Journal of Food Engineering 76 (2006) 261–271

www.elsevier.com/locate/jfoodeng

Simulation of continuous physical refiners for edible

oil deacidification
Roberta Ceriani, Antonio J.A. Meirelles *

EXTRAE (Laboratory of Extraction, Applied Thermodynamics and Equilibrium), Food Engineering Department,
State University of Campinas (UNICAMP), Cidade Universitária Zeferino Vaz, Campinas, São Paulo 13083-862, Brazil

Received 12 February 2004; received in revised form 5 May 2005; accepted 25 May 2005
Available online 2 August 2005

Abstract

Physical refining, also known as deacidification by steam distillation, is a process in which free fatty acids and other volatile com-
pounds are distilled off from the oil, using an effective stripping agent (usually steam) and suitable processing conditions. In general,
vegetable oils are formed by triacylglycerols, free fatty acids, partial acylglycerols and some minor compounds, constituting a com-
plex mixture.
This work presents the implementation of a simulation program applied to the physical refining of edible oils, following a multi-
component-stripping tray-column design. Two singular vegetable oils, from palm and coconut, were selected. Their entire compo-
sitions were considered within the simulations, including acylglycerols and free fatty acids. To describe the required physical
properties, group contribution methods and empirical equations were selected. The results were analyzed and compared with the
literature in terms of neutral oil loss and final oil acidity, as a function of the processing conditions. Murphree efficiencies and
entrainment were considered for each plate.
 2005 Elsevier Ltd. All rights reserved.

Keywords: Simulation; Multicomponent-stripping column; Physical refining; Vegetable oils; Vapor–liquid equilibria; Distillative neutral oil loss;
Murphree efficiency; Entrainment

1. Introduction Petrauskaité, DeGreyt, & Kellens, 2000; Verleyen

et al., 2001).
The physical refining process has been under interest Physical refining of vegetable oils is based on the
and research, as a suitable alternative for the caustic higher volatility of free fatty acids (FFA) compared with
refining process (Hodgson, 1996; Stage, 1985; Tandy & triacylglycerols (TAG) at the processing conditions
McPherson, 1984). In the case of palm oil, it practically (high temperatures and low pressures). Throughout the
replaced the use of chemical refining (Hodgson, 1996), process, complex chemical and physical phenomena do
as a consequence of the high acidity of this oil. The effect occur, and their main physical effects are the vaporiza-
of the processing parameters in the final products tion of volatile compounds and the entrainment or
(refined oil and distillate) has also been widely studied mechanical carry-over of neutral oil droplets within
by many authors (Maza, Ormsbee, & Strecker, 1992; the stripping steam (Maza et al., 1992). Hydrolysis of
TAG is one of the chemical reactions, and its impor-
tance is related to the generation of FFA during the
*
Corresponding author. Tel.: +55 19 37884037; fax: +55 19
physical refining process. Nevertheless, the hydrolysis
37884027. reaction has only a slight influence on the final product
E-mail address: [email protected] (A.J.A. Meirelles). quality. According to Carlson (1996), the final FFA

0260-8774/$ - see front matter  2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jfoodeng.2005.05.026
262 R. Ceriani, A.J.A. Meirelles / Journal of Food Engineering 76 (2006) 261–271

content can be reduced to a value not lower than 0.005% should be included in a future work. According
due to the hydrolysis effect, but such value is very low to Ruiz-Méndez, Márquez-Ruiz, and Dobarganes
if compared with the limit required (0.3% w/w) by the (1996), hydrolysis can be avoided using nitrogen as the
Codex Alimentarius (1999). stripping gas. Economics aspects would dictate this
Besides the loss of neutral oil due to entrainment, choice.
there is also a distillative loss, caused by the vaporiza-
tion of an acylglycerol part from the oil, known as neu-
tral oil loss (NOL) (Petrauskaité et al., 2000; Verleyen 2. Modeling a continuous physical refiner
et al., 2001). As highlighted by some authors (Ceriani
& Meirelles, 2004a; Petrauskaité et al., 2000), this loss As described by Ahrens (1998), an industrial contin-
is determined mainly by the distillation of monoacylgly- uous physical refiner is based on a series of steam-agi-
cerols (MAG) and diacylglycerols (DAG). tated vertical trays stacked in a cylindrical shell. In
In addition to the high FFA content of coconut and this equipment, the oil is refined by flowing over a series
palm oils, which is an indicative for the use of physical of four to five trays (Stage, 1985), wherein the intimate
refining (Tandy & McPherson, 1984), both oils have sin- contact with the stripping steam is achieved. The refined
gular compositions. Coconut oil typifies lauric oils, with oil is collected at the bottom (first stage) of the equip-
its high level of short-chain fatty acids (FA) and low ment (Balchen, Gani, & Adler-Nissen, 1999). It is inter-
unsaturation (Petrauskaité et al., 2000), while palm oil esting to note that two feasible flow patterns are used
is rich in palmitic and monounsaturated FA (Firestone, industrially: countercurrent flow (CCF, Fig. 1A) and
1999). cross-flow (CRF, Fig. 1B). Tray physical refiners are
Computer simulation has been used in the investiga- usually designed in CRF while thin-film equipments
tion of some traditional processes of the food industry, are based in the countercurrent contact between gas
as batch steam refining of coconut oil (Ceriani & Meir- and oil (Balchen et al., 1999). In this work, we compared
elles, 2004a), continuous deodorization of a selection of the performance of both column configurations. In a
mono- and polyunsaturated vegetable oils (Ceriani & previous work, Ceriani and Meirelles (2004c) simulated
Meirelles, 2004c), and whisky manufacture (Gaiser the deodorization of soybean and canola oils in CRF
et al., 2002). The good performance of the simulation tray deodorizers.
programs developed in these works emphasizes the The multicomponent-stripping column modeling pro-
applicability of this kind of tool for complex systems. posed in this work was based on the method described
Ceriani and Meirelles (2004c) compared with success by Naphtali and Sandholm (1971) and the computa-
the tocopherol retention in soybean oil submitted to tional program was developed in MatLab. The general
deodorization in industrial scale equipments (18,000– equations, which include material- and energy balances
23,000 kg/h) following data from Maza et al. (1992). and Murphree efficiencies coupled with vapor–liquid
Unfortunately, we found few research papers reporting
experimental data for continuous physical refining of
coconut and palm oils. One of interest was given by
Chu, Baharin, Quek, and Man (2003) about the content
of fatty compounds in the palm fatty acid distillate
(PFAD): 85% of FFA and 13.7% of acylglycerols. Some
feasible comparison with this work was done along the
text.
This work presents the implementation of a simula-
tion program, developed with MatLab, for a multi-
component-stripping column applied to the physical
refining of edible oils. The vapor–liquid equilibria
(VLE) of these fatty systems were described by a group
contribution equation for vapor pressures (Ceriani &
Meirelles, 2004b) and the UNIFAC model for activity
coefficients (Ceriani & Meirelles, 2004b; Fornari, Bot-
tini, & Brignole, 1994; Reid, Prausnitz, & Poling,
1987). The entire oil composition in terms of fatty com-
pounds (TAG, DAG, MAG and FFA) was considered
within the simulations in order to analyze the losses of
neutral oil due to vaporization and entrainment during
physical refining. Acidity generation (from hydrolysis Fig. 1. Schematic flow diagram of a continuous tray design: (A) in
of acylglycerols) was not considered at this time and countercurrent flow, and (B) in cross-flow.
 R. Ceriani, A.J.A. Meirelles / Journal of Food Engineering 76 (2006) 261–271 263

equilibrium relationships, are described separately for approach developed by Ceriani and Meirelles (2004b)
CRF and CCF configurations in the Appendixes A and the UNIFAC model. As confirmed by previous
and B. This method uses Newton–Raphson as the itera- works (Ceriani & Meirelles, 2004b; Fornari et al.,
tive procedure for simultaneous convergence until the 1994), UNIFAC r 3/4 is more suitable in the prediction
true values of ln,i (the component liquid molar flow), of VLE of fatty mixtures.
vn,i (the component vapor molar flow) and the tempera-
ture T are found. Input data to the calculations were as 2.2. Estimation of oil composition and physical properties
follows:
Crude vegetable oils are formed by TAG, DAG and
• Identification and division of the mixture compo- MAG, in addition to a variety of other minor com-
nents in usual groups, according to each group con- pounds. This work is specifically concerned with the
tribution method selected. behavior of fatty components (FA and acylglycerols)
• Data bank with parameters for model UNIFAC within the physical refining process. To estimate the
(Fornari et al., 1994; Reid et al., 1987) and for predic- entire composition of palm and coconut oils in terms
tive equations for critical properties (Reid et al., of TAG, the methodology suggested by Antoniosi
1987), second virial coefficients (Reid et al., 1987), Filho, Mendes, and Lanças (1995), which is based in a
vapor pressures (Ceriani & Meirelles, 2004b; DIPPR, statistical procedure, was used. The TAG compositions
2003) and enthalpies (DIPPR, 2003; Tu & Liu, 1996; were estimated from the FA compositions, considering a
Reid et al., 1987). percentage of trisaturated TAG: 7.9% for palm oil
• Specification of the processing variables: number of (Chuan Ho, 1976) and 84% for coconut oil (Theme,
stages, Murphree efficiencies, feed location and com- 1968) approximately. The compositions in DAG and
position, feed flow rates and thermal states, and col- MAG were obtained from the estimated TAG compo-
umn pressure. sition in the following way: each TAG was split into
• Initial estimation for the total molar flow rates for 1,2- and 1,3-DAG; each DAG was then split into
each stage and for the temperature profile. MAG following the stoichiometric relations of the prior
compounds. This procedure was already applied suc-
cessfully by Ceriani and Meirelles (2004a, 2004c).
2.1. Vapor–liquid equilibria In order to enhance our analyses, two different com-
positions for each oil (one more volatile, MV, rich in
Physical refining is essentially a mass transfer process short-chain fatty acids, and other less volatile, LV, rich
and therefore its VLE has to be described. To calculate in long-chain fatty acids), were selected using the con-
the VLE, for each compound in each tray, the thermo- centration ranges given by Firestone (1999). Table 1
dynamic approach of Ceriani and Meirelles (2004b) brings some relevant aspects on the composition of the
was applied. The VLE model adopted in this work is oils selected. For the sake of simplicity, Table 2
shown below: gives the complete composition for each class of
y i ci  fio fatty compounds just for the more volatile palm oil
ki ¼ ¼ ð1Þ (MVPO).
xi P  /i
Due to the high number and diversity of compounds
where found in the physical refiner, predictive group contribu-
 L  tion methods were selected to describe the physical
V i  ðP  P vp
i Þ
fio ¼ P vp sat
i  /i  exp ð2Þ properties required for the enthalpy balances and equi-
RT
librium conditions. To calculate activity coefficients,
where fio is the standard-state fugacity, xi and yi are the UNIFAC r 3/4 was used (Fornari et al., 1994). Fugacity
molar fractions of component i in the liquid and vapor coefficients were calculated using the virial equation
phases, respectively, P is the total pressure, R is the truncated at the second term in combination with mix-
gas constant, T is the system absolute temperature, P vp i ing rules (Reid et al., 1987). Critical properties and acen-
and /sat
i are, respectively, the vapor pressure and the tric factors of the pure components are needed to
fugacity coefficient of the pure component i, ci is the calculate second virial coefficients. Well-known estima-
activity coefficient, /i is the fugacity coefficient and V Li tion methods as Jobacks technique for critical volumes
is the liquid molar volume of component i. The expo- and pressures, and Fedors group contribution equation
nential term in Eq. (2) corresponds to the Poynting fac- for critical temperatures were used (Reid et al., 1987).
tor (POY). As explained in Appendixes A and B, these The V Li values for all components can be obtained with
correlations are valid for each stage n, corrected by the the modified Rackett technique (Reid et al., 1987). For
Murphree efficiency. The vapor pressure P vp i and the the energy balances, vapor and liquid heat capacities,
activity coefficient ci are calculated for each fatty and enthalpies of vaporization were required. We se-
compound using, respectively, the group contribution lected Jobacks technique for ideal vapor heat capacities,
264 R. Ceriani, A.J.A. Meirelles / Journal of Food Engineering 76 (2006) 261–271

Table 1
Simplified composition of palm and coconut oils
Fatty acid Palm oil Coconut oil
Formula Trivial name More volatile (MVPO) Less volatile (LVPO) More volatile (MVCO) Less volatile (LVCO)
(abbreviation)
Mass (%) Mass (%) Mass (%) Mass (%)
C6:0 Caproic (Co) – – 0.60 –
C8:0 Caprilic (Cp) – – 9.40 4.60
C10:0 Capric (C) – – 7.80 5.50
C12:0 Lauric (L) 1.00 0.10 48.20 43.00
C14:0 Miristic (M) 1.50 0.90 16.80 20.60
C16:0 Palmitic (P) 46.40 41.80 7.70 10.20
C16:1 Palmitoleic (Po) 0.30 0.10 – –
C18:0 Stearic (S) 4.20 5.10 2.30 3.50
C18:1 Oleic (O) 37.30 40.70 5.60 9.90
C18:2 Linoleic (Li) 9.10 10.0 1.50 2.10
C18:3 Linolenic (Ln) – 0.60 0.05 0.20
C20:0 Arachidic (A) 0.20 0.70 – 0.20
C20:1 Gadoleic (G) – – 0.05 0.20
Class of compounds Mass (%) M g/gmol Mass (%) M g/gmol Mass (%) M g/gmol Mass (%) M g/gmol
TAG 88.25 839.76 91.99 847.71 93.75 645.21 93.61 682.50
DAG 7.75a 585.36 5.34a 590.09 3.00b 442.28 3.00b 476.12
MAG 0.51a 338.68 0.44a 340.11 1.00b 267.21 1.00b 283.92
FFA 3.49 267.68 2.23 270.34 2.25 202.67 2.39 241.92
Molecular weight 752.33 786.41 596.55 630.91
Iodine value (IV) 48.12 53.99 7.58 12.83
IV rangesc 49–55 5–13
a
From Chuan Ho (1976).
b
From Loncin (1962).
c
From Firestone (1999).

Rowlinson–Bondis method for liquid heat capacities N Dg ¼ rL =ðlL  U VN Þ; N Sc ¼ lL =ðqL  DLK Þ;

(Reid et al., 1987) and the method of Tu and Liu N Re ¼ qV  U VN  hw =ðlL  FAÞ ð3cÞ
(1996) for estimation of the enthalpy of vaporization
2
of each compound involved. The physical properties where rL is the surface tension of the liquid (kg/s ), lL is
(V Li , vapor and liquid heat capacities, and enthalpies the liquid viscosity (kg/s m), UVN is the vapor superficial
of vaporization) of water (stripping steam) were calcu- velocity (m/s), qL and qV are the liquid and vapor den-
lated using the equations from the Design Institute for sities (kg/m3), respectively, hw is the weir height (m), FA
Physical Properties Chemical Database (DIPPR, 2003). is the fractional free area (the ratio of the fractional area
of the holes to the column free cross-sectional area) and
2.3. Murphree efficiencies and entrainment (or DLK is the liquid diffusivity (m2/s) at infinite dilution
mechanical carry-over) from Wilke–Chang equation (Reid et al., 1987). All
the physical properties mentioned above were calculated
For vapor–liquid separation process, considering that using well-known methods from Reid et al. (1987), ex-
both liquid and vapor phases are mixed perfectly, a mea- cept the oil viscosity, which was obtained using the
sure of the efficiency of mass transfer for a component i method of Rabelo, Batista, Cavaleri, and Meirelles
on nth stage, is the Murphree efficiency (Ludwig, 1995). (2000).
As suggested by Ludwig (1995), the Murphree efficiency All gn,i values (Murphree efficiencies for each compo-
(g) can be estimated using the equations of MacFarland, nent at each tray; see Appendixes A and B) were consid-
Sigmund, and Van Winkle (1972), as a function of ered equal to g, which is a simplifying approximation.
dimensionless groups: the surface tension number, Rigorous methods to estimate gn,i have already been
NDg, the liquid Schmidt number, NSc, and the modified developed (Rao, Goutami, & Jain, 2001), but their
Reynolds number NRe. The two models are given implementation is very complicated and out of the scope
below: of this work. To calculate g, we considered a typical load
of palm oil (4425 kg/h) at 3 mmHg, 260 C and 1% of
g ¼ 7.0  ðN Þ0.14  ðN Þ0.25  ðN Þ0.08
Dg Sc Re ð3aÞ stripping steam. The mixture involved in this calculation
0.1 0.115
was characterized by pseudo-compounds: an equivalent
g ¼ 6.8  ðN Re  N Sc Þ  ðN Dg  N Sc Þ ð3bÞ TAG (same molecular weight and unsaturations of the
 R. Ceriani, A.J.A. Meirelles / Journal of Food Engineering 76 (2006) 261–271 265

Table 2
Complete estimated composition of the more volatile palm oil (MVPO)
TAG DAG
a b
Group Major TAG M (g/gmol) Mass (%) Mole (%) Acylglycerol compound Mb (g/gmol) Mass (%) Mole (%)
44:0 LPP 750 0.41 0.46 LM– 484 0.04 0.04
46:0 MPP 778 0.59 0.64 LP– 512 1.33 1.52
48:0 PPP 806 5.11 5.32 MP– 540 2.28 2.47
50:0 PPS 834 1.28 1.29 PP– 568 23.43 24.15
52:0 PSS 862 0.16 0.16 PS– 596 3.75 3.68
44:1 LOM 748 0.08 0.08 PA– 624 0.35 0.33
46:1 LOP 776 1.79 1.94 LO– 538 1.10 1.20
48:1 MOP 804 2.78 2.91 MO– 566 1.68 1.74
50:1 POP 832 33.73 34.04 PO– 594 46.37 45.69
52:1 POS 860 5.66 5.52 SO– 622 1.73 1.63
54:1 POA 888 0.49 0.46 AO– 650 0.04 0.04
46:2 LLiP 774 0.44 0.48 LLi– 536 0.35 0.38
48:2 MLiP 802 1.09 1.14 MLi 564 0.78 0.81
50:2 PLiP 830 9.19 9.30 PLi– 592 10.77 10.65
52:2 POO 858 19.32 18.90 OO– 620 4.50 4.25
54:2 SOO 886 1.63 1.54 OLi– 618 1.50 1.42
56:2 OOA 914 0.07 0.07 MAG
48:3 LOLi 800 0.23 0.24 L– 274 1.27 1.57
50:3 MOLi 828 0.38 0.39 M– 302 2.26 2.53
52:3 POLi 856 8.85 8.68 P– 330 54.83 56.28
54:3 OOO 884 3.23 3.08 S– 358 2.9 2.74
52:4 PLiLi 854 1.11 1.09 O– 356 31.65 30.11
54:4 OOLi 882 1.93 1.84 Li– 354 6.93 6.63
54:5 OLiLi 880 0.45 0.43 A– 386 0.16 0.14
a
Isomer set including different triacylglycerols, but all with the same number of fatty acid carbons and double bounds. For example, Group 50:3
means the isomer set of triacylglycerols with 50 fatty acid carbons and three double bounds. Moreover, groups with a total concentration of TAG
lower than 0.05% were ignored.
b
M, molecular weight (g/gmol). For other abbreviations see Table 1.

palm oil), an equivalent FFA (same molecular weight to the densities of the vapor and liquid, and to the per-
and unsaturations of the palm oil acidity) and water. centage of flood. The w value from our calculations was
The calculated values of g were 61.876% using Eq. 0.024 moles of liquid entrained/moles of liquid down-
(3a) and 49.986% (gn,i = 0.50) using Eq. (3b). In this flow. The entrainment term, en (moles of liquid en-
case, Eq. (3b) gives a more conservative design basis, trained/moles of vapor), of Eqs. (A.1), (A.2), (A.6)
and was chosen to represent the Murphree efficiency in and (A.7) is related to w, by the following equation
our work. The approximation of the whole multicompo- (Ludwig, 1995):
nent mixture by an equivalent one, containing pseudo-
w Ln
compounds, was used only in the case of efficiency en ¼  ð4Þ
estimation. 1w Vn
Mechanical entrainment of oil is defined as the carry The calculated value for en, using Eq. (4), was 0.4324
of oil droplets upward from the free surface by the vapor moles of liquid entrained/moles of vapor. In particular,
to the outlet of the equipment (Carlson, 1996). Accord- for CCF tray equipments, part of the liquid of lower
ing to Carlson (1996), entrainment separators are capa- concentration of the plate below (n  1) is carried by
ble of preventing the mechanical carry-over of oil the vapor to the plate above (n), reducing its liquid con-
droplets, thereby reducing entrainment losses to less centration with respect to the more volatile compounds.
than 0.1–0.2%, with an additional loss of about 10% As a consequence, the vapor rising from this plate (n)
of the FFA content in the feed. In order to evaluate will be also of lower concentration, reducing the net
the effects of mechanical entrainment in some of our amount of mass transfer and the efficiency as well. The
simulations, we used the graphical method of Fair & Murphree efficiency must be corrected to take this effect
Mathews (see Ludwig, 1995) for bubble–cap and perfo- into account (Ludwig, 1995). The wet Murphree effi-
rated plates. This correlation was developed on the basis ciency, gw, can be easily calculated as a function of w
of a flow parameter (Ln/Vn Æ (qV/qL)0.5) and a capacity (Ludwig, 1995):
parameter (UVN Æ (qV/(qL  qV))0.5). It relates the frac-  
tional entrainment w (moles of liquid entrained/ moles w
gw ¼ g 1þg ð5Þ
of liquid downflow) to the liquid and vapor flow rates, 1w
266 R. Ceriani, A.J.A. Meirelles / Journal of Food Engineering 76 (2006) 261–271

All gn,i values in the equilibrium relationships of of TAG changed slightly, increasing from the 1st to the
countercurrent pattern (see Appendix A) were consid- 5th stage. In general, the CCF configuration generated
ered equal to gw (0.494). For CRF deodorizers gn,i val- lower liquid mass flow rates of FFA, MAG, DAG and
ues (see Appendix B) were set as equal to g (0.50). TAG for all stages. Fig. 2B and E give the profiles of
For CCF deodorizers, the pressure drops of the trays the vapor mass flow rates of FFA, MAG, DAG and
are added so that the pressure at the bottom of the col- TAG, respectively for the CCF and the CRF columns.
umn is greater than the top pressure. For CRF deodor- It is interesting to note that all mass flow rate curves
izers, the pressure drop is negligible. of Fig. 2B had higher values, indicating that this behav-
ior was a consequence of the flow pattern, i.e., for the
same operational conditions, more volatiles were carried
3. Results and discussion by the vapor to the plate above in the case of the CCF
configuration. For both flow patterns, there was a sharp
To evaluate the physical refining of palm and coconut increase in the values of the liquid mass flow rate of
oils, we selected from the literature (Ahrens, 1998; Maza FFA and MAG from the 4th to the 5th stage, but this
et al., 1992; Petrauskaité et al., 2000; Rossi, Gianazza, enhance was even higher for the CCF column, because
Almprese, & Stanga, 2001; Willems, 1985) usual pro- at the 5th stage the fresh oil contacted a higher mass
cessing conditions for oil load (kg/h), percentage of flow rate of vapor (total feed of steam plus the volatiles
stripping steam, temperature and pressure. Two compo- of all stages below). Note that the DAG curve also had
sitions were also considered for each type of oil: MVPO, this increase only in the CCF configuration, due to the
MVCO, LVPO and LVCO. The processing parameters same reason.
studied in this work are shown in Table 3. As one can As shown in Fig. 2C and F, the difference between
see, the combination of these parameters led to a consid- temperatures at the top and the bottom of the column
erable number of possibilities for simulation. The most was less than 0.9 C and 1.5 C, respectively for the
interesting ones are discussed below. CCF and the CRF patterns. As one can see, the differ-
From the parameters given in Table 3, we selected a ences between the temperatures of each stage of the
typical physical refining (267 Pa, 1% of stripping steam CCF and the CRF configurations varied from 0.3 C
and 260 C) of the MVPO to show the profiles of tem- to 0.9 C. Note that the temperatures of the CRF col-
perature and of the vapor and liquid flow rates (kg/h) umn were always higher. At a first insight, one might ex-
within the stripping column (see Fig. 2). The simulations pect that the temperature profile in the CCF column
considered Murphree efficiencies, despising however would be higher than in the CRF column, due to the
entrainment effects. Because our simulation program pressure drop effect that increases the pressure of each
permitted the investigation of different flow patterns, stage in the CCF column. It is important to observe
we studied the effect of CCF (Fig. 1A) and CRF that, besides this effect, the amount of the vapor stream
(Fig. 1B) configurations on the results. Fig. 2A and D that pass through each stage in the CCF column, and
show the profile of the mass flow rates of FFA, MAG, therefore the concentration of water dissolved in the
DAG and TAG in the liquid within the column, respec- liquid phase, are much higher than in the CRF configu-
tively for CCF and CRF equipments. As the oil flowed ration. According to Ceriani and Meirelles (2004a), very
down into the column (from stage 5 to 1), the liquid small amounts of steam (water), that condense and dis-
mass flow rates of FFA, MAG and DAG decreased as solve in the oil, decrease the boiling temperature of the
a consequence of their vaporization. The mass flow rate mixture. Our simulations showed that this second effect
(amount of steam) prevailed, i.e. the temperatures of
each stage in the CCF column were lower than the ones
Table 3
Operating conditions for the continuous physical refiner simulations in the CRF column. In fact, in the CCF column, the
water was present in the liquid phase in mass fractions
Parameter Palm oila Coconut oilb
c
between 3.0E8 to 1.1E6 while in the CRF column
% Stripping steam 1.0, 2.0 and 3.0 0.5, 1.0 and 2.0
this range was 1.7E8 and 1.8E7. To confirm this
Temperature (C) 240, 260 and 270 190, 210, 230 and 260
Top pressure (Pa) 267, 400 and 667 133, 267 and 400 statement, we simulated the processing conditions of
Oil load 4425 kg/hd Fig. 2 in CCF pattern but with 0.2% of stripping steam,
Number of trays 5e in a way that the amount of steam passing through each
a
From Rossi et al. (2001) and Willems, 1985. stage was in the same range of the CRF with 1.0% of
b
From Petrauskaité et al., 2000. stripping steam (0.2% in each of the five stages). As
c
Calculated as percentage of total oil load. The stripping steam is expected, the temperatures of the CCF column were
distributed equally in the five stages of the column in the CRF design
always higher, because now the pressure drop effect
and as a single feed (1st stage) in the CCF design.
d
Ahrens (1998). The oil feed is introduced into the column at the top prevailed.
stage (5th). For the processing conditions of Fig. 2 (CRF pat-
e
Stage (1985). tern), 188.49 kg/h of distillate was formed with 79.8%
 R. Ceriani, A.J.A. Meirelles / Journal of Food Engineering 76 (2006) 261–271 267

Fig. 2. Profiles for the mass flow rates of FFA and acylglycerols in the liquid phase for the (A) CCF and (D) CRF design, in the vapor phase for the
(B) CCF and (E) CRF design, and for the temperature in the (C) CCF and (F) CRF stripping column. Code: Vapor phase FFA (,), TAG (h), DAG
(s) and MAG (n); Liquid phase FFA (.), TAG (j), DAG (d) and MAG (m).

of FFA and 19.8% of acylglycerols, values not far from of NOL (0.86% for CRF and 0.83% for CCF) and OA
the observation of Chu et al. (2003). For comparison (0.084% for CRF and 0.085% for CCF, in % of oleic
purposes, we report that, at 240 C, the mass flow rate acid) were achieved for both configurations, but in the
of distillate reduced to 155.37 kg/h, composed by case of CRF, the steam consumption must be kept at
86.2% of FFA and 12.5% of acylglycerols. In general, 1.0%. This result is very interesting since it indicates that
the concentrations of acylglycerols and FFA in the dis- with the CCF configuration, the same quality of the final
tillate varied from 9.4% to 26.0%, and from 72.7% to product would be obtained, with half of the steam con-
89.3%, respectively, for all the processing conditions sumption. Ahrens (1999) informs that the steam con-
studied for the MVPO. sumption of thin-film deodorizers (also in CCF) is cut
In comparison with the values found for CRF design, to a third of the amount required in conventional
the CCF always generated lower levels of final oil acidity deodorizers.
(OA) and higher values of NOL for all the conditions To illustrate the feasible separation of the classes of
studied for the physical refining of the MVPO (see fatty compounds (FFA, MAG, DAG and TAG) within
Fig. 3). To elucidate this point, in the case of CCF, we the column, we report the Kn,i (distribution coefficient)
simulated a condition with a reduced percentage of values (see Eq. (2)) for stage number 5 at one of the pro-
steam (0.5%, 50% less), while maintaining the other cessing conditions of Fig. 3 (260 C, 267 Pa and 1% of
parameters fixed (260 C and 267 Pa). Similar values stripping steam, CRF). The comparison of Kn,i values
268 R. Ceriani, A.J.A. Meirelles / Journal of Food Engineering 76 (2006) 261–271

Fig. 3. Final oil acidity (OA) and neutral oil loss (NOL) for physical refining of the MVPO with 1% of stripping steam for (A) CCF and (B) CRF
patterns. Code: OA (h) and NOL (n).

gives an idea about the facility of the classes of Table 4

Effects of processing conditions on OA and NOL in the physical
compounds in a mixture to be distilled. The calculated
refining of coconut oil (MVCO and LVCO) with 1% of stripping steam
values for the MVPO were as follows: K5,TAG = 0.002,
P (Pa) T (C) NOL (%) OAa (%)
K5,DAG = 0.27, K5,MAG = 13.11 and K5,FFA = 35.91.
Note that the Kn,i values were inversely proportional LVCO
133 190 0.69 0.575
to the molecular weight of the compound class (see
210 1.09 0.231
Table 1). 230 1.77 0.053
In general, the volatility of FFA increased with the 260 4.52 0.002
temperature, vacuum magnitude and percentage of 267 190 0.50 0.793
stripping steam. At higher temperatures, DAG and 210 0.86 0.384
MAG volatility also reached important values and these 230 1.37 0.120
compounds were stripped off from the oil, increasing 260 3.24 0.006
NOL. The limiting factor for choosing processing condi- 400 190 0.40 0.951
tions is the neutral oil loss (NOL), which was the object 210 0.72 0.501
of our analysis. Table 4 gives the final OA and NOL val- 230 1.18 0.189
260 2.66 0.013
ues, for the MVCO and the LVCO. Note that the NOL
values were more significant for the MVCO, even at MVCO
lower temperatures, because its acylglycerol fractions 133 190 0.89 0.369
were more volatile (see Table 1) in comparison to the 210 1.47 0.141
LVCO fractions. As a consequence of the increase in 230 2.57 0.032
260 6.95 0.001
the volatility of the acylglycerol fractions, 31–52% more
losses of neutral oil were observed. The same association 267 190 0.66 0.530
can be made comparing the NOL values found for the 210 1.13 0.238
230 1.91 0.072
physical refining of the MVPO and LVPO.
260 4.92 0.004
The results discussed in the previous paragraphs evi-
dence the important contribution of the vaporization of 400 190 0.54 0.657
210 0.95 0.317
partial acylglycerols in the NOL values. Besides losses 230 1.60 0.115
caused by distillation, there is also a loss due to entrain- 260 3.98 0.008
ment. In fact, NOL is a consequence of two different a
Expressed as percentage of lauric acid (% C12:0) in the refined oil.
contributions: the vaporization of acylglycerols (mainly
DAG and MAG, which are more volatile than TAG)
and the mechanical carry-over of liquid oil droplets, to report the results for the simulations considering
composed mostly by TAG. entrainment just for the CCF configuration. Note, how-
For a better understanding of these two effects, we ever, that similar conclusions would arise from the anal-
separated them in Fig. 4 for the physical refining of ysis of the results for the CRF column. As one can see,
the MVCO at different processing conditions. Because in Fig. 4A, the contribution of distillation for the losses
in the CCF design, the mechanical carry-over of the of neutral oil increased with temperature and percentage
oil by the rising vapor stream affects the VLE and of stripping steam. At 230 C and 260 C, losses due to
changes the phase compositions of each stage, we chose distillation were even more important than the mechan-
 R. Ceriani, A.J.A. Meirelles / Journal of Food Engineering 76 (2006) 261–271 269

Fig. 4. Contribution of distillation and entrainment to (A) the total loss of neutral oil (NOL) at 133 Pa and to (B) the loss of TAG at 210 C in the
physical refining MVCO.

ical carry-over (except for the condition 230 C and results were less than 0.001 in the vapor phase, and
0.5% of steam), which was a direct consequence of the 0.00003 in the liquid phase. In general, the /sat
i values
increase in the volatility of the acylglycerols at these were not lower than 0.85 for water and 0.96 for the fatty
temperatures. Fig. 4B shows the values for TAG losses, compounds. In this way, it might be interesting to use,
as a function of the percentage of stripping steam and eventually, the /sat
i values only for water and some other
system pressure. As one can see, even at 210 C, the selected fatty compounds of low molecular weight.
TAG fraction of coconut oil was vaporized. As ex-
pected, the entrainment increased TAG losses several
times (minimum of seven times at 133 Pa and 2% of 4. Conclusion
steam; maximum of 23 times at 400 Pa and 0.5% of
steam). It should be emphasized that physical refiners This work indicated that the tools developed by
are usually equipped with entrainment separators to Ceriani and Meirelles (2004b) for calculating vapor
trap small oil droplets, reducing entrainment losses pressures and evaluating the VLE of multicomponent
(Carlson, 1996). In our simulations, we considered fatty mixtures, coupled with the multicomponent-strip-
entrainment for all stages, including the Nth stage, ping column program, allowed the simulation of
where this kind of apparatus acts. The NOL values re- different processes of the vegetable oil industry. As illus-
ported may be over-estimated but they are an indicative trated here, it is possible to build a computerized routine
of the necessity of using such improvements in physical to design a multicomponent-stripping column, includ-
refiners. It should be emphasized that this work did not ing the rigorous estimation of the phase equilibria
consider losses of acylglycerols by hydrolysis during the involved.
process. The software developed in the present work could be
The main effect related to the inclusion of Murphree a suitable tool for the design of stripping columns, since
efficiencies within the equilibrium relationships was an it permits to investigate the composition (quality) of the
increase in the final oil acidity. In this case, more stages refined oil and byproducts (as the distillate), as a func-
would be required to achieve the same final content of tion of processing variables (temperature, % stripping
FFA in the refined oil. steam and pressure). Murphree efficiencies and entrain-
An important consideration about the VLE model ment had also important effects on the final products.
adopted has to be discussed. As already highlighted by The combination of all these aspects could also lead to
Ceriani and Meirelles (2004c), the calculated values of a better primary selection of the operating conditions
/i and POY for all components in the system (acylglyce- of a real plant.
rols + FFA + water) were equal or very close to unity
while the /sat
i values, specifically for water and short-
chain fatty acids, deviated notably from unity, as a con- Acknowledgements
sequence of the high values of vapor pressures for these
compounds at the operating conditions. This work The authors wish to acknowledge FAPESP (Funda-
showed however, that the results from the rigorous ção de Amparo à Pesquisa do Estado de São Paulo—
VLE model (Eq. (1)) produced no significant difference 03/04949-3 and 05/53095-2) and CNPq (Conselho
if compared with the results obtained considering the Nacional de Desenvolvimento Cientı́fico e Tecnológico
vapor phase as ideal. The differences in the mass fraction —521011/95-7) for the financial support.
270 R. Ceriani, A.J.A. Meirelles / Journal of Food Engineering 76 (2006) 261–271

Appendix A. Equations for the continuous multitray liquid flows, based on an estimated value for Ln (total
countercurrent flow design oil feed despising acidity) and for Vn, which is set as
the total steam feed plus acidity. The derivatives of test
For an arbitrary stage n of a stripping column, the re- functions (Jacobian matrix) with respect to temperature
lated nomenclature can be set as follows. are found analytically, while those with respect to com-
Subscript n: flow from stage n, n = 1, 2, . . .,NS; sub- ponent flow rates are found numerically.
script i: component i, i = 1, 2, . . ., NC; H = vapor phase
enthalpy (J/h); h = liquid phase enthalpy (J/h); hf =
liquid feed enthalpy (J/h); Hf = vapor feed enthalpy Appendix B. Equations for the continuous multitray
(J/h); V = total vapor flow (mol/h); v = component cross-flow design
vapor flow (mol/h); L = total liquid flow (mol/h); l =
component liquid flow (mol/h); f = component feed Component balances (total: NS · NC relations)
flow as liquid (mol/h); F = component feed flow as
ln;i
vapor (mol/h). F 1ðn;iÞ ¼ ln;i þ vn;i þ V n  en   lnþ1;i  fn;i  F n;i ¼ 0
For each stage n, a set of dependent relationships Ln
(test functions Fk(n,i)) must be satisfied. In the equations ðA:6Þ
below, the entrainment term (en) is already introduced. Enthalpy balances (total: NS relations)
Component balances (total: NS · NC relations)
hn
ln;i F 2ðnÞ ¼ hn þ H n þ en  V n   hnþ1  hf ;n  H f ;n ¼ 0
F 1ðn;iÞ ¼ ln;i þ vn;i þ V n  en   vn1;i  V n1 Ln
Ln
ðA:7Þ
ln1;i
 en1   lnþ1;i  fn;i  F n;i ¼ 0 ðA:1Þ Equilibrium conditions derived from the definitions of
Ln1
the vapor phase Murphree plate efficiency, gn,i (total:
Enthalpy balances (total: NS relations) NS · NC relations)
hn ln;i
F 2ðnÞ ¼ hn þ H n þ en  V n   H n1  en1 F 3ðn;iÞ ¼ gn;i  K n;i  V n   vn;i þ ð1  gn;i Þ
Ln Ln
hn1 F n;i
 V n1   hnþ1  hf;n  H f;n ¼ 0 ðA:2Þ VnP ¼0 ðA:8Þ
Ln1
i F n;i
Equilibrium conditions derived from the definitions of
the vapor phase Murphree plate efficiency, gn,i (total:
NS · NC relations)
References
ln;i
F 3ðn;iÞ ¼ gn;i  K n;i  V n   vn;i þ ð1  gn;i Þ Ahrens, D. (1998). Comparison of tray, thin-film deodorization.
Ln
Inform, 9(6), 566–576.
Vn
 vn1;i  ¼0 ðA:3Þ Ahrens, D. (1999). Industrial thin-film deodorization of seed oils with
V n1 SoftColumn technology. Fett/Lipid, 101(7), 230–234.
Antoniosi Filho, N. R., Mendes, O. L., & Lanças, F. M. (1995).
The above relationships comprise a vector of test
Computer prediction of triacylglycerol composition of vegetable
functions oils by HRGC. Journal of Chromatographia, 40(9–10), 557–
562.
FðxÞ ¼ fF1 ; F2 ; F3 g ¼ 0 ðA:4Þ
Balchen, S., Gani, R., & Adler-Nissen, J. (1999). Deodorization
principles. Inform, 10, 245–262.
which contains NS(2NC+ 1) elements, and which may
Carlson, K. F. (1996). Deodorization (5th ed.). In Y. H. Hui (Ed.).
be solved for equally many unknowns Baileys industrial oil and fat products (vol. 4, pp. 339–390). New
x ¼ fl; v; Tg ðA:5Þ York: Wiley-Interscience.
Ceriani, R., & Meirelles, A. J. A. (2004a). Simulation of batch physical
where the vector l contains all the elements ln,i, v all the refining and deodorization processes. Journal of the American Oil
elements vn,i and T all elements Tn. Chemists Society, 81(3), 305–312.
Once ln,i, vn,i and Tns are known, the product compo- Ceriani, R., & Meirelles, A. J. A. (2004b). Predicting vapor–liquid
equilibria of fatty systems. Fluid Phase Equilibria, 215, 227–236.
sitions, the product flow rates, the concentration and Ceriani, R., & Meirelles, A. J. A. (2004c). Simulation of continuous
temperature profiles in the column follow readily. The deodorizers: effects on product streams. Journal of the American Oil
iterative Newton–Raphson method solves Eq. (A.4) Chemists Society, 81(11), 1059–1069.
using the prior set of values of the independent variables Chu, B. S., Baharin, B. S., Quek, S. Y., & Man, Y. B. C. (2003).
Separation of tocopherols and tocotrienols from palm fatty acid
(Eq. (A.5)). A first estimative is necessary to initiate the
distillate using hydrolysis–neutralization–adsorption chromatogra-
calculations. This estimative considers a linear profile phy method. Journal of Food Lipids, 10(2), 141–152.
for temperature, based on the oil and stripping steam Chuan Ho, O. (1976). Studies in palm oil crystallization. Journal of the
feed temperature in each stage and for the vapor and American Oil Chemists Society, 53(10), 609–617.
 R. Ceriani, A.J.A. Meirelles / Journal of Food Engineering 76 (2006) 261–271 271

Codex Alimentarius (1999). Codex Standard for Named Vegetable Rabelo, J., Batista, E., Cavaleri, F. W., & Meirelles, A. J. A. (2000).
Oils (vol. 8, pp. 1–16). Codex Stan 210-1999, Rome. Viscosity prediction for fatty systems. Journal of the American Oil
DIPPR Student Chemical Database Login. Available from: <http:// Chemists Society, 77(12), 1255–1261.
dippr.byu.edu/students/chemsearch.asp> accessed November 2003. Rao, D. P., Goutami, C. V., & Jain, S. (2001). A direct method for
Firestone, D. (1999). Physical and chemical characteristics of oils, fats incorporation of tray-efficiency matrix in simulation of multicom-
and waxes. Washington: AOCS Press. ponent separation processes. Computers and Chemical Engineering,
Fornari, T., Bottini, S., & Brignole, E. A. (1994). Application of 25, 1141–1152.
UNIFAC to vegetable oil–alkane mixtures. Journal of the American Reid, R. C., Prausnitz, J. M., & Poling, B. E. (1987). The properties of
Oil Chemists Society, 71(4), 391–395. gases and liquids. New York: McGraw-Hill.
Gaiser, M., Bell, G. M., Lim, A. W., Roberts, N. A., Faraday, D. B. Rossi, M., Gianazza, M., Almprese, C., & Stanga, F. (2001). The effect
F., Schulz, R. A., et al. (2002). Computer simulation of a of bleaching and physical refining on color and minor components
continuous wisky still. Journal of Food Engineering, 51, 27–31. of palm oil. Journal of the American Oil Chemists Society, 78(10),
Hodgson, A. L. (1996). Refining and bleaching (5th ed.). In Y. H. Hui 1051–1055.
(Ed.). Baileys industrial oil and fat products (vol. 4, pp. 157–212). Ruiz-Méndez, M. V., Márquez-Ruiz, G., & Dobarganes, M. C. (1996).
New York: Wiley-Interscience. Comparative performance of steam and nitrogen as stripping gas in
Loncin, M. (1962). LHydrolyze Spontanée des Huiles Glycéridiques et physical refining of edible oils. Journal of the American Oil
en Particulier de lHuile de Palme. Maison DEditon, Couillet. Chemists Society, 73(12), 1641–1645.
Ludwig, E. E. (1995). Applied process design for chemical and Stage, H. (1985). The physical refining process. Journal of the American
petrochemical plants (vol. 2, pp. 42–44) (3rd ed.). Houston: GPC. Oil Chemists Society, 62(2), 299–308.
MacFarland, S. A., Sigmund, P. M., & Van Winkle, M. (1972). Predict Tandy, D. C., & McPherson, W. J. (1984). Physical refining of edible
distillation efficiency. Hydrocarbon Processing, 51(7), 111–114. oil. Journal of the American Oil Chemists Society, 61(7), 1253–1258.
Maza, A., Ormsbee, R. A., & Strecker, L. R. (1992). Effects of Theme, J. G. (1968). Coconut oil processing. Rome: FAO.
deodorization and steam-refining parameters on finished oil Tu, C. H., & Liu, C. P. (1996). Group-contribution estimation of the
quality. Journal of the American Oil Chemists Society, 69(10), enthalpy of vaporization of organic compounds. Fluid Phase
1003–1008. Equilibria, 121, 45–65.
Naphtali, L. M., & Sandholm, D. P. (1971). Multicomponent Verleyen, T., Verhe, R., Garcia, L., Dewettinck, K., Huyghebaert, A.,
separation calculations by linearization. AIChe Journal, 17(1), & De Greyt, W. (2001). Gas chromatographic characterization of
148–153. vegetable oil deodorization distillate. Journal of Chromatography
Petrauskaité, V., DeGreyt, W. F., & Kellens, M. J. (2000). Physical A, 921, 277–285.
refining of coconut oil: Effect of crude oil quality and deodorization Willems, M. G. A. (1985). Palm oil; Quality requirements from a
conditions on neutral oil loss. Journal of the American Oil Chemists customers point of view. Journal of the American Oil Chemists
Society, 77(6), 581–586. Society, 62(2), 454–459.