Journal of Food Engineering 100 (2010) 1–11

Contents lists available at ScienceDirect

Journal of Food Engineering

journal homepage: www.elsevier.com/locate/jfoodeng

Prediction of cooking times and weight losses during meat roasting

Sandro M. Goñi a,b,c, Viviana O. Salvadori a,c,*
a
Centro de Investigación y Desarrollo en Criotecnología de Alimentos (CIDCA, CONICET-La Plata), Fac. de Cs. Exactas – UNLP, 47 y 116, B1900AJJ La Plata, Argentina
b
Departamento de Ciencia y Tecnología, UNQ, R.S. Peña 352, B1876BXD Bernal, Argentina
c
MODIAL, Área Deptal. Ing. Qca., Fac. de Ingeniería, UNLP, 1 y 47, B1900TAG La Plata, Argentina

a r t i c l e i n f o a b s t r a c t

Article history: In this work, a meat oven cooking model is developed, and its ability to predict three main process vari-
Received 22 September 2009 ables – evaporative loss, dripping loss and cooking time – is evaluated. Heat transfer is modelled by Fou-
Received in revised form 19 February 2010 rier’s law, while the internal moisture content variation is modelled as a function of water demand, which
Accepted 13 March 2010
depends on the water holding capacity of beef. Experimental cooking of semitendinosus muscle samples
Available online 19 March 2010
was carried out in a convective oven to obtain general information about the process and to assess the
model accuracy. Simulations were done by means of the finite element method, using three-dimensional
Keywords:
irregular geometries as simulation domains. The model predictions were in good agreement with the
Meat cooking
Simulation
experimental ones; the average absolute relative error was 3.91% for cooking time prediction, and
Dripping loss 7.96% for total weight loss prediction.
Evaporative loss Ó 2010 Elsevier Ltd. All rights reserved.

1. Introduction weight loss is computed by establishing an evaporative flux at

the surface. In this sense, several authors have studied different
Oven cooking (or roasting) of meat is a food operation that af- meat products, i.e. beef (Goñi et al., 2008b; Obuz et al., 2002,
fects both quality attributes and microbiological safety of pro- 2004; Singh et al., 1984; Townsend et al., 1989a,b), whole un-
cessed products. In spite of the increasing importance of stuffed turkeys (Chang et al., 1998). Goñi et al. (2008b) and Obuz
industrial cooking, several aspects related to this process have et al. (2002) found that the average error between predicted and
not been sufficiently explored. From a physical point of view, oven experimental weight losses is 20–22%. Recently, Bottani and Volpi
cooking involves heat transfer from the surrounding ambient to (2009) simulated cooking of beef and turkey samples in industrial
the food surface, and consequently induces a temperature gradient steam ovens. As cooking was performed in an oven with steam
inside the product which results in an increase of internal temper- injection, the effect of water vapourization was neglected in the
ature. In addition, a significant decrease of food weight is often ob- energy balance. Their experimental measurements indicated that
served during the process, which is generally attributed to water also in this cooking condition, weight losses ranged from 15% to
loss, neglecting other food components (i.e. proteins, lipids) losses. 18%. The exhaustive analysis of these reported works indicates that
Internal transport of liquid water is due to thermal protein dena- this model will generally fail to predict total weight loss because
turation which causes the shrinkage of the meat fibres network, the liquid water loss by dripping is neglected.
resulting in a mechanical force that expels the excess interstitial Other researchers incorporated the inner mass transfer by
water towards the surface (Godsalve et al., 1977). Depending on means of the Fick’s law in the mathematical model; cooking of
heat and mass transfer conditions at surface, this expelled liquid meatballs (Huang and Mittal, 1995) and chicken patties (Chen
water can be lost by evaporation or dripping. et al., 1999) were studied on this concept. This approach can be
Regarding the mathematical modelling of meat oven cooking, useful to predict total weight loss in elaborated meat products,
different models have been proposed depending on the assump- formulated with other ingredients that reduce drip losses (i.e.
tions about internal heat and mass transfer mechanisms. A first food additives that retain water during cooking). For raw beef,
approach is to consider internal heat transfer occurring by conduc- unrealistic large diffusion coefficients are needed in order to pre-
tion (Fourier’s law) while neglecting internal mass transfer; then, dict total weight loss (Burfoot and Self, 1989). Recently, van der
Sman (2007a,b) proposed a mathematical model for meat cooking
by using the Darcy’s law to describe the water flux through the
fibres. His work is in good agreement with the commonly ac-
* Corresponding author at: Centro de Investigación y Desarrollo en Criotecnología cepted description of liquid water transport stated by Godsalve
de Alimentos (CIDCA, CONICET-La Plata), Fac. de Cs. Exactas – UNLP, 47 y 116,
B1900AJJ La Plata, Argentina. Tel./fax: +54 221 425 4853.
et al. (1977). This approach can predict both evaporative and drip-
E-mail address: [email protected] (V.O. Salvadori). ping losses, though the resulting model is complex (involving

0260-8774/$ - see front matter Ó 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jfoodeng.2010.03.016
2 S.M. Goñi, V.O. Salvadori / Journal of Food Engineering 100 (2010) 1–11

Nomenclature

a1, a2, a3, TR parameters of Eq. (14) Greek symbols

aw water activity e beef emissivity
c moisture content, dry basis /w volumetric fraction of water
cwb moisture content, wet basis k latent heat of evaporation (J kg1)

C average moisture content, dry basis q meat density (kg m3)
CP specific heat of meat (J kg1 K1) r Stefan–Boltzmann constant (5.67  108 W m2 K4)
DL dripping weight loss (kg) C surface of geometric model
EL evaporative weight loss (kg) X domain of geometric model
h convective heat transfer coefficient (W m2 K1)
k thermal conductivity of meat (W m1 K1) Subscripts
kg mass transfer coefficient (kg Pa1 m2 s1) exp experimental
KW parameters of Eq. (4), kg dry solid (s1) i initial
ms dry solid mass (kg) o oven
Psat water vapour pressure (Pa) s surface
RH oven relative humidity sim simulated
SD standard deviation
T temperature (°C)
WHC water holding capacity, dry basis

parameters difficult to estimate or either unavailable values for in these situations, total weight loss is mostly evaporative, and this
some properties). model offers acceptable results. Conversely, as it does not take into
This work is part of a comprehensive study on meat cooking, account the dripping phenomena, the total weight loss will be
where the general aim is the multi-objective optimization of the notoriously underestimated when the heat and mass transfer coef-
process. In this way, the specific objective of this work was to de- ficients present low or moderate values (Goñi et al., 2008b; Obuz
velop an accurate cooking model, which allows the estimation of et al., 2002), which is common in domestic and also in some indus-
cooking times and also both evaporative and dripping losses. Be- trial ovens. In consequence, Eq. (4) is coupled to the previous ones
sides, such model should be as simple as possible in terms of com- in order to evaluate the variation of liquid water content in the
putational cost involved in numerical simulation, bearing in mind meat.
the ultimate optimization. The proposed mathematical model was
dc
validated by comparing its predicted results with experimental ms ¼ K W ðc  WHCðTÞÞ ð4Þ
dt
data of cooking time and weight loss.
This balance establishes that water content variation is directly
proportional to water demand. The water demand is the difference
2. Beef cooking modelling
between the instant water content and an equilibrium value, equal
to the water holding capacity (WHC). At this point, it is important
In general, oven cooking involves heat transfer by convection
to mention that Eq. (4) does not imply an internal mass transfer
and radiation from the surrounding ambient to the food surface.
due to diffusion and/or convection mechanisms. A similar ap-
Then, the high surface temperature induces conductive heat trans-
proach has been successful used for water and lipid content predic-
fer towards the core of the product. For modelling purposes, meat
tion in contact-heating cooking of hamburgers (Pan et al., 2000)
can be considered as a solid matrix composed principally by pro-
and pan frying of hamburgers (Ou and Mittal, 2006, 2007).
teins and liquid water. Therefore, a simple mathematical model
For a given process time t, the evaporative loss (EL) is predicted
that describes heat transfer is expressed in the below equation:
by surface integration of evaporative mass flux:
@T Z t Z 
qC P ¼ rðkrTÞ ð1Þ
ELðtÞ ¼ jev ap ðC; tÞdC dt ð5Þ
@t
0 C
Due to high ambient temperature that largely surpasses 100 °C,
the surface of meat rapidly achieves temperature values near As the proposed model does not consider the inner flux of
100 °C. As a result, a significant fraction of liquid water is evapo- water, the dripping loss at a given process time (DL) is estimated
rated, which can be quantified as a water vapour flux transferred as the difference of liquid water content:
to the oven ambient, jevap. The heat associated to evaporative flux 
DLðtÞ ¼ ms ðci  CðtÞÞ ð6Þ
is added in the boundary condition for the energy balance, Eq.  is the volume average moisture content, obtained by vol-
where C
(2). As well, this mass flux depends on the water activity of food
ume integration of the moisture profile.
surface and the relative humidity of the oven ambient (Eq. (3)).
Z
 nkrT ¼ hðT s  T o Þ þ er ðT 4s  T 4o Þ þ kjev ap ð2Þ  ¼1
CðtÞ cðX; tÞdX ð7Þ
V X
jev ap ¼ kg ðaw Psat ðT s Þ  RHP sat ðT o ÞÞ ð3Þ
Finally, the total weight loss is obtained by summing both evap-
Besides the water loss produced by superficial evaporation, orative and dripping losses.
there exist enough experimental evidence to say that the protein
matrix acts like a sponge, which losses a significant amount of li- 2.1. Thermophysical properties
quid water by dripping when receiving several stimuli (e.g.
stress–strain due to protein denaturation, pre-processing mechan- For the case of study, thermophysical properties of meat were
ical cuts) (Tornberg, 2005). For high values of the heat and mass computed according to moisture content (in wet basis, cwb) and
transfer coefficients, free liquid water can evaporate. Therefore, temperature, considering only water and proteins as major
 S.M. Goñi, V.O. Salvadori / Journal of Food Engineering 100 (2010) 1–11 3

components. Properties of individual components were deter- was placed on a coarse netting tray in the central region of the
mined according to Choi and Okos (1986). Thermal conductivity oven. Then, with the sample ready to cook, the oven was turned
was considered to be anisotropic, and the parallel (Eq. (9)) and on. Each experiment was finished, and the corresponding process
the perpendicular (Eq. (10)) values were computed from the volu- time was determined, when the core temperature reached 72 °C,
metric fraction of components. since this condition is required for microbiological safety of pro-
cessed meat products (McDonald et al., 2001). Also, this internal
cwb q
/w ¼ ð8Þ temperature corresponds to ‘‘medium” degree of doneness, accord-
qw ðTÞ ing to the Beef Steak Color Guide (AMSA, 1995, cited by López
kjj ¼ /w kw ðTÞ þ ð1  /w Þkp ðTÞ ð9Þ Osornio et al. (2008)).
1 /w 1  /w To quantify shrinkage during cooking, initial and final charac-
¼ þ ð10Þ teristic dimensions, i.e. length, width, and height, of samples were
k? kw ðTÞ kp ðTÞ
measured by using a calliper. In order to calculate total weight loss,
C P ¼ cwb C P;w ðTÞ þ ð1  cwb ÞC P;p ðTÞ ð11Þ
the samples were weighed before and after cooking. Furthermore,
1 cwb ð1  cwb Þ we experimentally attempted to determine the evaporative and
¼ þ ð12Þ
q qw ðTÞ qp ðTÞ dripping contributions to total weight loss. For this aim, an ad
Beef emissivity, which appears in the energy boundary condi- hoc procedure was performed for each cooking test (see Fig. 1):
tion (Eq. (2)), was considered equal to 0.9 (Townsend et al.,
1989a). Water activity at beef surface, needed to evaluate evapora- 1. A roasting pan with a known mass of water is placed under-
tive flux (Eq. (3)), was expressed according to van der Sman neath the tray with the sample, at the bottom of the oven. In
(2007a,b): this way, any amount of water expelled from the meat is col-
lected (note that the use of an empty pan would lead to rapid
0:08 evaporation of the dripped water). Previous experiments are
aw ¼ 1  ð13Þ
c done to determine the necessary quantity of water to prevent
its complete evaporation in the pan.
2. During cooking, the pan collects the dripped liquid from meat
3. Materials and methods which is weighed at the end of the cooking test.
3. In an independent experiment but under the same operating
3.1. Samples and cooking tests conditions as cooking test, only the pan with liquid water is
placed in the oven and the weight variation is registered.
Six cooking tests were performed to validate the proposed 4. The dripping loss during meat cooking can be calculated by the
mathematical model, where thermal histories, cooking time, weight difference of both pans (i.e. with and without the sam-
weight loss (with evaporative and dripping contributions), and ple being cooked).
shrinkage were determined. Half pieces of semitendinosus muscle 5. Finally, evaporative loss can be estimated by difference
(from 0.5 to 1 kg) acquired at local markets were used to perform between total weight and dripping losses.
the experiments. Prior to cooking, superficial fat was removed.
Then, the samples were packaged and stored at room or refrigera- 3.2. Water holding capacity and KW estimation
tion temperature during several hours to ensure uniform initial
temperature. Initial moisture content of samples was determined The water holding capacity (WHC) describes the ability of meat
according to AOAC official method 950.46 (AOAC, 1995). to resist the removal of liquid that could result from squeezing the
Cooking of each sample was performed in a domestic electrical beef or from gravity (Bengtsson et al., 1976). In our case, it was de-
oven (ARISTON FM87-FC, Italy), using the forced convection heat- sired to determine this capacity as a function of temperature, in the
ing mode and setting a different oven temperature for each piece absence of water evaporation. So, water holding capacity of sam-
(detailed in Table 1). Meat (at least surface and core) and oven ples was measured according to Bengtsson et al. (1976). Thin slices
temperature profiles were measured using T-type thermocouples of meat (3–4 mm thickness) were packaged into plastic pouches
(Omega, USA) connected to a data logger (Keithley-DASTC, USA). and immersed in a thermostatic bath using different combinations
Thermocouples were located before cooking, and then the sample of time (from 2.5 to 30 min) and bath temperature (from 40 to

Table 1
Sample characteristics and experimental results of the cooking tests.

Sample #1 #2 #3 #4 #5 #6
Initial weight (kg) 1.0799 0.7406 0.9718 0.6325 0.4900 0.7795
Initial water content 2.65 2.97 3.31 3.09 2.72 3.64
Initial T (°C) 20.63 13.30 7.50 12.95 7.00 15.34
To, at regime (°C) 212.89 223.50 185.41 193.80 197.26 172.83
Cooking time (min) 87.50 75.00 91.50 74.00 63.00 78.50
Weight loss (kg)
Total 0.4068 0.2276 0.2674 0.1679 0.1135 0.2081
Evaporative 0.2358 0.1154 0.1214 0.0785 0.0338 0.1042
Dripping 0.1710 0.1122 0.1460 0.0894 0.0797 0.1039
Shrinkage, initial and (final) characteristic dimensions (cm)
Height 8.2 (8.5) 7.3 (8.2) 7.7 (8.1) 6.5 (7.1) 5.7 (6.4) 6.7 (7.5)
Width 10.6 (8.0) 10.0 (8.0) 10.5 (8.0) 8.5 (7.6) 9.0 (7.5) 11.2 (9.3)
Length 17.7 (14.6) 15.7 (12.2) 17.5 (13.9) 14.0 (11.5) 13.0 (10.7) 14.3 (11.3)
Geometric modelling
No. of slices 10 10 8 8 9 8
Average thickness (cm) 15.2 12.2 16.3 14.5 11.2 14.4
4 S.M. Goñi, V.O. Salvadori / Journal of Food Engineering 100 (2010) 1–11

Initial mass of Roasting pan

water (known)

Cooking Control
With the same time /
Evaporative temperature used in cooking
water loss test
Netting tray

Evaporation of
Dripping loss water from pan

Collection of dripped liquid water from meat

Final weight: W1 Final weight: W2

Experimental dripping loss

DL: W1 – W2
Fig. 1. Schematic representation of experimental procedure used to estimate dripping and evaporative contribution to total weight loss.

100 °C). After the thermal treatment, samples were blotted for equivalent diameter, equal to the average value between the width
1 min using absorbent paper towels at room temperature. At each and height of the raw sample; (ii) thermophysical properties of air
temperature, water content was then obtained by weighing the are evaluated at an average temperature between the ambient and
sample; the final value (at 30 min) was defined as WHC. These data surface values, once the oven reached a regime condition.
were fitted to a sigmoid function (van der Sman, 2007a,b): The air velocity inside the oven was measured using a hot wire
a1 anemometer (Solomat MPM2000, United Kingdom). The velocity
WHCðTÞ ¼ ci  ð14Þ measured in the central region of the oven (v = 0.85 m s1) was
1 þ a2 expða3 ðT  T R ÞÞ
used to evaluate the Re number, which indicated that forced con-
where ci is the initial moisture content of raw sample. The unknown vection cooking mode was used. Under this condition, Eq. (15)
parameters of Eq. (14), i.e. a1, a2, a3, TR, were estimated by non-lin- was used (Perry and Green, 1997) to determine the convective heat
ear regression using the Levenberg–Marquardt method. transfer coefficient:
In addition, the experimental values of water content variation
were used to estimate the parameter KW, which is required to solve Nu ¼ 0:683Re0:466 Pr 1=3 ; 40  Re  4000 ð16Þ
the mass balance (Eq. (4)). In order to analyze the dependence of The mass transfer coefficient was then estimated by using a
KW with temperature, we assumed that the sample temperature heat-mass transfer analogy (Obuz et al., 2002). This approach, sim-
quickly reaches the water bath temperature, i.e. isothermal condi- ilar to the Chilton–Colburn’s analogy, relates both convective
tion, which is actually true for a thin slice of meat. Then, the value transfer coefficients:
of KW can be calculated by means of the analytical solution of Eq.
(4): h
kg ¼ ð17Þ
  64:7k
K W ðTÞ
cðtÞ ¼ WHCðTÞ þ ðci  WHCðTÞÞ  exp  t ð15Þ
ms
3.4. Geometric modelling

3.3. Heat and mass transfer coefficients Beef samples were considered as three-dimensional solid ob-
jects having an irregular shape for simulation purposes. In order
The convective heat transfer coefficient was estimated from to obtain the geometrical representation of each sample, the fol-
known relationships of Nusselt (Nu) vs. Reynolds (Re) and Prandtl lowing procedure was performed:
(Pr) dimensionless numbers, for the corresponding experimental
cooking conditions. For this aim, the following considerations were a. After cooking, the sample was packaged and cooled at 4 °C
done: (i) the meat sample is resembled to a finite cylinder with an during several hours; then, it was sliced along the axial axis,
 S.M. Goñi, V.O. Salvadori / Journal of Food Engineering 100 (2010) 1–11 5

and a digital image (in RGB format) of each slice was

obtained using a computer vision system, i.e. a digital cam-
era (Professional Series Network IP Camera, Intellinet Active
Networking, USA) connected to a PC.
b. Image processing was performed to obtain the irregular
boundary of slices, following several sub-steps:
1. Conversion of original RGB images to grey-scale format.
2. Noise reduction by filtering to enhance image quality
(optional).
3. Segmentation through a threshold value which was
obtained from the grey-scale image histogram. A binary
image is obtained where black color (pixel value equal
to 0) represented the background and white color the
sample (pixel value equal to 1).
4. Boundary detection and interpolation of a subset of
boundary pixels by a closed B-Spline curve (a continuous
approximation to the discrete boundary of binary
images).

The obtained B-Spline curves approximating the real bound-

aries of slices were assembled by means of a lofting technique in
COMSOL™ Multiphysics (COMSOL AB, Sweden) and MATLAB. In
this way, a closed surface is obtained, which was then transformed
in a 3D solid object. For further details about the developed proce-
dure to perform geometric modelling, the reader should be re-
ferred to Goñi et al. (2007, 2008a).
Note that the geometric models were obtained from the cooked
samples, which were actually different from the raw pieces since a
significant volume change (i.e. shrinkage) occurred during the pro-
cess. However, the geometric model of the raw sample is required
for modelling and simulation purposes. So, such model was ob-
tained by scaling the former, using three scaling factors calculated Fig. 2. (a) Experimental setup. (b) Thermocouple locations.
as the ratio between raw and cooked characteristic dimensions of
sample, i.e. length, width and height.
during cooking can be seen, while Fig. 2b shows a slice of cooked
beef, indicating the positions where the thermocouples were in-
3.5. Numerical solution
serted. In this way, Fig. 3 illustrates two representative tempera-
ture profiles obtained during cooking tests. Note that the oven
Simulation of meat cooking model (Eqs. (1)–(7)) was performed
temperature profile presents a delay (15 to 30 min) before reach-
by using the finite element method implemented in COMSOL™
ing the selected oven temperature. This is due to experimental pro-
Multiphysics. Initial uniform temperature and moisture content
cedure adopted (see Section 3.1). After this warm-up period, the
were considered for all simulations (obtained from experimental
oven automatically controls the set value with an accuracy of ±5 °C.
measurements). For each sample, the corresponding geometric
Respect to meat results, in all cases the lowest temperature cor-
model and the experimental profiles recorded during the cooking
responded to the core, and the highest to the surface. At long pro-
tests were used.
cess times, a plateau ca. 100 °C was observed for surface
A mesh consisting of ca. 6400 (in average) deformed tetrahe-
temperature. This behaviour can be correlated with a constant (or
drons was used. The solver used is an implicit variable time-step-
near constant) drying rate period, in which the surface is suffi-
ping scheme combined with Newton’s method to solve the
ciently wet and water evaporation takes place, therefore limiting
resulting non-linear equation system. Simulation time was fixed
a temperature increase. Furthermore, the evaporative front is main-
to 10 min more than experimental cooking times; solution time
tained near the surface, avoiding the formation of a crust (i.e. dehy-
was about 2–3 min using a PC with a 1.86 GHz Intel Core2Duo Pro-
drated outer layer). This observation can be verified through Fig. 2b,
cessor and 3.25 GB RAM.
where only a thin darker (mainly due to browning reactions) layer
The goodness of all fitting procedures as well as the simulated
is visualized, mostly in the upper regions. Other researchers found
temperature profiles were assessed by means of the absolute aver-
similar behaviour during meat cooking in domestic ovens (Bengts-
age relative deviations (Eq. (18)), where n is the size of data set.
son et al., 1976; Singh et al., 1984; Obuz et al., 2002). Finally, Table 1
n  
100 X v aluesim  v alueexp 
 summarizes the cooking times for all tested conditions. It is worth
AARD ¼   ð18Þ
n i¼1 v alueexp to note that actual cooking times could be slightly different from
the experimental ones due to the difficult task of exactly locating
a thermocouple at the sample core, especially for irregular shaped
4. Results and discussion materials suffering volume change during the process. The oven
temperature values shown in Table 1 correspond to the average va-
4.1. Cooking tests lue once the oven reaches the regime.

4.1.1. Temperature profiles 4.1.2. Weight loss

Fig. 2a shows a sample inside the oven with inserted thermo- The initial water content, raw weight and weight loss for all
couples, where also the pan used to measure the dripped water samples are detailed in Table 1, and Fig. 4 shows the total weight
6 S.M. Goñi, V.O. Salvadori / Journal of Food Engineering 100 (2010) 1–11

(a) 180

160
Oven temperature
140 Flux sensor
120

Temperature (ºC)
100

80

60

40

20

0
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75
Time (min)

(b) 200

180

160

140
Temperature (ºC)

120

100

80

60

40

20

0
0 5 10 15 20 25 30 35 40 45 50 55 60
Time (min)

Fig. 3. Measured temperatures: (s) core; (}) surface; (h) between core and surface. (a) Sample #6. (b) Sample #5. Continuous line is oven temperature. Dashed lines are the
model prediction for the same locations.

loss and the contribution of both evaporative and dripping mech- 4.1.3. Sample size variation
anisms (respect to raw weight). Average total weight loss was Characteristic dimensions measured before and after cooking
28.72%. Based on measured values, it can be say that dripping is are summarized in Table 1. In all cases, it was found that the final
very important in the tested operative conditions, since its contri- height was greater than the initial one, with an average size ratio
bution is of the same order of evaporative losses. In average, (i.e. cooked/raw value) of 109.11% (SD 3.83%), while the width
53.22% of the total weight loss was produced by dripping; a similar and length of the sample were reduced, with an average size ratio
value is reported by Bengtsson et al. (1976). Moreover, a tendency of 81.24% (SD 5.19%) and 80.52% (SD 2.05%), respectively. Similar
for the contribution of each mechanism to weight loss could be results were also reported by other authors (Godsalve et al.,
established respect to product size: for small samples (i.e. 0.5– 1977; van der Sman, 2007b).
0.7 kg initial weight), dripping would be the responsible for most
of weight loss, while for large samples (>1 kg initial weight) evap- 4.2. Water holding capacity and KW estimation
oration would dominate. In addition, big pieces of meat seem to
suffer more weight reduction than small ones. Further analysis Fig. 5a shows the variation of water content of meat samples
and experiments are needed to explain the contribution of differ- measured during the experiments described in Section 3.2. As it
ent mechanisms of weight loss as a function of sample size. was expected, water content diminished as temperature increased.
During the cooking tests, an interesting phenomenon was ob- The moisture value obtained at long immersion time (i.e. 30 min
served: liquid water expelled from meat by dripping remained heating) for each temperature was defined as WHC. In this way, as
‘‘adsorbed” to the surface until large droplets were formed; at this the temperature increased, the ability of meat to retain water was re-
point, water was allowed to flow down to lower regions of the beef duced, which is due to thermal protein denaturation. Subsequently,
and finally to oven bottom (where the pan with water was placed; the experimental values of WHC (as a function of temperature) were
see Section 3.1). This behaviour surely depends on the gravitational fitted to Eq. (14). Firstly, the parameter a1 was set to (ci – 0.961) kg
force and the surface tension that the droplet must support and water/kg dry solid, since a final moisture content or WHC of 0.961
overcome, as well as the viscosity of the dripped fluid. In conse- (in average) was obtained for high temperature (100 °C) indepen-
quence, a fraction of the dripped water was evaporated at the sur- dently of initial water content of samples. Then, the following values
face, so the obtained experimental values for dripping contribution for other parameters of Eq. (14) were found by regression:
are probably lower than the actual ones. a2 = 3.2674; a3 = 9.0027  102 °C1; TR = 48.27 °C. Fig. 5b shows
 S.M. Goñi, V.O. Salvadori / Journal of Food Engineering 100 (2010) 1–11 7

40

35

30

Weight loss (%)

25

20

15

10

0
0.45 0.55 0.65 0.75 0.85 0.95 1.05
Initial weight (kg)

Fig. 4. Total percentage weight loss () and the contribution of both evaporative (h) and dripping (4) mechanism (respect to raw weight).

(a) 3.5
3.0
Moisture content (dry basis)

2.5

2.0

1.5

1.0

0.5
0 3 6 9 12 15 18 21 24 27 30
Time (min)

(b) 3.5 6.0E-05

3.0 5.0E-05
KW (kg dry solid s )
-1

2.5 4.0E-05
WHC (dry basis)

2.0 3.0E-05

1.5 2.0E-05

1.0 1.0E-05

0.5 0.0E+00
10 20 30 40 50 60 70 80 90 100
Temperature (ºC)

Fig. 5. (a) Experimental (symbols) water content variation and fit of Eq. (15) (lines) at different water bath temperatures: (e) 40 °C; () 50 °C; (h) 60 °C; (j) 70 °C; (4) 80 °C;
(N) 90 °C; (s) 100 °C. (b) Values of water holding capacity (}) and fit of Eq. (14) (continuous line); KW parameter (N) and fit of Eq. (19) (dashed line).

the experimental WHC data and the corresponding fitted sigmoid we analyzed the dependence of KW with temperature (Fig. 5b).
expression, which have an AARD of 4.10%. With the aim of using this parameter in the mathematical model
Data of water content as a function of time were used to esti- for roasting, the following expression was proposed to describe
mate the parameter KW for each tested temperature, through Eq. its dependence with temperature:
(15). The performance of parameter estimation is shown in
Fig. 5a, besides the AARD was found to be 3.18%. Furthermore,
K W ðTÞ ¼ 4:669  106 expð4:495  102 ðT  40ÞÞ ð19Þ
8 S.M. Goñi, V.O. Salvadori / Journal of Food Engineering 100 (2010) 1–11

Note that Eq. (19) was obtained by neglecting the data corre- cedure demonstrated its ability to accurately reproduce the shape
sponding to 100 °C, since the KW value for this temperature should of the samples.
be at least equal to the one at 90 °C. We attribute these results to
estimation of KW by using all data of water content as a function
4.4. Numerical simulation of cooking
of time. At high temperatures (80–100 °C), the equilibrium condi-
tion (WHC) was achieved rapidly, i.e. all variation was observed be-
Numerical simulation of meat cooking was performed by using
fore 10 min (Fig. 5a). In this way, the sensitivity of Eq. (15) is low
the finite element method. Each experimental condition was repre-
for long times and high uncertainty is related to estimation of KW
sented by the corresponding values of transfer coefficients, being
at high temperature.
the average values of heat and mass transfer coefficients
9.676 W m2 °C1 (SD 0.456) and 6.502  108 kg Pa1 m2 s1
4.3. Geometric modelling (SD 3.669  109), respectively. The goodness of the developed
model was evaluated by comparing the experimental and simu-
Fig. 6 illustrates the developed geometric modelling procedure lated thermal histories at surface and core (coldest point), the
for a representative sample. As can be seen, the obtained geometric cooking time, and the weight loss of the samples.
model is in good agreement with the shape of the actual raw sam- Fig. 3 shows the evolution of both experimental and simulated
ple. Table 1 details the number of slices and the average slice thick- temperature profiles at three internal positions for two different
ness used for each beef sample. Fig. 7 shows the meshed geometric beef samples. As can be seen, the model well describes the tem-
models for the six samples used to validate the developed mathe- perature variation in all positions. The values of AARD for core
matical model. and surface temperature profiles as well as experimental and pre-
In order to perform an objective evaluation of this procedure, dicted cooking times are detailed in Table 2. In general terms,
the volume of the cooked samples, measured by liquid displace- simulated core temperature is slightly higher than experimental
ment method, was compared with the volume calculated from one, producing a small under–prediction of the process time,
the constructed geometric models (before scaling), according to equivalent to 3.17 min, in average; the AARD for this parameter
Goñi et al. (2007). The AARD between both values was 3.90% was 3.91%. Fig. 8 shows the simulated temperature and water
(equivalent to 18.26 cm3). Additionally, the density of the samples content profiles for a whole sample (#3), at a specific time
was determined by using the measured initial weight and the esti- (40 min).
mated volume of raw sample (from geometric models after scal- Regarding the weight loss, Table 3 shows the measured and pre-
ing); an average density of 1064.3 kg m3 was calculated. This dicted values for all beef samples. The predicted total weight loss,
value agrees with the one calculated from both experimental obtained from Eqs. (5) and (6), presented an AARD equal to 7.96%,
weight and volume, i.e. 1065 kg m3. Therefore, the presented pro- equivalent to 0.0152 kg. This result is a good indicator that the

Fig. 6. (a) Original images used to construct the geometric model. (b) Result of applying the procedure to approximate the shape of each slice. (c) Image of whole raw sample
and (d) constructed geometric model (which is already scaled to raw sample dimensions and meshed).
 S.M. Goñi, V.O. Salvadori / Journal of Food Engineering 100 (2010) 1–11 9

Fig. 7. Meshed geometric models for the six samples used in experimental determinations and model simulations. The geometric models are already scaled to raw
dimensions.

proposed model is accurately representing the water loss occurring Respect to individual contributions, AARD for dripping and
during the used cooking conditions. evaporative losses were equal to 21.71% and 26.44%, respectively.
Furthermore, simulated dripping contribution to total weight loss
was 55.21% (average value), a similar value to the experimental
one reported in Section 4.1.2. Respect to these high AARD values,
Table 2 it is important to remark at this instance that the proposed cooking
Measured and predicted cooking times and temperature profiles error.
model allows the prediction of each contribution to total weight
Sample Cooking time (min) AARD (%) loss, representing a significant advance in beef cooking modelling.
Measured Predicted Core Surface At this instance, it is important to mention that the parameter
#1 87.50 82.50 2.93 1.73
KW, required to solve the proposed mathematical model, can be
#2 75.00 71.50 3.90 5.39 determined by means of a simple experimental procedure, i.e. it
#3 91.50 87.50 3.10 5.98 does not imply its estimation through the solution of an inverse
#4 74.00 70.50 3.84 2.06 problem.
#5 63.00 62.50 3.63 5.32
Finally, it is worth to note that the use of high realistic represen-
#6 78.50 76.00 2.74 1.78
tation of the sample reduces the geometry influences on the
10 S.M. Goñi, V.O. Salvadori / Journal of Food Engineering 100 (2010) 1–11

Fig. 8. Simulated profiles of (a) temperature (°C) and (b) water content (wet basis), at 40 min cooking for the sample #3.

Table 3
the computational cost of simulations. Finally, the developed mod-
Measured and predicted weight loss with evaporative and dripping contributions. el appears as a valuable tool to optimize and control the oven cook-
ing (or roasting) of beef.
Sample Measured weight loss (kg) Predicted weight loss (kg)
Total Evaporative Dripping Total Evaporative Dripping
#1 0.4068 0.2358 0.1710 0.3963 0.1928 0.2035 Acknowledgments
#2 0.2276 0.1154 0.1122 0.2531 0.1257 0.1274
#3 0.2674 0.1214 0.1460 0.2610 0.1191 0.1419
We thank Dr. Emmanuel Purlis for helpful comments during the
#4 0.1679 0.0785 0.0894 0.1915 0.0783 0.1132
#5 0.1135 0.0338 0.0797 0.1270 0.0663 0.0607 development of this work. Authors acknowledge Consejo Nacional
#6 0.2081 0.1042 0.1039 0.2197 0.0696 0.1501 de Investigaciones Científicas y Técnicas (CONICET), Agencia Nac-
ional de Promoción Científica y Tecnológica (ANPCyT 2007-
01090), and Universidad Nacional de La Plata (UNLP, 11I140) from
Argentina for their financial support.
simulated results, and we can focus on the error associated with
model assumptions and development.
References
5. Conclusions American Meat Science Association, 1995. Research Guidelines for Cookery, Sensory
Evaluation and Instrumental Tenderness Measurements of Fresh Meat. National
In this work, a simultaneous heat and mass transfer model to Live Stock and Meat Board, Chicago.
AOAC, 1995. In: Helrich, K. (Ed.), Official Methods of Analysis of the Association of
simulate meat oven cooking is proposed and accordingly validated. Official Analytical Chemists, fifth ed. AOAC, Washington, DC.
The model describes internal heat transfer by Fourier’s law and the Bengtsson, N.E., Jakobsson, B., Dagerskog, M., 1976. Cooking of beef by oven
concept of water demand (related to the water holding capacity of roasting: a study of heat and mass transfer. Journal of Food Science 41 (5),
1047–1053.
meat) is used to describe the inner water content variation during Bottani, E., Volpi, A., 2009. An analytical model for cooking automation in industrial
the process. Bearing in mind the ultimate objective of our work, i.e. steam ovens. Journal of Food Engineering 90 (2), 153–160.
multi-objective optimization of roasting, the model is focused on Burfoot, D., Self, K.P., 1989. Predicting the heating times of beef joints. Journal of
Food Engineering 9 (4), 251–274.
predicting cooking time and weight loss. In this way, two mecha- Chang, H.C., Carpenter, J.A., Toledo, R.T., 1998. Modeling heat transfer during oven
nisms of water loss are incorporated: evaporation and dripping. roasting of unstuffed turkeys. Journal of Food Science 63 (2), 257–261.
Through experimental data, it is demonstrated that the approach Chen, H., Marks, B.P., Murphy, R.Y., 1999. Modeling coupled heat and mass transfer
for convection cooking of chicken patties. Journal of Food Engineering 42 (3),
adopted well described such a complex process, including both
139–146.
temperature and weight loss variations. Choi, Y., Okos, M.R., 1986. Effects of temperature and composition on the thermal
In order to reduce errors associated to experimental measure- properties of foods. In: Le Maguer, M., Jelen, P. (Eds.), Food Engineering and
ment of variables such as temperature and water content, realistic Process Applications: Transport Phenomena. Elsevier Applied Science,
Amsterdam.
geometric models are used. It is shown that this methodology COMSOL AB. COMSOL Multiphysics User’s Guide. Version: September 2005,
helps to improve the model predictions as well as not increasing COMSOL 3.2.
 S.M. Goñi, V.O. Salvadori / Journal of Food Engineering 100 (2010) 1–11 11

Godsalve, E.W., Davis, E.A., Gordon, J., Davis, H.T., 1977. Water loss rates and Ou, D., Mittal, G.S., 2006. Double-sided pan frying of unfrozen/frozen hamburgers
temperature profiles of dry cooked bovine muscle. Journal of Food Science 42 for microbial safety using modelling and simulation. Food Research
(4), 1038–1045. International 39, 33–45.
Goñi, S.M., Purlis, E., Salvadori, V.O., 2007. Three-dimensional reconstruction of Ou, D., Mittal, G.S., 2007. Single-sided pan frying of frozen hamburgers with
irregular foodstuffs. Journal of Food Engineering 82 (4), 536–547. flippings for microbial safety using modeling and simulation. Journal of Food
Goñi, S.M., Purlis, E., Salvadori, V.O., 2008a. Geometry modelling of food materials Engineering 80 (1), 33–45.
from magnetic resonance imaging. Journal of Food Engineering 88 (4), 561–567. Pan, Z., Singh, R.P., Rumsey, T.R., 2000. Predictive modeling of contact-heating for
Goñi, S.M., Purlis, E., Salvadori, V.O., 2008b. Evaluating a simple model for meat cooking a hamburger patty. Journal of Food Engineering 46 (1), 9–19.
cooking simulation. In: Proceedings of CIGR 2008 – International Conference of Perry, R.H., Green, D.W., 1997. Perry’s Chemical Engineers’ Handbook, seventh ed.
Agricultural Engineering – XXXVII Congresso Brasileiro de Engenharia Agrícola, McGraw-Hill, New York.
Iguassu Falls City, Brazil. Singh, N., Akins, R.G., Erickson, L.E., 1984. Modeling heat and mass transfer
Huang, E., Mittal, G.S., 1995. Meatball cooking – modeling and simulation. Journal of during the oven roasting of meat. Journal of Food Process Engineering 7 (3),
Food Engineering 24 (1), 87–100. 205–220.
López Osornio, M.M., Hough, G., Salvador, A., Chambers IV, E., McGraw, S., Fiszman, Tornberg, E., 2005. Effects of heat on meat proteins – implications on
S., 2008. Beef’s optimum internal cooking temperature as seen by consumers structure and quality of meat products. Journal of Food Engineering 70 (3),
from different countries using survival analysis statistics. Food Quality and 493–508.
Preference 19, 12–20. Townsend, M.A., Gupta, S., Pitts, W.H., 1989a. The roast: nonlinear modeling and
McDonald, K., Sun, D.-W., Kenny, T., 2001. The effect of injection level on the quality simulation. Journal of Food Process Engineering 11 (1), 17–42.
of a rapid vacuum cooled cooked beef product. Journal of Food Engineering 47 Townsend, M.A., Gupta, S., Pitts, W.H., 1989b. Optimal roasting. Journal of Food
(2), 139–147. Process Engineering 11 (2), 117–145.
Obuz, E., Powell, T.H., Dikeman, M.E., 2002. Simulation of cooking cylindrical beef van der Sman, R.G.M., 2007a. Moisture transport during cooking of meat: an
roasts. Lebensmittel-Wissenschaft und-Technologie 35 (8), 637–644. analysis based on Flory–Rehner theory. Meat Science 76 (4), 730–738.
Obuz, E., Dikeman, M.E., Erickson, L.E., Hunt, M.C., Herald, T.J., 2004. Predicting van der Sman, R.G.M., 2007b. Soft condensed matter perspective on moisture
temperature profiles to determine degree of doneness for beef biceps femoris transport in cooking meat. American Institute of Chemical Engineers 53 (11),
and longissimus lumborum steaks. Meat Science 67 (1), 101–105. 2986–2995.