An assessment of the Silt Density Index based on RO membrane ...

230 S. G. Yiantsios, A.J. Karabelas /Desalination 151 (2002) 229-238
<strong>of</strong> <strong>the</strong> deposits detected in membrane autopsies
throughout <strong>the</strong> world [l]. The issues, <strong>the</strong>refore,
<strong>of</strong> prediction and mitigation <strong>of</strong> colloidal and
organic fouling remain critical after many years
<strong>of</strong> successful application <strong>of</strong> membrane technologies
to a variety <strong>of</strong> water sources including
well and surface, brackish and seawater, as well
as wastewater <strong>of</strong> various origins.
The most common and widely accepted tool
for prediction <strong>of</strong> colloidal fouling is <strong>the</strong> <strong>Silt</strong>
<strong>Density</strong> <strong>Index</strong> (SDI). O<strong>the</strong>r proposed indices are
<strong>the</strong> Modified Fouling <strong>Index</strong> (MFI), which was
developed in <strong>the</strong> Ne<strong>the</strong>rlands [2] and gained
acceptance mainly <strong>the</strong>re, and <strong>the</strong> Mini Plugging
Factor <strong>Index</strong> (MPFI ). All three are <strong>based</strong> on
batch filtration <strong>of</strong> feed waters through a 0.45 pm
Millipore micr<strong>of</strong>ilter. More recently, <strong>the</strong> UF-
MFI was introduced which uses ultrafiltration
membranes instead <strong>of</strong> <strong>the</strong> Millipore micr<strong>of</strong>ilter
[3,4]. O<strong>the</strong>r measurements such as turbidity,
particle counts, and particle electrophoretic
mobility are also used but are not considered
reliable criteria for <strong>assessment</strong> <strong>of</strong> fouling trends.
Membrane manufacturers recommend that SD1
should not exceed 4 or 5 and set limits <strong>of</strong> membrane
productivity depending on <strong>the</strong> SD1 [5] (i.e.,
8-14 gfd for SD1 2-4, 14-18 gfd for SD1 ~2 and
20-30 gfd for SD1 cl). It should be emphasized
that <strong>the</strong> SD1 test is, strictly speaking, an empirical
one and a very poor simulation <strong>of</strong> actual RO
conditions. The absence <strong>of</strong> actual membranefoulant
interactions, <strong>the</strong> absence <strong>of</strong> shear (as
opposed to <strong>the</strong> cross-flow operation <strong>of</strong> membrane
modules), <strong>the</strong> significantly higher permeation
rate (>l cm/s in <strong>the</strong> filter vs. c.001 cm/s in RO
membranes), and <strong>the</strong> doubtful rejection <strong>of</strong>
particles smaller than 0.45 pm (nominal pore size
<strong>of</strong> <strong>the</strong> filter) are <strong>the</strong> most important concerns.
These poorly rejected particles are more likely to
cause severe fouling to a membrane since <strong>the</strong>
resistance <strong>of</strong> a cake layer is inversely proportional
to <strong>the</strong> square <strong>of</strong> particle size. Schippers et
al. [6] state that particles smaller than 0.05 pm
are responsible for flux decline in RO mem-
branes. <strong>An</strong>o<strong>the</strong>r criticism is that <strong>the</strong> index is not
linearly related to colloid particle concentration
and does not distinguish between fouling
mechanisms during <strong>the</strong> test (pore blocking, cake
filtration, cake consolidation).
Despite <strong>the</strong> significance <strong>of</strong> colloidal fouling
problems in desalination, <strong>the</strong>re are relatively few
experimental studies at laboratory scale that
represent typical water sources, realistic foulants
and well controlled conditions at <strong>the</strong> same time.
Most studies use syn<strong>the</strong>tic waters <strong>of</strong> simple
composition (e.g., distilled water to which NaCl
is added) and model colloidal particles, which
cannot be considered typical <strong>of</strong> <strong>the</strong> foulants
encountered in practice [7-91. A second related
problem is <strong>the</strong> duration <strong>of</strong> <strong>the</strong> laboratory
experimental studies. First <strong>of</strong> all, long times are
required in order to obtain a stable membrane
flux under non-fouling conditions due to phenomena
referred to as “membrane setting”. In
addition, due to <strong>the</strong>ir high resistance to water
flow, RO membranes are not very sensitive to
colloidal fouling, in <strong>the</strong> sense that relatively thick
deposits <strong>of</strong> <strong>the</strong> typically employed particles
(which usually exceed 100 nm in diameter) need
to be developed in order to observe a detectable
flux reduction. This necessitates long experimental
times and accelerated fouling conditions
with unrealistically high particle concentrations
(200 ppm or more). It may also be added that in
a typically closed-loop laboratory system it is
difficult to maintain constant conditions due to
<strong>the</strong> limited stock <strong>of</strong> fouling particles and to
possible aggregation phenomena.
The present experimental efforts include two
steps: first, to generate reliable RO fouling data
under realistic, yet, well controlled conditions;
second, on <strong>the</strong> basis <strong>of</strong> such results, to assess <strong>the</strong>
performance <strong>of</strong> existing techniques for colloidal
fouling prediction (i.e., SDI, MFI). Iron oxide
was selected as a typical inorganic colloidal
foulant due to its importance as evidenced by <strong>the</strong>
frequent appearance in membrane autopsies [ 11,
and <strong>the</strong> specific reference in manufacturers’