20th European Symposium on Computer Aided Process Engineering ESCAPE20 S. Pierucci and G.

. Buzzi Ferraris (Editors) 2010 Elsevier B.V. All rights reserved.

Numerical Solution of Batch Crystallization Models

Qamar S.a,b, Seidel-Morgenstern A.a
a

Max-Planck Institute for Dynamics of Complex Technical System, Sandtor Str. 1, 39106 Magdeburg, Germany, [email protected] b COMSATS Institute of Information Technology , Park Road Chak Shahzad , Islamabad, Pakistan, qamar@ mpi-magdeburg.mpg.de

Abstract
An efficient and accurate numerical technique is introduced for the simulation of a batch crystallizer equipped with a fines dissolution unit incorporating a time-delay. The dissolution of small crystals (fines dissolution) improves the product quality and facilitates the downstream processes. The proposed method follows two steps. In the first step, a coupled system of ordinary differential equations (ODEs) for moments and solute mass is numerically solved in the time domain of interest, giving the discrete values of growth and nucleation rates. In the second step, theses discrete values are used along with the initial crystal size distribution (CSD) to construct the final CSD. The method of characteristics and Duhamels principle are employed for deriving an expression for CSD from the given population balance model (PBM). An alternative quadrature method of moments (QMOM) is introduced for approximating integrals in the ODE system of moments and mass balance. In this technique, orthogonal polynomials, obtained from the lower order moments, are used to find the quadrature abscissas and weights. The numerical results of our scheme are validated against the results of high resolution finite volume scheme results. Our scheme was found to be efficient, accurate, and free from numerical dissipation and dispersion. Keywords: Batch crystallization, nucleation and growth, fines dissolution, time-delay, method of characteristics, Duhamels principle.

1. Introduction
Crystallization from solution is an important industrial operation due to marketing of large number of materials as crystalline particles. It is therefore one of the most widely separating and purifying technique in chemical, pharmaceutical, semiconductor, and food industries. Population balance equations (PBEs) are widely used for modelling crystallization processes. Several numerical methods have been developed for solving PBEs, see for example Ramkrishna (2000) and Qamar et al. (2009). In Qamar et al. (2009), a batch crystallization models without fines dissolution was considered. In this article, the numerical method in Qamar et al. (2009) is extended for solving a batch crystallization model with fines dissolution. The main ingredients of the scheme are the initial data for CSD and solute mass, a coupled system of ODEs for moments and solute mass, as well as an expression for CSD obtained by utilizing the method of characteristics and Duhamels principle. To overcome the closure problem, a Gaussian quadrature method based on orthogonal is employed for approximating integrals appearing in the moment-system of ODEs polynomials (e.g. in Qamar et al. (2009). In this manuscript, a third order orthogonal polynomial, needing first six moments, is used. A comprehensive discussion on other QMOM methods and their limitations can be found in Grosch et al. (2007). The current QMOM was found robust and the problems

Qamar and Seidel-Morgenstern mentioned by Grosch et al. (2007) were not observed. The proposed method is efficient, accurate, and easy to implement compared to other numerical methods available in the literature. Further details about this method can be found in Qamar et al. (2009). For validation, the numerical results of our scheme are compared with the finite volume scheme of Koren (1993).

2. Mathematical Model
In one-dimension, the crystals size is represented by a characteristic length x . The CSD, i.e. n(t , x) , represents the number of crystals per crystal length. A balance for the number of crystals in an infinitesimal interval of crystal length leads to the following PBE (e.g. Randolph and Larson (1988)) & n(t , x) [G(t , x)n(t , x)] V = h( x)n(t , x) + B0 (t ) ( x x0 ), n(0, x) = n0 ( x), (1) t x Vc where, (t , x) 2 + and + := (0, ). Here, n0 ( x) denotes the CSD of seed crystals, G (t , x) is the size-dependent crystal growth rate, B0 (t ) is the nucleation rate at minimum crystal size x0 and represents Dirac delta distribution. Moreover, Vc is the & is volumetric flow rate from the crystallizer to the dissolution volume of crystallizer, V unit, and the death function h( x) describes the dissolution yield of small particles below some critical size. The ith moment of CSD is defined as

i (t ) = x i n(t , x)dx , i = 0,1,2,L.

0

(2)

A balance for the liquid phase yields an ODE for the solute mass

dm(t ) & in (t ) m & out (t ) 3c kv x 2G (t , x)n(t , x)dx , m(0) = m0 , =m dt 0

(3)

where, c is crystals density, and k v is a volume shape factor defined such that the
& out is the mass flux in the liquid phase volume of a crystal is k v x 3 . In equation (3), m & in is the which is being taken out from the crystallizer to the dissolution pipe, and m incoming mass flux in the crystallizer from the dissolution unit. They are defined as &, & out (t ) = w(t ) soluV m & (4) k V 3 & in (t ) = m & out (t t p ) + v c x3h( x)n(t t p , x)dx kv c x0 m B0 (t t p ), Vc 0

where, w(t ) = m(t ) /(m(t ) + msolv ) represents the mass fraction. Here, solu is the density of the solution and msolv is the mass of the solvent. Moreover, t p 0 represents & , where V p denotes the residence time in the dissolution unit. It is defined as t p = V p / V
volume of pipe. A size dependent growth rate can be defined as G (t , x) = k g ( S (t ) 1) g (1 + 1 x) 2 := g1 (t ) g 2 ( x), (5)

where k g is the growth rate constant. The exponent g denotes the growth order, and 1 and 2 are constants. Moreover, S (t ) = w(t ) / weq is the supersaturation of the dissolved component which is decreasing with time. Here, weq is the equilibrium mass fraction. The nucleation rate is defined as

Numerical solution of batch crystallization models

B0 (t ) = k b ( S (t ) 1) b 3 (t ), (6) where, k b is nucleation rate constant and the exponent b gives the nucleation order. Both growth and nucleation are of empirical nature. The above model reduces to the case of fines dissolution without time-delay when t p = 0 and to without fines dissolution case when the second last term on the right-hand side of (1) and the first two terms on the right-hand side of (3) are zero. Then equation (4) is not required.

3. Numerical Technique
Here, we give a brief overview of our numerical scheme, named as constructed scheme. A complete derivation of the scheme can be found in the recent article by Qamar et al. (2009). The method has following two steps:
Step 1: In this step of the method, we solve the moment-system of ODEs, obtained by using the PBE (1), equation (2), and the mass balance equation (3) & di (t ) V i = i xi 1G (t , x)n(t , x)dx xi h( x)n(t , x)dx +x0 B0 (t ), i = 0,1,2, L ,5, dt V c 0 0 (7) dm(t ) & in (t ) m & out (t ) 3c kv x 2G (t , x)n(t , x)dx. =m dt 0

In the case of time-delay, the conditions in the pipe are taken equal to the initial conditions in the crystallizer for t < t p in equation (4). The above equation can not be solved by a standard quadratur method due to the closure problem, therefore the quadrature method of moments (QMOM) is needed. By using QMOM, we obtain N & N d i (t ) V i i 1 i = i wk xk G (t , xk ) wk xk h( x k ) + x0 B0 (t ), dt V p k =1 k =1 (8) N dm(t ) 2 & & = min (t ) mout (t ) 3 c k v wk xk G (t , xk ), i = 0,1,2,L,5. dt k =1

where, xk are the abscissas and wk are the quadrature weights. A third order orthogonal polynomial is given as (see Qamar et al. (2009)): 2 2 12 5 2 3 ) x 2 ( + 0 2 5 + 13 p3 ( x) = x3 + 1 2 4 3 0 3 4 2 2 2 0 2 4 212 3 0 3 1 4 2 2 2 4 + 2 32 ) x (12 5 0 35 134 + 0 4 + (9) 3 2 2 0 2 4 212 3 0 3 12 4 2 2 3 (22 34 + 135 14 2 5 2 3 ) x + . 3 2 2 0 2 4 212 3 0 3 12 4
The roots of above polynomial are the abscissas xk of Gaussian quadrature in (8). After having the abscissas, the next step is to calculate weights wk . According to Press et al. (2007) the weights are given as p2 p2 2 (10) wk = , p 2 p 2 = n(t , x) p 2 dx, k = 1,2,3 , p 2 ( x k ) p ( x k ) 0 where, the second order orthogonal polynomial p 2 ( x) is defined as
p 2 ( x) =
2 x 2 ( 0 2 12 ) + x( 1 2 0 3 ) + 1 3 2 . 2 0 2 1

(11)

Qamar and Seidel-Morgenstern

Step 2: After solving the ODE system (8), the discrete values of six moments and the corresponding growth and nucleation rates are available in the time domain of interest. The next step is to get an expression for the CSD. This can be obtained by solving the PBE (1) with the method of characteristics and Duhamels principle (see e.g. Qamar et al. (2009)). Let (t , x) be a location where the characteristic through (t , x) intersects the x-axis in the backward direction. Then, equation (1) gives (e.g. Qamar et al. (2009)) g ( x ) B ( ) & t V 2 0 0 exp h( (t s, x))ds Vc , (t , x) < 0, (12) n(t , x) = nH (t , x) + G( , x) 0, otherwise where, & t V G(t , (t , x))n0 ( (t , x)) (13) nH (t , x) = exp h( (t s, x))ds . G(t , x) Vc 0

The analytic expression for (t , x) is given as (e.g. Qamar et al. (2009))

t 1 g1 ( ) d 1 , = 1, 0, 0 2 1 1 1 + ( x)e 1 t 3 1 3 (t , x) = 1 + [ ( x)] 1 3 g1 ( )d , 2 1, 1 0, 0 1 t x g1 ( )d , 1 = 0, 0

(14)

where ( x) = (1 + 1 x) and 3 = 1 2 . Note that (0, x) = x and (t , x) < x for t > 0 because the characteristics moves from left to right in the x t plane. The last step is to find needed for any (t , x) < 0 by finding the root of
(t , x) . Thus, for a given t and (t , x) < 0 , one can find by using Newtons formula Q( ) = (t , x) x0 , Q( ) =

(15)

Q k ( ) , Q k ( ) where, k represents the iteration steps. Note that, G (t , x) and B0 (t ) are only available at discrete points in the time domain. Therefore, a linear interpolation is used for calculating their values at any [0, t ].

k +1 = k

4. Numerical Test Problem

In order to validate our numerical schemes for the given model, we consider the following numerical test problem. The initial data are given as
mseeds 2 n(0, x) = e 2 , m(0) = 0.009915. (16) kv c 3 (0) 2 Here, c = 1250 kg/m3, kv = 0.029 , mseeds = 0.0025 kg, x = 0.0014 m, = 0.00032 m. The minimum and maximum crystal sizes considered are x0 = 0 and xmax = 0.005 m,
( x x )2

Numerical solution of batch crystallization models

respectively. The interval [ x0 , xmax ] is subdivided into 300 grid points and the final simulation time is 800 minutes. In the case of size-dependent growth rate we have chosen 1 = 200 m-1 in equation (5), while 2 = 1 in all cases. For size-independent growth rate, 1 = 0 . Moreover, k g = 1.37 105 m/min, g = 0.73 , kb = 3.42 107 (m3min)1 & = 2 105 , b = 2.35 , weq = 0.091 , msolv = 0.802 kg, solu = 103 kg/m3, Vc = 103 m3, V m3, V p = 2.4 104 m3, and T = 33 oC. The death function h( x) is considered as
1 1 h( x ) = e 1 , 1 = , = 1.1547 103. (17) 21 0.6 2 In Figure 1, the final CSDs for size-independent and size-dependent growth rates are given. The CSDs of our scheme are also compared with those obtained from the high resolution finite volume scheme (FVS) of Koren (1993). In the figure, symbols denote the results of our scheme (constructed) and lines are used for FVS results. The fines dissolution unit dissolves small crystals below some critical size and therefore reduces the number of small crystals in the crystallizer as depicted in the figure. The dissolution of small crystals enhances the solute mass in the solution and therefore could increase the growth rate of the seed crystals. However, in the case of fines dissolution without time-delay, the dissolution of a large number of small nuclei produces negligible effect on the growth rate of seeds crystals. In this case, small nuclei get no time to grow and instantaneously dissolve back in the solution as soon as they achieve a stable size. On the other hand, the fines dissolution with time-delay allows the small nuclei to grow in the crystallizer and the more concentrated solution from the dissolution unit comes back to the crystallizer after certain time-delay. Hence, the concentration of the solution increases in the crystallizer. In other words, the seeds crystals and those introduced by nucleation grow at a faster rate. The numerical results of finite volume scheme and our scheme are almost overlapping. However, our scheme has better resolved the sharp discontinuities. Moreover, Figure 1 also shows plots of the normalized moments. One can see that, the total number ( 0 ), size ( 1 ) and surface area (2) of the crystals x
2

reduces, but the total volume ( 3 ) increases with fines dissolution. Table 1, gives the errors in mass balances and CPU time of our scheme. It is clear from the table that our scheme preserves mass balance. Table 1. Errors in mass balances for the size-independent growth rate Absolute error Relative error CPU time (sec) Description Without fines dissolution Fines dissolution without delay Fines dissolution with delay
1.39 107 1.39 107 5.8 104 1.37 106 1.37 106 4.16 103

11.6 11.8 15.8

5. Conclusions and Remarks

In this manuscript an efficient and accurate numerical technique is derived for solving one-dimensional batch crystallization model with fines dissolution and time-delay. The main ingredients of the scheme are the initial CSD and solute mass, a coupled ODE system of moments and solute mass, as well as an expression for the CSD obtained by employing the method of characteristics and Duhamels principle. A Gaussian quadrature method based on orthogonal polynomials was used to close the momentsystem. The method was found to be efficient, accurate, and free from numerical

Qamar and Seidel-Morgenstern dissipation and dispersion. Moreover, no grid refinement is needed for improving the quality of the solution. This property of the numerical method is fruitful in the multidimensional case or in situations where more computations are involved. The significant effects of fine dissolution and delay on CSD are illustrated in a case study.
3.5 3 2.5
7 x 10 sizeindep. growth, lines:constructed, symbols:FVS

x 10

sizedep. growth, lines: constructerd, symbols: FVS without fines dissolution fins dissolution without delay fines dissolution with delay

without fines dissolution fines dissolution without delay fines dissolution with delay

2.5

2
2 1.5 1 0.5 0 0

CSD [1/m]
1 2 length [m] 3 4 x 10 5
3

CSD [1/m]

1.5

0.5

0 0

2 length [m]

4 x 10

5
3

9 8

without fines diss. and sizeindep. growth

0(t)/0(0) 1(t)/1(0) 2(t)/2(0) 3(t)/3(0)

fines diss. with delay and sizeindep. growth

11 10 9 0(t)/0(0) 1(t)/1(0) 2(t)/2(0) 3(t)/3(0)

normalized moments

normalized moments
300 400 500 600 700 800

7 6 5 4 3 2 1 0

8 7 6 5 4 3 2 1

100

200

time [min]

0 0

100

200

300

400

500

600

700

800

time [min]

Figure 1: A comparison of results, Lines: our scheme (constructed), symbols: FVS.

6. Acknowledgements
This work is partially supported by Higher Education Commission (HEC) of Pakistan.

References
R. Grosch, H. Briesen, W. Marquardt, M. Wulkow, (2007), Generalization and numerical investigation of QMOM, AIChE J. 53, 207-227.

H.M. Hulburt, S. Katz, (1964), Some problems in particle technology, Chem. Eng. Sci. 19, 555-574. B. Koren, (1993), A robust upwind discretization method for advection, diffusion and source terms. In: C.B. Vreugdenhil and B. Koren, editors, Numerical Methods for Advection-Diffusion Problems, Volum 45 of Notes on Numerical Fluid Mechanics, Vieweg Verlag, Braunschweig, 117-138. S. Qamar, S. Mukhtar, A. Seidel-Morgenstern, M.P. Elsner, (2009), An efficient numerical technique for solving one-dimensional batch crystallization models with sizedependent growth rates, Chem. Eng. Sci. 64, 3659-3667. D. Ramkrishna, (2000), Population balances: Theory and applications to particulate systems in engineering, Academic Press. D. Randolph, M.A. Larson, (1988), Theory of particulate processes, second edition, Academic Press.