98 Solid/Liquid Separation: Scale-up of Industrial Equipment
coincided with increased detection of Cryptosporidium oocysts in
treated drinking water causing a number of cryptosporidiosis outbreaks
(e.g. Milwaukee in 1993).
On-line particle counters typically count filtrate particles in a range of
pre-determined size bands from 2 \im up to around 100 ^im.
Cryptosporidium oocysts are around 5 \im in size so are easily detected
using a particle counter. Most on-line particle counters use an optical
method to detect, size and count the particles (more details can be
found in Colton et al, 1996). Other particle counter/sizers are available
which use electrical resistance created by particles suspended in an
electrolyte to estimate particle size. These are not generally suitable for
continuous on-line monitoring. A typical output from an on-line
particle counter is shown in Figure 3.11 where the counts for 2-5 |Lim
and >5 jxm particles are displayed.
25% flow increase, 1 hour into a filter run
09:12 10:12 11:12 12:12 13:12 14:12 15:12 16:12 17:12
Time
2-5um >5um Flow l/min
F i g u r e 3.11 Typical output from an on-line particle counter for a laboratory column experiment
investigating flow changes
3.4 Full scale filters
Tests can be made on full scale filters by varying the pre-treatment, or
the rate of flow if the associated hydraulic equipment is suitable. It is
more difficult to make changes in media due to the quantities involved.
Consequently, tests on full-scale filters do not generally have the
flexibility of model tests.
3 • Deep bed filters 99
As in the case of models, observations of filtrate quality changes and
total head loss during the filter run can reveal something of the filter
performance. The interpretation of the data is identical to that for
models. However, the observation of pressure profiles through the
media is much more informative giving the same insight into the
behaviour of the various media layers as with the models.
Full-scale filters are rarely equipped with manometer probes through
the media depth when they are constructed. One example of this was at
the Chapel Hill (North Carolina) waterworks, where tubes were
incorporated in the concrete wall of a filter during construction. Probes
were later inserted through the wall into the media, to which a bank of
plastic manometers was connected. Similar probes were inserted
through holes burned with a thermic lance through the concrete filter
wall at Staines waterworks (North Surrey Water Co.). These probes
were used to monitor the performance of an anthracite-sand media
filter; an example of the pressure profiles obtained is in Figure 3.12.
This also shows the pressure curves obtained from a model column of
15 cm diameter operating under identical conditions at the same
waterworks. The close similarity of the curves proves the point that
there is no scale effect in deep bed filtration providing that the media,
operating regime and washing processes are identical. Similar data
have been obtained from the Vereeniging (Rand Water Board)
waterworks and Kempton Park Experimental Plant of Thames Water
Utilities Ltd.
Pressure ins H2O
80 90 100 110
Hours
-Model Full scale
F i g u r e 3 . 1 2 Pressure profiles through anthracite-sand-gravel media at Staines waterworks during
94 hr run. Note the close agreement between the model (15 cm diameter) and full scale readings
100 Solid/Liquid Separation: Scale-up of Industrial Equipment
Currently, more and more filters now have transducers installed to
record the pressure loss across individual filter beds.
Backwashing tests are more difficult to evaluate on ftill-scale filters
due to the difficulty in observing visually the media expansion. A
useful device for measuring the expansion consists of a rod with small
trays (for example, 5 cm square by 1.5 cm deep) attached at 3 cm
intervals. The rod is set so that the tray at its lower end rests on the
undisturbed media. During backwashing, the rod is held secure in this
position and media retained in the trays after backwashing will indicate
the height to which expansion took place.
3-5 Filter operation
3.5.1 Flow control
3.5.1.1 Head loss and permeability
In order to maintain a steady output from a water treatment works it is
usual to keep a control on the outflow of each filter. As each filter
retains particles (typically fine hydroxide fiocs) in the pores of its
sand, or other granular media, the resistance to flow in the filter bed
increases, as it loses permeability. Without some form of flow control
this would lead to progressive diminution of the flow volume dis-
charged from the filter. Consequently there is a hydraulic requirement
to compensate for the loss of permeability due to the clogging of the
filter media.
An exception to the requirement for constant output from each filter is
the process of declining rate filtration, where filters in a group are
operated at various stages of decline in output, starting from above
average outflow rate. A strict sequence of backwashing maintains the
output of the group to meet the requirements of the waterworks supply.
A minimum of 4 filters is required, which makes model simulations more
complex. A comparison of flow control systems, including declining rate
filtration can be found elsewhere (Committee Report, 1984).
The clean filter media, after backwashing, offers relatively little
resistance to flow - the equivalent of a small number of centimetres of
head loss, T/Q? which can be calculated from the well-known Kozeny-
Carman equation (laminar flow in porous media):
I g P s^ ¥ d
3 • Deep bed filters 101
where ^o is the clean filter porosity, JA and p are the water viscosity and
density, d and ip are the grain diameter and sphericity (see Section
3.6.2.2), and k^ is the Kozeny constant. There have been attempts to
produce a form of the Kozeny-Carman equation for a filter bed in the
process of clogging. The resulting equations are difficult and unreliable
as the porosity is reducing with time of operation, by different amounts
at different depths in the filter media. In addition, the filter grains are
changing shape and size due to coatings of deposits which are also
varying with time and depth. A third uncertainty lies in the Kozeny
coefficient (A:^), usually taken as 5.0, which may vary with time and
depth in an unpredictable manner.
These problems can all be solved by utilising filter models or full-scale
filters equipped with manometers or pressure transducers, set at
intervals throughout the media depth as described previously. The
readings of head loss with time and depth are then directly applicable
and there is no need to deal with ill-defined parameters.
The progressive loss of permeability of the filter media leads to a decline
in filter flow and so the increasing resistance of the clogging filter has to
be met by a decreasing resistance in a rate controller, or an increasing
inlet head to compensate. A rate controller provides an additional loss of
head in the influent line (upstream control) or effluent line (downstream
control). Adjusting this loss of head, usually reducing it, keeps the
supply of water or discharge of filtered water at a desired constant value.
3.5.1.2 Flow controllers
The most common types of filter rate control on full-scale filters are
shown diagrammatically in Figure 3.13.
As can be seen in Figure 3.13 all the systems require one or two
variable control (typically butterfly) valves. It is the mechanical
resistance of these valves which makes miniaturisation impractical for
model-scale filters, as mentioned previously. There are several more
complex alternatives to the use of butterfly valves for variable flow
control: these include pump regulators, venturi devices and siphonic
systems. It is unlikely that any of these would translate to the model
scale; also they are more expensive. It is possible to use a laboratory
peristaltic pump to maintain outlet flow on a model filter, but it needs
to be appreciated that the plastic tube changes its flexibility with time.
Also, if there are particles in the filtrate they may deposit over time in
the flexible tube altering its characteristics.
In practical installations in water treatment works there are several
rapid filters acting in parallel, with one out-of-service for backwashing.
102 Solid/Liquid Separation: Scale-up of Industrial Equipment
ZZZZZ2ZZZ
max. w.L
min. w.i.
<^ . calibrated
max. W.L
min. W.L
'^f^HNU
max. wJ.
conduit V^//u///^^
max. w.l.
min. W.L
Systems: A and B have upstream water level control, downstream rate control;
C and D have upstream rate control, downstream water level control;
A and C have constant inlet water level, variable filtered water level;
B and D have variable inlet water level, constant filtered water level.
Figure 3,13 Systemsof filter control
while the others sustain the waterworks output. This involves a small
increase in flow rate in the remaining filters. Although the changes are
small, it is important in practice to make such transitions slowly to
avoid disturbance and detachment of the deposits in the filter pores.
This is not usually a significant factor in scale-up operations, although
it is possible to simulate flow rate increases in a model to determine
3 « Deep bed filters 103
appropriate rates of change to avoid detachment of deposits. A
significant advantage in the operation of an array of rapid filters is to
maintain a common inlet water inlet level in all. This is achieved by
having a common inlet channel with large openings to each filter,
giving a negligible resistance to flow.
Whilst most textbooks on water treatment do not contain enough detail
for these practical matters of flow control to be presented, a good
introduction to the subject can be found in either of the following
references: AWWA (1990) or Stevenson (1997).
3.5.2 Filter washing
Granular media filters need regular backwashing to remove clogging
deposits and maintain efficient operation. Backwashing of a gravity
filter involves flow reversal to dislodge the deposits. There are various
techniques of backwashing:
• fluidising water wash
• fluidising water wash plus surface jets
• air scour followed by a fluidising water wash
• simultaneous air and water wash followed by a fluidising water rinse.
All of the above backwash procedures are conducted at varying upflow
rates and bed expansions in practice. Backwashing can be applied to
laboratory filters, which can be used to investigate backwash options.
Backwash should be optimized to give best cleaning but to minimise
water use.
If a filter is backwashed by upflowing water then the bed of granular
material will become fiuidised in accordance with the theory and
observations to be discussed below.
3.5.2.1 Fiuidised beds and fluidisation theory
A fiuidised bed consists of solid particles or grains suspended by a
fluidising medium, e.g. a gas or a liquid.
Consider a vessel such as a filter shell filled with a granular material, e.g.
sand, of bed depth, / (see sketch below). A fluidising medium (usually
water) enters at the base of the vessel and flows upwards through the bed
of sand with a superficial velocity, v. The granular material has a fixed
bed voidage of e. When the bed is fiuidised the particles or grains are
suspended in equilibrium by the fluid drag forces exerted on them. The
bed expands to a depth ^ and a consequent voidage s^.
