Crystalliser Design
Crystalliser Design
Crystalliser Design
Abstract
The main challenge in the design of industrial crystallisers is to predict the influence of crystalliser geometry, scale,
operating conditions and process actuators on the process behaviour and product quality. The quality characteristics,
such as the crystal size distribution, inclusion content and morphology determine to a large extent the product
performance and are therefore of importance. The quality of the product crystals is basically determined by the rates at
which crystals are born and attrited, grow or dissolve and agglomerate in the different regions of the crystalliser. An
analysis technique is therefore introduced to describe the various crystallisation phenomena as a function of local process
conditions such as supersaturation and energy dissipation. This technique is based upon:
The advantage of this technique over conventional techniques is illustrated for an evaporative DTB crystalliser. 1999
Elsevier Science B.V. All rights reserved.
0022-0248/99/$ — see front matter 1999 Elsevier Science B.V. All rights reserved.
PII: S 0 0 2 2 - 0 2 4 8 ( 9 8 ) 0 1 1 7 9 - 8
730 H.J.M. Kramer et al. / Journal of Crystal Growth 198/199 (1999) 729–737
solvent—solute system, crystalliser geometry and tions on crystallisation behaviour. Although it has
set of operating conditions, it is thus necessary to already long been recognised, that industrial crys-
take into account the non-uniform distribution tallisers are far from ideally mixed, geometrically
throughout the vessel of process variables, such as lumped descriptions of the crystallisation phe-
the solids density, crystal size distribution (CSD), nomena in a vessel and their effect on product
supersaturation and energy dissipation, and also quality still prevail.
the system dynamics, an inherent quality of most The need for a description of the crystallisation
crystallisation processes. phenomena on a local scale in an industrial crystal-
Traditional scale-up strategies for crystallisers do liser, i.e. on a scale where process conditions are
not take these geometric and time-dependent vari- uniform, is widely recognised:
ations into account, and it is therefore hardly sur-
E A geometrically lumped description only yields
prising that these strategies practically fail without
one average value for the supersaturation, thus
exception. In this paper, we will discuss an ap-
ignoring possible peak values. In evaporative
proach to the full scale design of industrial crystal-
crystallisers the supersaturation will be above
lisers for a predefined set of product specifications.
average in the boiling zone and lower, possibly
Main emphasis will be upon:
undersaturated, after the heat exchanger. Near
1. The need for a more detailed and separate de- feed points of a cooling crystalliser, where
scription of the hydrodynamics and kinetics to a saturated stream is mixed with the much cooler
analyse and optimise the behaviour of design contents of the vessel, high supersaturation
alternatives. values can be expected, possibly leading to pri-
2. Setting up hydrodynamic models for industrial mary nucleation or scaling. In both these exam-
scale crystallisers (typically non-MSMPR), in ples, a lumped description of the crystallisation
which the crystalliser volume is divided into process will probably ignore or definitely under-
hydrodynamically well-defined regions, i.e. re- estimate the occurrence of certain phenomena,
gions in which the supersaturation, energy dissi- such as dissolution and primary nucleation.
pation, solids density and CSD are uniformly E Whenever there is a difference in material density
distributed. of the solid and liquid phase, particle segregation
3. Obtaining scale and geometry independent kin- will occur to some extent. To what extent, de-
etic parameters, for a certain kinetic model, from pends on the internal circulation rate, which is
laboratory scale (MSMPR) experiments. The related to the specific power input of a circula-
key problem here is how to obtain pure kinetic tion pump or impeller. If particle segregation is
parameters, that are not polluted with hy- such that large particles will have a significantly
drodynamic information. shorter residence time in the boiling zone,
4. Description of the crystallisation phenomena the growth rate of these particles will appear to
in full scale crystallisers as a function of local be below average. In a geometrically lumped
process conditions, in order to properly predict description of such a process, this effect can only
the influence of changes in crystalliser geo- be described by lowering the growth rate con-
metry, scale and operating conditions on the stant of the larger particles. The error in this
resulting product quality and its related product approach becomes evident when the internal cir-
performance. culation rate is increased, particle segregation
decreases and the growth rate of these particles
approaches average values.
2. Drawback of current design practice These examples also obviate the problems en-
countered in practice, when a kinetic model, of
The main problem with the current design prac- which the parameters were estimated from the be-
tice, is the lacking of an adequate tool to predict the haviour of a non-ideally mixed crystalliser, is
influence of geometry, scale and operating condi- applied to another crystalliser. These kinetic
H.J.M. Kramer et al. / Journal of Crystal Growth 198/199 (1999) 729–737 731
dimensions, geometry and operating conditions. model, is illustrated for ammonium sulphate in
In this phase, the majority of the rates of flows water using a 22 l evaporative DT crystalliser. For
connecting the various compartments is deter- this system, first of all the kinetic model derived by
mined roughly. Ö Meadhra et al. [7] was selected. This model is
E Secondly, the compartments are checked for in- discussed below, in order to highlight the depend-
ternal gradients in local energy dissipation. encies of the various kinetic phenomena on super-
Strong gradients may be expected around the saturation and crystal size. Note that no intrinsic
impeller or inside the pump and near feed loca- dependency of changes in energy dissipation are
tions. If they are indeed present, one or more of present in the model.
the compartments selected in the first step will be In a recently published new model for secondary
split up. nucleation [5], the direct relation is formulated
E In the third place, all compartments are checked between the crystallisation kinetics and the fre-
for the presence of internal supersaturation quency and energy of particle—impeller collisions
gradients, or in other words the degree of liquid- depending on a number of material properties of
phase mixing is analysed. For this purpose, the the crystals. This promising new model, which is
half lifetime for supersaturation decay must discussed in the contribution of Neumann et al. [8],
be one order of magnitude larger than the resi- will open the possibilities for the formulation of the
dence time in the compartment. Depending on direct relation between the energy dissipation and
the crystallisation system, the half lifetime varies crystallisation kinetics.
from a few seconds to several minutes. The growth kinetics are dependent on both the
E Finally, classification functions are calculated to supersaturation and the crystal size. The size de-
describe particle segregation or the non-uniform pendency is a result of internal stress in the smaller
distribution of the solids phase. These functions particles. This stress is a consequence of the attri-
act upon the flows connecting the individual tion process, by which these particles were formed
compartments, which effectively makes the resi- (secondary nucleation).
dence time of the crystals in a compartment size
¸ N
dependent. The classification function of a slurry G (¸,p)"p pN 1!(1!p ) exp . (1)
p
stream is related to its superficial flow velocity,
flow direction, viscosity and density difference The attrition rate function accounts for the fact that
between the liquid and solid phases. the smaller particles do not develop sufficient kin-
etic energy to undergo attrition. From a certain
It must be noted that the same material, energy
size onwards the crystals become more and more
and population balances are used to describe each
susceptible to attrition:
compartment in a compartmental model [1]. This
also applies to the physical and thermodynamic 1
G (¸)"p 1! . (2)
property relations as well as the kinetic rate expres- 1#(¸/p )N
sions. Different nucleation, growth, dissolution, at-
trition, breakage and aggregation rates are purely The effective growth rate of a crystal thus depends
a result of different local conditions, not of different on its size and the supersaturation:
sets of parameters. G (¸,p)"G (¸,p)!G (¸). (3)
The volumetric rate of attrition is given by
5. Crystallisation kinetics derived from laboratory
scale experiments »Q "3k G (¸)n(¸)¸ d¸. (4)
The analysis of a kinetic model structure and of The secondary nucleation rate is related to the
a particular laboratory scale crystalliser’s suitabil- attrition rate by means of a supersaturation-depen-
ity to derive pure kinetic parameters for that kinetic dent survival efficiency ("p p). Furthermore, an
734 H.J.M. Kramer et al. / Journal of Crystal Growth 198/199 (1999) 729–737
Fig. 5. Six compartment model of a 1.1 m DTB crystalliser. of the draft tube, a compartment to describe the
classification in the annular zone and temperature
increase in the external heat exchanger, and a
the external circulation loop in which fine compartment in which fine particles are dissolved
crystals are dissolved and the internal circulation after the heat exchanger (see Fig. 5). For compari-
through the draft tube are essential for larger son purposes, the crystalliser was also modelled as
crystallisers. a classical one-compartment model with an ideal
Changes in the fines removal rate influence the fines dissolver [7]. In both the six- and one-com-
cut size of the fines classification system, as a direct partment model, the particle segregation in the
consequence of an increased vertical velocity in annular zone was described using a model de-
the annular zone. This flow rate also determines veloped by Eek et al. [4], which was based on CSD
the residence time in the heat exchanger and the measurements of the fines and product streams.
temperature rise of the fines flow. The degree of The median crystal size trend after start-up
dissolution is determined by a combination of these simulated with the one compartment model and
effects. ideal fines dissolver is shown in Fig. 6 for fines
The behaviour of a DTB crystalliser is thus flows of 2.5 and 3 l s\. Comparison with the
strongly influenced by its geometry, actuator design MSMPR trend given in the same figure, denotes
and operating conditions. The analysis of DTB that ideal fines dissolution strongly affects the me-
crystalliser design alternatives therefore requires dian crystal size, while the process dynamics are
a detailed description of crystallisation phenomena hardly affected. In reality, however, often a much
on a local scale. Such a detailed analysis will be smaller effect is observed on the steady-state value
illustrated for a 1.1 m evaporative DTB crystal- of the median crystal size, whereas the dynamics are
liser. For this purpose, a six-compartment model strongly affected, occasionally resulting in severe
was derived, consisting of a boiling zone, a stirrer cyclic behaviour of the crystalliser [4,8,9]. This
zone, two zones for the inner and outer sections discrepancy between model and reality, makes the
736 H.J.M. Kramer et al. / Journal of Crystal Growth 198/199 (1999) 729–737
7. Conclusions
Acknowledgements
[2] S.K. Bermingham, H.J.M. Kramer, G.M. van Rosmalen, [7] R. Ö Meadhra, H.J.M. Kramer, G.M. van Rosmalen,
Proc. Escape 8, 24—27 May, Brugge, Belgium, 1998. AIChE J. 42 (1996) 973.
[3] P.J. Daudey, Ph.D. Thesis, Delft University of Technology, [8] A.M. Neumann, S.K. Bermingham, H.J.M. Kramer, G.M.
1987. van Rosmalen, J. Crystal Growth 198/199 (1999) 723.
[4] R.A. Eek, Sj. Dijkstra, G.M. van Rosmalen, AIChE J. 41 (3) [9] A.M. Neumann, S.K. Bermingham, H.J.M. Kramer,
(1995) 571. G.M. van Rosmalen, Modelling the dynamic behaviour of
[5] C. Gahn, A. Mersmann, Trans I Chem E 75 (A) (1997) 125. a 22 liter evaporative DT crystallizer, in: Proc. Int. Symp. on
[6] J. Jager, Ph.D. Thesis, Delft University of Technology, 1990. Industrial Crystallisation, Tianjin, China, 1998, pp. 222—226.