quantity of contaminant. ume. As the wash ratio increases to several
Theoretically such curves could be used to times the pore volume, the mechanism of axial
determine the residual mother liquor content in dispersion becomes more important, and the
the cake from analyses of the filtrate. In practice effect of a given quantity of wash liquid
Filtration, 1. Fundamentals 23
becomes better for thick cakes than for thin in the cake and the momentary concentration in
ones. the effluent will vary stochastically. Scale-up
Figure 17 B shows how at the end of an from a laboratory test filter with only small-
experiment the washing time is the most rele- scale fingering can be misleading as well as the
vant parameter for the result. In this example results of few small samples taken from a big
this seems to be true below a residual concen- cake.
tration of
Instabilities Shrinking. If small electrodes
c are placed in a filter cake during the washing
103
c0 process, they indicate wildly varying local
conductivity. This reveals the existence of small
cracks in changing size and positions even in
3.3. Test Procedures and Pitfalls cakes with no visible fissures or shrinkage. This
is not surprising because the contact forces
As can be seen in Figure 17, washing of filter between the particles change dramatically
cakes gives scattering results even under labo- when the ion content of the surrounding liquid
ratory conditions. This explains why only few is washed off (see Chap. Interparticle Forces
experimental results are found in the literature: and Forces Between Particles and Filtermedia,
many experimentators are discouraged by DLVO Theory). The phenomenon has not yet
inconclusive test results and never publish been investigated thoroughly, but its stochastic
them. The reason of scattering results are sto- effects on the washing results should be similar
chastic influences: to the above-mentioned fingering.
If shrinking during washing presents a seri-
Fingering. In most washing processes the ous problem, it can be helpful to reslurry the
wash liquid has a lower viscosity than the cake in wash liquid and filter it again. Many
mother liquor. It tends to flow through the nutsche filters (! Filtration, 2. Equipment,
cake in finger-like streams past isles of viscous Chap. 3) are used this way. The second cake
fluid, as shown in Figure 18 [42]. This is a (with less ions in the liquid during cake build-
stochastic process, and the local concentration up) is often less porous and less prone to
24 Filtration, 1. Fundamentals
Figure 21. Test installation for measuring the capillary pressure pk [45]
a) Filter vessel; b) Filter cake; c) Semipermeable membrane; d) Draine; e) Entry for gas
26 Filtration, 1. Fundamentals
saturated filter cake (b) is placed in the filter Further increase of the pressure displaces
(a) on a filter media (c) (e.g., semipermeable liquid from finer pores and reduces the moisture
microporous membrane, the membrane is per- down to an irreducible saturation S1 after
meable for liquid, but it is impermeable for gas infinite time beyond which no further reduction
at the applied pressure). Gas pressure is applied is reached. This corresponds to the amount of
via port (e). The quantity of liquid displaced liquid, which is trapped in isolated domains
from the cake is drained by the valve (d) and within the cake.
collected. The standpipe is used to obtain an Figure 22 also shows the imbibition curve.
exact liquid level when the pressure gauge is When the pressure is reduced again, and pro-
adjusted to zero. vided that the bottom of the cake is still in
A typical capillary pressure curve is shown contact with the expelled liquid, then the cake
in Figure 22 (material: incompressible cake of will be reimbibed by capillary suction. At equal
glass beads with a mean diameter of 79 mm and moisture content the suction pressure will be
water). Starting from the completely saturated smaller than the capillary pressure difference
cake (S ¼ 1) the gas pressure is slowly for deliquoring. This is easily explained
increased, at each set point the equilibrium is because the capillary pressure for imbibition
established and the quantity of displaced liquid depends on the larger waist-diameter of the
is registered, yielding a point of the deliquoring pores (see Fig. 22):
curve. Liquid is displaced from the pores when
4 s cosd
the pressure exceeds a certain threshold pres- pci ¼
dpore;waist
sure. This is the capillary entry pressure pce (in
this example 0.062 bar). Each time when a pore Also imbibition does not go to full satura-
is emptied, the gas pressure overcomes the tion, because some air remains trapped in the
capillary pressure at the neck of the pores pores.
with a diameter dpore,neck: The following conclusions can be drawn
from the capillary pressure curve:
4 s cosd
pce ¼ ð23Þ
dpore;neck 1. Deliquoring is caused by pressure difference
between gas and liquid. Gas flow is nor-
mally required to maintain this pressure
difference, but not for the deliquoring itself.
No gas flow is necessary if the cake is placed
on a semipermeable membrane, which is
impermeable for gas. (This principle can
also be used technically for dewatering
without gas flow [46])
2. A certain threshold pressure must be
exceeded before liquid is displaced from
the pores; this is the capillary entry pressure
pce. This implies technical consequences:
3. Vacuum filters are restricted to pressure
differences < 1 bar. This means that cakes
cannot be deliquored by suction if their pce is
near or above this limit. A pressure filter
with Dp > 1 bar has to be considered for
such cakes, even if its installation is much
more expensive than that of a vacuum filter.
Theoretically suction can dewater filter
cakes with pores > ca. 3 mm: with pure
water and a small contact angle (d 0)
Figure 22. Example of a capillary pressure curve for Equation (23) yields p ¼ 1 bar for dpore ¼
imbibition and for deliquoring [45] 2,9 mm.
Filtration, 1. Fundamentals 27
Filter cakes of submicronic particles like liquid flow simultaneously through the cake,
wastewater sludge have such a high capillary the pores filled with liquid and the pores filled
entry pressure that their pores cannot be dewa- with gas form two separate capillary systems,
tered mechanically: with pure water and a small each of them with reduced permeability. The
contact angle Equation (23) yields p > 10 bar local permeabilities (for liquid or air) in relation
for dpore < 0.29 mm. to the permeability of the saturated cake are
called relative permeabilities krel. They depend
4.1.2. Kinetics of Deliquoring by Gas on the local saturation of the cake and can be
Pressure represented as:
for the wetting fluid (the liquid)
In most cases there is no semipermeable mem-
brane below the cake but a normal filter kw a
krel;w ¼ ¼
medium that is permeable for gas, like a filter k aw
S
SR ¼
1 S1
dimensionless time
pce t
Q¼
aH h H2 ð1 S1 Þ e
Figure 27. Residual saturation of filter cakes dewatered by centrifugal force as a function of dimensionless time Q0
[50]
30 Filtration, 1. Fundamentals
Figure 28. Residual saturation of filter cakes dewatered by gas blowing as a function of dimensionless time Q0 [50]
(this is why the fit is generally rather good, see a problem of great practical impact. The mech-
Figure 29 [50]). anism of crack formation is represented in
Figure 30. During filtration the viscous forces
4.1.5. Shrinking and Cracks in from the flowing liquid compress the cake. As
Filter Cakes described in Section 2.2, only the layer near
the filter medium is subjected to the full amount
Deliquoring of compressible filter cakes by gas of pressure, while the “upper” layer of the cake
pressure very often produces cracks in the cake, remains uncompressed (Fig. 31 (1)). When
Figure 29. Residual saturation of filter cakes dewatered by centrifugal force as a function of the modified dimensionless time
Q1 [50]
expressed, is [55] the straight line from the start of this regen-
eration to the end of filtration is tangential to
H 1 HðtÞ the yield curve. In other words: the filtration
Uc ¼ ¼ 1 B expðC tÞ
H1 H1 should be stopped when the instantaneous
flow (as indicated by a flow meter) falls
where Uc is the consolidation rate, H (t) the below the mean throughput according to the
cake thickness, H1 the original cake thickness definition:
and H1 the thickness after infinite time. B and
C are creep constants and tc is the consolidation V
V_ mean ¼ ð24Þ
(¼ compression) time. A comparison of differ- t þ treg
ent compression models can be found in [56].
This criterion can for example be
implemented in a computerized process
5. Optimal Filtration Cycle Time control. For design purposes it can be
useful to calculate the optimal cycle time
When sizing a filter installation, the right fil- in advance. For this purpose a rule-of-
tration cycle time has to be chosen (example: thumb will be derived assuming cake fil-
the same task can be performed in a small filter tration and negligible resistance of the
which must be cleaned frequently or a bigger filter medium (b ¼ 0). The quantity of
one with longer cycles). With a long filtration filtrate according to Equation (10) or (11)
time the filter cake becomes thick and the flow is then
rate per filter area declines. On the other hand
for very short filtration the downtime for fre- V ¼ c1 t1=2 ð25Þ
quent cleaning will reduce the capacity. This
leads to two questions: (1) What is the optimal and Equation (24) becomes
cycle time with the highest overall throughput?
(2) What is the optimal cycle time with the c1 t 1=2
V_ mean ¼
lowest overall cost? t þ treg
1. Figure 32 shows the quantity of filtrate The highest V_ mean is obtained if the
produced as a function of time. Before the filtration is stopped at topt, which is defined
start of filtration a time treg is needed for by the differential equation
regenerating the filter (extracting the cake
from the previous cycle, cleaning the filter, d 1
¼0¼
1 1=2 treg 3=2
t t
rearranging the filter elements, etc.). The dt V_ mean 2c1 opt 2c1 opt
highest overall throughput is obtained when
This leads to the rule-of-thumb for
maximum throughput:
topt;1 ¼ treg
Regeneration represents a considera- of this charge z is 120 mV < z < 120 mV, and
ble cost, including downtime, labor, new it depends on the pH of the liquid. The charge
filter elements, and disposal of waste. If reaches relatively far, depending on the ion
this cost is called creg, the mean specific content of the surrounding liquid. The reach
cost is is characterized by the Debye length rD. In pure
distilled water this length is exceptionally long
cost c2 t þ creg with 0.9 mm. In most technical aqueous solu-
¼
filtrate V tions it is much smaller, in sea water for exam-
ple only 0.4 nm [57].
With the approximation of Equation
(25) this is Attraction. Van der Waals forces are part of
the cohesive forces between the molecules
cost c2 1=2 creg 1=2
¼ t þ t within the solid particle. They reach also a
filtrate c1 opt;2 c1 opt;2
certain distance beyond the surface, but their
The minimum is defined by range is less than that of electrostatic forces.
The decline is proportional to the distance
d cost c2 1=2 creg 3=2 according to
¼0¼ t t
dt filtrate 2c1 opt;2 2c1 opt;2
const
Eattraction ¼
As result the rule-of-thumb for maxi- r
mal yield per cost is obtained:
When two particles approach to distance
c2 topt;2 ¼ creg
zero, these attractive forces become very strong
and should in principle become equal to the
or cost for the filter run ¼ cost for regen- cohesive force inside the particle. The reality is,
erating the filter. however, more complicated because there are
always absorbed molecules on the surface.
These rules have been derived here for con- An example of the superposition of both
stant pressure filtration. They can also be forces according to DLVO theory is represented
derived for filter runs with different but con- in Figure 33. The graph shows the resulting
stant flow rates until a pre-set final pressure. energy potential. A positive gradient of the
Practitioners therefore apply the rule-of-thumb curves describes attraction, negative gradient
rather generally (e.g., also for filtration with a repulsion. At very small distances, attraction
pump with a characteristic curve anywhere
between constant pressure and constant flow).
prevails. At larger distance electrostatic repul- enhance mutual adhesion and the formation
sion prevails (if it is not zero). of loose flocs.
Attraction and repulsion between two parti- In comparison to particles with the same
cles in suspension. The superposition of elec- surface charge, it is covered with a repulsive
trostatic repulsion and attraction by van der skin of a thickness near the Debye length. This
Waals forces is represented by the resulting skin has no roughness and no friction; hence it
energy potential. High ion content in the sur- is rather slippery
rounding liquid reduces the reach of electro- The surface charge of suspended particles
static repulsion (Debye length) can be measured as a zeta potential (! Col-
loids; ! Emulsions). This potential varies with
Interparticle contact. Electrostatically the suspending liquid, especially with its pH:
charged particles repulse one another. In com- High pH means a high concentration of OH
parison to their equally charged neighbors they ions, which can be absorbed, creating a nega-
show some kind of smooth and slippery repul- tive surface charge. Low pH means high con-
sive skin. (The smooth repulsive skin can be centration of protons and a positive surface
made visible with modern atomic force micros- charge. At a certain pH the surface is neutral
copy, see [58].) The Debye length characterizes (point of zero charge, isoelectric point). At this
the thickness of this skin (see Fig. 34). particular pH, the particles can more easily
Without electric charge, there is no such agglomerate and are easier to filter. This opti-
skin. The particles touch their neighbors with mal pH is often determined empirically without
their rough surfaces and tend to adhere due to knowledge of the zeta potential.
van der Waals’ forces.
Particles in aqueous suspension with weak
electric charge most probably have oppositely 7. Mathematical Simulation of
charged patches on their surface what may Filtration and Cake Formation
Different attempts have been made to simulate
numerically the structure and porosity of filter
cakes. Until now such calculations are
restricted to spherical particles with homoge-
neous surface. Starting from randomly chosen
locations the trajectories of particles in a filter
flow are calculated. When they touch the cake
surface, the deposition is simulated using either
a sticking angle [59, 60], or a friction angle and
an adhesive force [61, 62], or by estimating the
interparticle forces from van der Waals’ and
electrostatic forces according to the DLVO
theory [63]. After calculating the deposition
of a large number of particles, the packing
density of the resulting cake is found (see
Figure 35 A and B). These calculations help
to understand and explain some empirical facts:
suspension of filter aid. Then the solution to 21 S. Ripperger, J. Altmann, Filtrieren Separieren 11 (1997)
be cleaned is filtered through this layer. no. 3, 109–113.
22 Sterilizing of Liquids, Tech. Rep. Supp. PDA J. Pharm. Sci.
After filtration the precoat layer is broken Technol. 62 (2008) no. 62, 5.
in half. The dirt should form a separate layer 23 S. Sundaram et al: “Application of Membrane Filtration for
on the top and should not have penetrated Removal of Diminutive Bioburden Organisms in Pharm. Prod-
into the precoat ucts and Processes”, PDA J. Pharm. Sci. Technol. 53 (1999)
186–201.
3. Selection of the quantity of filter aid. As a 24 N. Starbard: Beverage Industry Microfiltration, Wiley & Sons;
first guess the quantity of filter aid is calcu- New York 2008.
lated which gives a cake volume equal to the 25 M.W. Jornitz, T.H. Meltzer: Filtration and Purification in the
Biopharmaceutical Industry, 2nd ed., Informa Healthcare, New
volume of dirt to be separated. It is then York 2008.
admixed to the suspension and the filterabil- 26 M.W. Jornitz, T.H. Meltzer: Sterile Filtration, Marcel Dekker,
ity of this mixture determined. If the flow New York 2001.
rate is to low, the next trial should be made 27 P.S. Stinavage: “Validation of the Filter and of the Filtration
Process”, in T.H. Meltzer, M.W. Jornitz (eds.): Filtration in the
with a higher quantity of filter aid Biopharmaceutical Industry, 2nd ed., Informa Healthcare, New
York 2008, pp. 371–388.
28 Committee D19: “F838-05 Standard Test Method for Deter-
References mining Bacterial Retention of Membrane Filteres Utilized for
Liquid Filtration”, American Society for Testing and Materials
1 H. Darcy: “Les fontaines publiques de la ville de Dijon,”Paris (ASTM), West Conshohoken, PA 2005.
1856. 29 21 CFR Food and Drug Administration, US Code of Federal
2 J. Kozeny, Ber. Math.-naturwiss. Abt. Akad. Wien. (1927) Regulations, Office of the Federal Register National Archives
271–306. and Records, Washington 2001.
3 A. E. Scheidegger: The Physics of Flow Through Porous 30 EudraLex: “The Rules Governing Medicinal Products in the
Media, 3rd ed., University of Toronto Press, Toronto 1974, European Union”, Vol. 1–5, EU Guidelines to Good Manu-
pp. 137ff. facturing Practice, Medicinal Products for Human and Veteri-
4 H. Rumpf, A. R. Gupte, Chem. Ing. Tech. 43 (1971) no. 6, nary Use, Commission Europeenne, Bruxelles 2001.
367–375. 31 Guidance for Industry, Sterile Drug Products Produced by
5 W. G€ osele, Filtrieren Separieren 9 (1995) no. 1, 14–22. Aseptic Processing, Current Good Manufacturing Practice,
6 VDI-Richtlinie “Filtrierbarkeit von Suspensionen” VDI 2762, Food and Drug Administration, Rockville 2004.
1997. 32 M.E. Chamberland: “Emerging Infectious Agents: Do They
7 “Filtrierbarkeit von Suspensionen,” VDI-Richtlinie 2762 Pose a Risk to the Safety of Transfused Blood and Blood
(1997). Products?”, Clin. Infect. Dis 34 (2002) 797–805.
8 J. Tichy: Fortschritt–Berichte VDI, Reihe 3, Nr. 877, Sontho- 33 K. Brorson, L. Norling, E. Hamilton, S. Lute, K. Lee, S.
fen (2007). Curtis, Y. Xu: “Current and Future Approaches to Ensure
9 L. Svarovsky: Solid–Liquid Separation, Butterworths, London the Viral Safety of Biopharmaceuticals”, Dev. Biol. 118 (2004)
1977, p. 175. 17.
10 M. Tiller, T. Cleveland, R. Lu, Ind. Eng. Chem. Res. 38 (1999) 34 M. Farshid, R.E. Taffs, D. Scott, D.M. Asher, K. Brorson:
590–595. “The Clearance of Viruses and Transmissble Spongiform
11 P. H. Hermans, H. L. Bredee, Rec. Trav. Chim. 54 (1935) 680– Encephalopathy Agents from Biologicals”, Curr. Opin. Bio-
700. technol. 16 (2005) 561–567.
12 A. Rushton, A. S. Ward, R. G. Holdich: Solid–Liquid 35 Note for Guidance on Virus Validation Studies: The Design,
Filtration and Separation Technology, VCH Verlagsgesell- Contribution and Interpretation of Studies Validating the
schaft, Weinheim, Germany, 1996, pp. 64 ff. Inactivation and Removal of Viruses, CPMP/BWP/268/95,
13 K. Luckert, Wiss. Z. Techn. Univ. Magdeburg 36 (1992) no. 5/6, European Medicines Agency, London 1995.
74–80. 36 K. Brorson, P.G. Swann, J. Brown, B. Wilcox, M. Shapiro:
14 P. L. Boucher, J. Proc. Inst. Civ. Eng. (1946–1947) 415–445. “Considerations for Developing Biopharmaceuticals: FDA
15 A. Rushton, A. S. Ward, R. G. Holdich: Solid–Liquid Perspective”, in J. Knablein (ed.): Modern Biopharmaceuticals
Filtration and Separation Technology, VCH Verlagsgesell- Design, Development and Optimization, vol. 4, Wiley-VCH
schaft, Weinheim, Germany, 1996, p. 198. Verlag, Weinheim 2005, pp. 1637–1668.
16 A. Rushton, A. S. Ward, R. G. Holdich: Solid–Liquid 37 M.W. Jornitz, T.H. Meltzer: “Filtration and Purification in the
Filtration and Separation Technology, VCH Verlagsgesell- Biopharmaceutical Industry”, in Drugs Pharm. Sci. 174 (2008)
schaft, Weinheim, Germany 1996, pp. 185 ff. 81–151.
17 K. J. Ives, in A. Rushton (ed.): Mathematical Models and 38 T. Burnouf, M. Radosevich: “Nanofiltration of Plasma-
Design Methods in Solid–Liquid Separation, NATO ASI series Derived Biopharmaceutical Products”, Haemophilia 9
E No. 88, Martinus Nijhoff, Dordrecht 1985, pp. 90 ff. (2003) 24.
18 G. Dierckx in: La Filtration Industrielle des Liquides, Tome III, 39 J. Carter, H. Lutz: “An Overview of Viral Filtration in
Societe Belge de Filtration, Liege 1978, pp. 295–314. Biopharmaceutical Manufacturing”, Eur. J. Parenteral Sci. 7
19 A. Palinski, W. Uhl, R. Gimbel, Vom Wasser 89 (1997) (2002) no. 3, 72.
175–189. 40 R.V. Reis, A. Zydney: “Bioprocess membrane technology”,
20 J. Altmann, S. Ripperger, J. Membrane Sci. 124 (1997) J. Membr. Sci. 297 (2007), pp. 16–50.
119–128.
38 Filtration, 1. Fundamentals
41 Z.H. Syedain, D.M. Bohonak, A.L. Zydney: “Protein Fouling 55 A. Rushton, A. S. Ward, R. G. Holdich: Solid–Liquid
of Virus Filtration Membranes: Effects of Membrane Orienta- Filtration and Separation Technology, VCH Verlagsgesell-
tion and Operating Conditions”, Biotechnol. Prog. 22 (2006) schaft, Weinheim, Germany 1996, pp. 455 ff.
pp. 1163–1169. 56 D. Leclerc, S. Rebouillat, in A. Rushton (ed.): Mathematical
42 W. Bender, Chem. Ing. Tech. 55 (1983) no. 11, 823–829. Models and Design Methods in Solid–Liquid Separation,
43 K.-H. Steiner, Chem. Ing. Tech. 61 (1989) no. 1, 1–8. NATO ASI series E No. 88, Martinus Nijhoff, Dordrecht
44 H. Anlauf, Fortschrittsber. VDI, Reihe 3, 114 (1986). 1985, pp. 356 ff.
45 H. Schubert: Kapillarit€ at in por€ osen Feststoffsystemen, 57 J. Gregory inK. J. Ives (ed.): The Scientific Basis of Floccu-
Springer Verlag, Heidelberg 1982, p. 184. lation, Sijdhoff & Noordhoff, Alphen aan de Rijn 1978,
46 H. Anlauf, Maschinenmarkt 95 (1989) no. 8, 26–30. p. 91.
47 R. J. Wakeman: “Cake Dewatering”, in L. Svarovsky (ed.): 58 H.-J. Jacobasch, P. Weidenhammer, Chem. Ing. Tech. 68
Solid–Liquid Separation, Butterworths, London 1977, p. 300. (1996) 1590–1594.
48 R. J. Wakeman, Filtr. Sep. 16 (1979) no. 6, 655–669. 59 D. Houi, R. Lenormand, in F. Family, D. P. Landau (eds.):
49 A. Rushton, A. S. Ward, R. G. Holdich: Solid–Liquid Kinetics of Aggregation and Gelation, Elsevier, Amsterdam
Filtration and Separation Technology, VCH Verlagsgesell- 1984.
schaft, Weinheim, Germany, 1996, pp. 348 ff. 60 P. Schmitz, B. Wandelt, D. Houi, M. Hildenbrand, J.
50 D. Redeker, K.-H. Steiner, U. Esser, Chem. Ing. Tech. 55 Membrane. Sci. 84 (1993) 171–183.
(1983) no. 11, 829–839. 61 W. H€oflinger, C. St€ockelmayer, A. Hackl, Filtr. Sep. (1994)
51 I. Nicolaou, W. Stahl, Aufbereit. Tech. 33 (1992) no. 6, 328– Dec., 807–811.
338. 62 W. H€oflinger, C. St€ockelmayer, Staub Reinhalt. Luft 55 (1995)
52 J. Walker, Spektrum Wissensch. 12 (1986) 188–197. 423–428.
53 W. Tiedemann, Fortschrittsber. VDI, Reihe 3, 453 (1996). 63 W.-M. Lu, C. C. Lai, K.-J. Hwang, Sep. Technol. 5 (1995)
54 M. Shirato, T. Murase, E. Iritani, S. Nakatsuka, Filtr. Sep. 24 45–53.
(1987) no. 2, 115–119.