Meat Science 95 (2013) 336–344

Contents lists available at SciVerse ScienceDirect

Meat Science
journal homepage: www.elsevier.com/locate/meatsci

Combined heat transfer and kinetic models to predict cooking loss

during heat treatment of beef meat
Alain Kondjoyan a,⁎, Samuel Oillic a, b, Stéphane Portanguen a, Jean-Bernard Gros c
a
INRA, UR370 Qualité des Produits Animaux, F-63122 St Genès Champanelle, France
b
ADIV, 10 rue Jacqueline Auriol, ZAC des Gravanches, F-63039 Clermont-Ferrand Cedex 2, France
c
Laboratoire de Génie Chimique et Biochimique, Université Blaise Pascal, Polytech, Clermont-Ferrand, 24 avenue des Landais, BP 206, F-63174 Aubière Cedex, France

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

Article history: A heat transfer model was used to simulate the temperature in 3 dimensions inside the meat. This model was com-
Received 28 September 2012 bined with a first-order kinetic models to predict cooking losses. Identification of the parameters of the kinetic
Received in revised form 23 April 2013 models and first validations were performed in a water bath. Afterwards, the performance of the combined
Accepted 26 April 2013
model was determined in a fan-assisted oven under different air/steam conditions. Accurate knowledge of the
heat transfer coefficient values and consideration of the retraction of the meat pieces are needed for the prediction
Keywords:
Model
of meat temperature. This is important since the temperature at the center of the product is often used to deter-
Cooking mine the cooking time. The combined model was also able to predict cooking losses from meat pieces of different
Meat sizes and subjected to different air/steam conditions. It was found that under the studied conditions, most of the
Heating water loss comes from the juice expelled by protein denaturation and contraction and not from evaporation.
Transfer © 2013 Elsevier Ltd. All rights reserved.
Juice

1. Introduction
During cooking, meat can lose a large quantity of its mass in the
form of meat juice. Water loss determines the technological yield of
the cooking operation, making it a critical factor in the industry.
Cooking time and losses also affect the quality of the cooked
meat: color, savor, juiciness, tenderness, micronutrient content,
etc. (Modzelewska-Kapitula, Dabrowska, Jankowska, Kwiatkowska, &
Cierach, 2012). Water debinding and migration in meat during cooking
are related to the denaturation and contraction of protein structures
caused by increasing temperature (Laroche, 1978; Lepetit, 2007;
Lepetit, Grajales, & Favier, 2000; Palka & Daun, 1999; Tornberg, 2005).
Up to 80% of the water can be lost during pan frying of beef burgers
(Oroszvari, Bayod, Sjöholm, & Tornberg, 2006). Van der Sman (2007)
modeled water transport in meat pieces during cooking by using
Flory–Rehner theory and Darcy law. This approach was validated
using a “rectangular roast” cooked in an air oven at two air temperatures,
175 and 225 °C. More recently, Dhall, Halder, and Datta (2012) have
developed a multiphase model to predict the transport of water and fat
during meat cooking. However, the effect of collagen contraction on
fluid flow is not taken into account in this model, despite the fact
that it leads to meat shrinkage and also dictates water transport in
non-ground meat (Bouhrara, Clerjon, Damez, Kondjoyan, & Bonny,
⁎ Corresponding author. Tel.: +33 473624492; fax: +33 473624089.
E-mail address: alain.kondjoyan@clermont.inra.fr (A. Kondjoyan).
2011, 2012). Dhall et al.'s (2012) modeling approach therefore essential-
ly remains dedicated to cooking hamburger patties. Extending the work
of Van der Sman (2007), Feyissa, Gernaey, and Adler-Nissen (2013)
have inserted swelling pressure and elastic modulus into the Darcy
law to model the effect of protein contraction on the water transport
inside roast meat. A mathematical relation was proposed to take into
account the effect of temperature on the elastic modulus, and the
parameters of this relation were fitted on the experimental data of
Tornberg (2005). This approach assumes that whole meat is a uniform
porous material, which permeability does not vary during heating. Per-
meability values for whole meat are unknown and Feyissa et al. (2013)
use reported data for ground meat and emphasized the need for more
quantitative knowledge of the effect of temperature on meat perme-
ability. Moreover, the assumption of juice circulating in a uniform
porous material is disputable for whole meat since it has been observed
that the juice expelled from the myofibers by heat denaturing and
contracting circulates in channels of different dimensions formed by
the shrinkage of the complex perimysium, endomysium and myofiber
bundle network (Bouhrara et al., 2012). Other combined heat transfer
and kinetics approaches are used to model the cooking of whole meat
(Goni & Salvadori, 2010). These approaches have recently been used
for a multi-objective optimization of beef roasting (Goni & Salvadori,
2012). The rate constant used in the kinetic models was explicitly
only dependent on meat temperature and was applied on both meat
slices and semitendinosus muscles, but an implicit variation of the rate
constant was also integrated into the model by dividing this rate

0309-1740/$ – see front matter © 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.meatsci.2013.04.061
 A. Kondjoyan et al. / Meat Science 95 (2013) 336–344 337

step was performed on rectangular cuboid (parallelepipeds with perpen-

List of symbols used dicular faces) samples heated in a water bath. These experimental results
have already been published in Oillic et al. (2011). In the second step,
A, B: parameters used in relation (3) model validation was performed in a fan-assisted oven. Experiments
Cp thermal capacity of the meat (J kg K −1) were performed on meat samples derived from one muscle type
[DM] percentage of dry matter (Semimembranosus muscle) whose size varied from thin steaks up to
d distance to the nearest meat surface (m) big muscle cuts. Different air/steam conditions were applied to analyze
Ea activation energy (J mol −1) the transition from wet air to dry air microenvironment, and a handful
h heat transfer coefficient (W m −2 K −1) of experiments were performed on muscles other than SM. All these
k and k0 rate constant of the kinetic models (s −1) results were analyzed to identify how far the combined heat transfer
M mass of the sample (kg) and kinetic models can be practicably used to predict and control the
R molar gas constant = 8.314 J mol −1 K −1 evolution of juice loss and meat quality during cooking.
T temperature, (°C) or (K)
t time, (s) or (min) 2. Modeling beef cooking
V volume of the sample (m −3)
X water content (kg of water per kg of dry matter) The first phase of the modeling process was to calculate local
λ thermal conductivity of the meat (W m −1 K −1) 3-dimensional temperature kinetics inside the sample. In-product
ρ density of the meat (kg m −3) heat transfer was assumed to be purely conductive (Oillic et al., 2011).
Energy exchanges at the boundaries were calculated classically by a
Newtonian law in which the difference between water bath or oven
Subscripts temperature and sample surface temperature was multiplied by the
raw sample thawed and uncooked convective transfer coefficient. Under these assumptions, the accuracy
cooked sample after cooking of the simulated temperatures was essentially dependent on how accu-
DM dry matter measured on the sample dried for 48 h at rately sample dimensions are known and on the accuracy of the param-
104 °C eter values that were introduced into the model. These parameters
0 value of the raw sample were thermal diffusivity of the sample (DT), and heat transfer coeffi-
eq value obtained at equilibrium cient value at the solid/fluid interface (h). Thermal diffusivity was calcu-
CL exp Experimental cooking loss lated from λ/ρCP using a density value of 1060 kg m−3 and the thermal
capacity of the meat, equal to 3200 J kg K −1 (Oillic et al., 2011; Tsai,
Unklesbay, Unklesbay, & Clarke, 1998). Thermal conductivity was
Muscle types 0.45 W m −1 K−1 and considered constant during the cooking process
IS Infraspinatus (Baghe-Khandan & Okos, 1981; Oillic et al., 2011). During water-bath
LT Longissimus thoracis heating or cooling treatments, heat transfer at the surface of the meat
MA Masseter was purely convective. In-oven energy exchanges by radiation were
SM Semimembranosus neglected, since the oven walls were made of polished stainless steel
ST Semitendinosus that has an emissivity of less than 0.1 (Kondjoyan, 2006). Evaporation
during dry-air treatment at 95 °C or sample cooling was taken into ac-
count using an apparent transfer coefficient value (Kondjoyan, 2006).
Heat transfer coefficients due to free or forced convection were mea-
constant value by the dry mass of the sample. It is important that the sured experimentally using aluminum objects of different shapes and
rate constant of the kinetic models does not solely depend on tempera- sizes and the method described by Ghisalberti and Kondjoyan (1999),
ture since it has been observed experimentally that there is a different and the effect of evaporation on heat transfer coefficient during oven
timescale between heat transfer and mass transfer in meat samples, treatments in air were determined from the author's previous experi-
this timescale being connected to the duration needed for the juice to mental studies (Kondjoyan, 2006; Kondjoyan et al., 2006a,b).
migrate from inside the meat sample to its surface (Oillic, Lemoine, Cooking losses were calculated from the time-course of water con-
Gros, & Kondjoyan, 2011). Van der Sman's and Goni and Salvadori's tent in the sample. This evolution was described using a first-order
validations were performed directly on roast meat of given size and kinetic models based on local water concentration X, with Xeq(T)
shape derived from one type of muscle and subjected to oven- being water concentration at equilibrium. Equilibrium was reached
cooking air conditions. This falls short, since a real determination of when loss was no longer observed whatever the duration of additional
model performance requires a wide range of sample sizes and mus- treatment. Xeq(T) was connected to the maximal protein denaturation–
cle types. Moreover, oven-cooking in dry air is not the best setting contraction and myofiber water debinding able to occur at a given
for a first test of model performance, since: (i) it is a complex situa- temperature.
tion where uncertainties on heat transfer “mix” with uncertainties
dX  
due to the mass transfer model and the phenomena driving crust for- ¼ −kðT; dÞ⋅ X−X eq ðT Þ ð1Þ
mation, and (ii) air-cooking makes it difficult to effectively separate dt
water loss by evaporation from water loss by protein denaturation–
contraction. The dependence of the reaction rate on temperature was given by
The work reported here is based on a combined heat transfer and an Arrhenius Eq. (2), with local temperature calculated using the heat
kinetics modeling approach to predict weight losses during the cooking transfer model described previously.
of beef meat. The rate constant of the kinetic models is calculated locally
kðT; dÞ ¼ k0 ðdÞ eð R⋅T Þ
−Ea

both from the meat temperature and from the distance between the
calculation point and the nearest meat surface. Introducing this distance
in the expression of the rate constant explicitly takes into account the
effect of sample thickness on mass transfer. The parameters of the
kinetic models were identified using a set of experiments performed
on meat cubes of different sizes heated in a water bath. A first validation
ð2Þ
The rate constant of the kinetic models was also rendered dependent
on distance between the calculation point and the meat surface. Prelim-
inary experiments proved that the dependency between k and product
thickness was a power law. Thus, a power law was retained to describe
338 A. Kondjoyan et al. / Meat Science 95 (2013) 336–344

the local dependency between k and the local distance to the nearest The combined heat transfer and kinetic models was implemented in
surface: COMSOL Multiphysics 3.4 which solves systems of nonlinear differential
equations by the finite element method. The boundary distance function
−B
k0 ¼ A⋅d ð3Þ of COMSOL Multiphysics 3.4, which determines the distance to the closest
boundary available on subdomains, was used to calculate k from d. The
with A and B being parameters whose values need to be determined. rectangular cuboid geometry of the sample was meshed using between
For very short distances to the surface, the power law (Eq. (3)) leads 3600 and 170,000 tetrahedrons depending on the sample's dimensions.
to an infinite k0 value when d tends to zero. This is not possible because The initial moisture content in the meat was always considered to be uni-
there are always some phenomena that slow down the expulsion of form. During water bath simulations, initial meat temperature was also
water, even at the meat surface (even when d = 0). Thus, the power assumed to be uniform, and external temperature was that of the bath.
law was only used for distances d greater than 2 mm. For shorter dis- Cooking–cooling durations were either decided before the simulation or
tances, the change in k0 was assumed to be linear with d, the slope of obtained by stopping the heating period as soon as the target temperature
the linear relation being the derivative of the power law function at at the center of the product had been reached. The solver used a conjugate
d = 2 mm. The variation of water content at the surface was determined gradient method and an algebraic multigrid preconditioning to solve the
by the relation (1) using the temperature simulated at the surface and equations. The solution took from 2 to 20 min depending on processing
the rate constant value calculated at d = 0. conditions, using a PC with a 2.33 GHz Intel Core2Duo Processor and
The water contents in meat at equilibrium (Xeq, dry basis) measured 32 Gb RAM. The values of parameters Ea, A and B in relations (2–3)
by Oillic et al. (2011) on different muscles and at different water bath were determined by minimizing the sum of the square difference
temperatures were used for numerical calculations. These average between the weight loss predicted by the model and that measured
values, which were calculated from at least 21 measurements at each on meat cubes at all temperature levels.
temperature in the case of SM, and 6 values for other muscles, were
comparable to those determined in the literature (Fig. 1). k(T) was con- 3. Cooking experiments
sidered isotropic in space, as previous water bath results have proven
that cooking losses are only very slightly dependent on meat fiber direc- The muscles used were from 1 to 4-year-old Charolais-breed cows,
tion (Oillic et al., 2011). During the cooling phase, meat temperature aged and then frozen under the same conditions, and prepared following
was simulated using the thermal model and the cooking loss went on the same procedure as in Oillic et al. (2011). The number of samples per
until X = Xeq, then it was stopped. muscle depends of their size, for example, only a single 70 × 70 ×
The integration of relation (1) from the initial time to time tf gives 280 mm sample could be cut in a SM muscle while thirty 10 ×
the local water concentration at this time, providing knowledge of 10 × 10 mm samples could be obtained in the same muscle. Experiment
initial water content X0: was repeated 3 times for each sample dimension and cooking condition.
Thirteen animals were used one muscle being cut per animal. Oven
tf   experiments were performed on rectangular cuboids (15 × 70 × 70;
X tf ¼ X 0 þ ∫ kðT Þ⋅ X eq ðT Þ−X t dt ð4Þ 60 × 60 × 110 and 110 × 110 × 110 mm) to ensure accurate position-
t¼0 ing of temperature sensors at different locations in the meat and to mea-
sure the dimensions of the cooked meat after retraction. Just before the
The percentage of cooking losses was calculated from: experiment, the frozen samples were slowly heated up to −2 °C. At
this temperature, the muscle are both, not too hard, but enough rigid,
X 0 −X tf to be cut very precisely. Thus, meat rectangular cuboids with flat surfaces
CLtf ¼ 100 ð5Þ
1 þ X0 can be obtained with accuracy on sample dimensions of ±1.0 mm. Ther-
mocouples were placed in the center of samples and, for roasts, at a dis-
tance of 5 and 27 mm from the surface (Fig. 2). To precisely position
these thermocouples, a 3.5 mm hole was drilled into the still-hard
X (kg water/kg DM) meat rectangular cuboids perpendicularly to the surface using a guided
3,5 drill. The thermocouple was then pushed into this hole (Fig. 2). The
accuracy on the initial thermocouple position was ±1 mm. The samples
3 were then wrapped and placed in a chamber at 10 °C for 60 to 120 min
to reach uniform temperature in the meat before cooking. Then, samples
2,5 were weighed (Mraw), placed in the center of an oven (Emeraude
III + model 635 Prestige Juste Température, HMI Thirode, Mitry-Mory,
France), then cooked at a control temperature and at different levels of
2 relative humidity corresponding to the following setting positions:
(1) no steam injection, (2) “10% steam injection”, (3) saturated steam.
1,5 Goni & Salvadori Air relative humidity measured at a 95 °C oven temperature using a
hygrometer (HTM 337 transmitter, VAISALA, Helsinki, Finland) during
the long period was (1) less than 1% (accuracy of the hygrometer at
1
SM ST LT the temperature of measurement), (2) 57% ± 10%, and (3) 80% ± 15%,
respectively, for the 3 setting positions. Preliminary measurements
0,5 proved that air temperature and velocity inside the oven were spatially
IS MA
Temperature (°C) uniform at sample locations. Cooking was stopped when the meat center
0 temperature reached a target value of 50–55 °C. Before being weighed
0 20 40 60 80 100 (Mraw), the samples were then cooked either in air at room temperature
(20 °C) or in an ice-water bath until meat temperature reached 5–10 °C.
Fig. 1. Variation of water content in the meat at equilibrium (longest cooking times), Xeq, at The model assumed evaporation during air cooling only affected juice
different water-bath temperatures. Black squares signify the value measured on ST muscle
by Goni and Salvadori (2010). Other symbols represent the measurements reported by
already expelled from the meat by protein denaturation–contraction.
Oillic et al. (2011) on IS, LT, MA, SM and ST muscles. The dashed line is the sigmoid function This assumption was checked by cooling vacuum-packed and non-
used by Goni and Salvadori (2010) to fit their experimental results. vacuum-packed meat previously heated in a water-bath. Packed and
 A. Kondjoyan et al. / Meat Science 95 (2013) 336–344 339

DRILLING

60

30 30 30

60
5
55

110
Fig. 2. Schematic representation of the positioning of the thermocouples in the 60 × 60 × 110 mm meat rectangular cuboids using the special procedure described in the paper and
a guided drill.

unpacked meat yielded the same cooking losses, thus verifying the
assumption. All subsequent samples were then air-cooled unwrapped.
The exact dimensions of each sample were measured before and after
cooking to determine heat-induced meat contraction. The experiment
was repeated three times for each muscle dimension and cooking condi-
tion. The experimental percentage of cooking loss was calculated directly
from the mass of the raw and the cooled sample:
Mraw −Mcooled
CL exp ¼ 100 : ð6Þ
M raw
During cooking, the mass of dry matter in the sample decreases as
part of it gets expelled into the juice. This dry matter loss is especially
important when samples are directly immersed in water. To take into
account this phenomenon and be able to compare the experimental
(6) and calculated (5) cooking losses, X values were determined on
the same dry matter basis, which was the dry matter measured on the
cooked–cooled sample MDM:
M sample −M DM
X¼
MDM
where Msample is the mass of sample in its raw state, in its cooked–
cooled state, or in its equilibrium state, to calculate X0, Xtf and Xeq
respectively.
To measure dry matter content, thin slices of meat were cut in the
cooked sample and placed in a calibrated cup. The cup-plus-sample
was weighed and then oven-dried for 48 h at 105 °C. After the 48-h dry-
ing period, the samples were taken out of the oven and re-weighed to
determine the dry matter mass (Oillic et al., 2011).
oven ranged from 3.8 to 5.0 m s −1, and the heat transfer coefficient
values were determined using the relations of Ghisalberti and
Kondjoyan (1999) for a flow turbulence intensity of 20–30%. The h
value of 51 Wm−2 K −1 measured in the oven under dry air conditions
was in the range of values (40.5 to 55.8 Wm −2 K −1) calculated from
these relations on bodies of comparable size. The h value measured
under “10% steam injection” (h = 134 Wm−2 K −1) was in between
this “dry air” value and the value determined by Kondjoyan and
Portanguen (2008) on a PTFE® sample of similar dimensions (h =
200–250 Wm−2 K −1) subjected to superheated steam jet. Condensa-
tion and evaporation taking place at the sample surfaces create phase
changes that vary the amount of energy available to heat the product,
and so the convective heat transfer coefficient value shall be replaced
by an effective transfer coefficient value for determining the boundary
condition at the meat surface (Kondjoyan, 2006). When the meat sam-
ple was subjected to mixed air/steam condition (“10% steam injection”),
relative humidity in the oven was 57%, which led to a dew point tem-
perature of 86 °C. Thus, during the first period of heating, water vapor
condensed at the surface of the meat sample and then evaporated as
soon as the product surface reached 86 °C. These phenomena also
ð7Þ
occurred on aluminum samples, and the h measure can be considered
an effective coefficient value that is directly applicable on meat samples.
During dry air treatment, the dew point temperature in the oven was
about 10 °C. Water evaporated throughout treatment on the meat
product that contained water but not on the aluminum sample which
was dry. Thus, the measure of h, which only accounted for convection,
Table 1
Heat transfer coefficient values measured in the water baths and in the oven under the
different treatment conditions applied in this study.

4. Results Cooking or cooling systems Conditions h (Wm−2 K−1)

Water bath Stirred 95 °C 2174 ± 174

4.1. Heat transfer Non-stirred 50 °C 821 ± 90
Non-stirred 70 °C 962 ± 98
Non-stirred 90 °C 1190 ± 101
4.1.1. Heat transfer coefficient value
Non-stirred ice-water ≈0 °C 374 ± 10
The convective heat transfer coefficients measured on aluminum Air cooling, free Ambient/room temperature 20 °C 6.5 ± 1
samples are coherent with literature measurements (Table 1). Santos, convection
Zaritzky, and Califano (2008) obtained an h value of 1615 Wm−2 K−1 Fan-assisted oven Dry air 100 °C 51 ± 5
using a stirred water bath to heat aluminum spheres and cylinders of 10% steam injection 100 °C 134 ± 13
steam 100 °C 291 ± 29
comparable dimensions to our samples. Air velocity in the fan-assisted
340
shall be decreased to take into account the effect of water evaporation
on energy exchanges at the meat surface. This decrease depends on
variation in water activity at the meat surface during heating. According
to the literature, water activity at the meat surface swiftly plummets
down to 0.10–0.15 (Kondjoyan et al., 2006a,b), leading to an effective
heat transfer coefficient value in the range 20 Wm −2 K−1 to
40 Wm−2 K −1 for an h value of 51 Wm−2 K −1. Surface evaporation
also occurred throughout air cooling, but it increased the convective
heat transfer coefficient value about 4-fold at the beginning of cooling
down to about 1.8-fold at the end. Our calculations for the entire treat-
ment thus used an average 2.5-fold increase in h value (Kondjoyan,
2006).
4.1.2. Comparison of simulated versus measured meat temperatures
The fact that we used rectangular cuboids samples and this experi-
mental procedure made it possible to accurately determine initial ther-
mocouple locations and product dimensions, and therefore reliably
compare simulated against measured temperatures, which is always
difficult in meat due to the soft and heat-deformable characteristic of
the material. The result of this comparison is illustrated here in the
case of the 60 × 60 × 100 mm roast oven-cooked under the “10%
steam injection” condition for the two extreme thermocouple posi-
tions: at the center of the product and 5 mm from its surface (Fig. 3a
and b). The standard deviations on the temperature measurements dur-
ing the heating period are on the average: ±3.4 °C at 5 mm from the
surface; and ±0.6 °C at the center of the sample. However, greater var-
iation can be noticed at specific cooking times. Model simulations can
prove that the observed variations are the result of meat contraction
which leads to thermocouple movement, and to juice expelling. During
the experiments, heat-shrinkage can lead to a decrease of 5 mm of the
lateral dimensions of the square section of the 60 × 60 × 110 mm
roast (55 × 55 mm section instead of the 60 × 60 mm initial dimen-
sions). Such a decrease in sample dimensions leads to a 3 °C difference
of the simulated temperature at the center of the product after 24 min
of heat treatment (Fig. 3a). This explains why the temperature mea-
sured at the core of samples during cooking can turn out different
even when the initial sample dimensions and thermocouple locations
at the start of the experiment are exactly the same, as heat-induced
meat contraction, which is not isotropic, does not occur in the exactly
same way in different samples. This also explains why cooking time
can be significantly different during accurately-controlled repetitions
when it is based on temperature measured at the “core of the product”.
Meat contraction shortens the distance between the meat surface and
the location of the temperature measurement. The effect is more pro-
nounced closer to the surface where temperature gradients are espe-
cially important. Simulations prove that a 1 mm shortening of the
a b
60
50
Temperature (°C)
40
30
20
10
0 0
0 10 20
Time (min)
Fig. 3. Temperatures measured in the central section of the 60 × 60 × 110 mm SM roasts subjected to the “10% steam injection” oven-cooking treatment, at core (a), and at 5 mm below the
surface (b), during cooking and air cooling (full lines with error bars). Temperatures calculated by the 3D conduction model, (a) at the center of the roast for a 5 mm decrease of the lateral
dimensions, and (b) of a variation of ±1 mm of the distance between the location of the thermocouple and the surface, for the measurement point which was the closest to the surface.
A. Kondjoyan et al. / Meat Science 95 (2013) 336–344
distance between the first thermocouple and the surface will vary by
3–5 °C the measured temperature (Fig. 3b). Juice migration can also
affect the kinetics of temperature. Juice velocity is not an explicit factor
of combined heat transfer-kinetic models. However, this velocity can be
assessed during the post-treatment from the calculated local water con-
centration in the meat. Such an approach was used to determine the ef-
fect of juice migration on the simulated temperature kinetics. The
simulated effect of juice migration on temperature was always less
than 0.4 °C at the center of the product, between 1 and 1.5 °C in the
major part of this product while it can reach 2–3 °C at 5 mm from the
surface. Thus, temperatures close to the meat surface are both very sen-
sitive to thermocouple movement and to sudden juice expelling which
easily explain the “oscillations” observed in the recorded temperature
(Fig. 3b). These under surface phenomena are very interesting to under-
stand juice migration and crust formation at the surface of the meat but
their effect on the average sample cooking loss is limited. Preliminary
analysis of model response has shown that the prediction of this
cooking loss is mostly sensitive to cooking time and average sample
temperature. This analysis leads to the conclusion that a simplifying as-
sumption of a heat transfer by conduction and the consideration of the
effect of meat contraction on cooking time (when determined from the
core temperature) are enough to predict average cooking loss. In prac-
tice, predictions of cooking losses can be very sensitive to a variation
of the boundary conditions and to the heat transfer coefficient values.
For example, in the case of steaks cooked in the oven under the “10%
steam injection” conditions, meat temperatures were highly sensitive
to small variations in the initial and final oven door opening, transient
time between cooking and cooling, etc., and in product dimensions (un-
certainties on cutting and effect of heat shrinkage). It shall also be
noticed that during the air-cooling period, part of the energy accumu-
lated close to the meat surface heated the center of the product, thus
increasing the temperature (Fig. 3a). Thus, juice expulsion due to
protein denaturation–contraction continued during cooling until
the temperature throughout the meat fell below 40 °C. This period,
which can be very long when the product is thick and air-cooled,
must absolutely, be taken into account for a sound prediction of
cooking losses.
4.2. Mass transfer
4.2.1. Determination of the parameters of the kinetic models on meat
samples
When the parameters of the heat transfer model are known, the var-
iation in water content in the sample depends on X0, Xeq and on the
parameters of the kinetic models: Ea, A and B (Eqs. (2) and (3)). The
values of these three last parameters were determined by minimization
100
80
Temperature (°C)
60
40
20
30 40 50 0 10 20 30 40 50
Time (min)
 A. Kondjoyan et al. / Meat Science 95 (2013) 336–344 341

of the absolute difference between calculated and measured cooking
losses on meat cubes of side dimensions 10, 30, 50 and 70 mm and
rectangular cuboids which dimensions were 30 × 30 × 120 and 70 ×
70 × 280 mm cut from the SM muscle and heated during different
times in a water bath at 50, 70 or 90 °C and finally cooled in air for
10 min. The water content at equilibrium Xeq was that given in Fig. 1,
and the value of X0 was that measured on each sample. It is important
to consider the initial water content of each of the samples, as it can
vary from one cut to another in the same muscle and can lead to appar-
ent cooking losses. Different sets of parameters Ea, A and B were
obtained in the range 15,000 ≤ Ea ≤ 19,000, 6.5 × 10−4 ≤ A ≤
8.5 × 10−4, 1.110 ≤ B ≤ 1.280, respectively, which led to very close
average differences between the measured and predicted weight losses.
Final values of Ea = 18,000–19,000, A = 6.8 × 10−4 and B = 1.275
were retained for the rest of this work. These parameter values led to
an algebraic average difference between calculated and measured
weight losses of 0.4%, thus, ensuring that model predictions do not sys-
tematically over- or underestimate the experimental outcomes. The
average absolute difference between predictions and measurements
was 1.7%, and thus close to the average absolute deviation on the mea-
surements which was 1.4% for all the cube dimensions and all the
heating conditions tested. Predicted and measured weight losses are
compared in Fig. 4. Model predictions agree with measurements on all
the meat cubes and whatever the sample dimensions (side of 10, 30
and 50 mm) and time–temperature conditions. There is only one case,
i.e. the 70 mm cube heated at 50 °C, where model predictions tend to
systematically underestimate the measurements. Note, however, that
it is at 50 °C that repeatability between sample measurements was
poorest due to the strong effect of muscle structure on cooking losses
(Oillic et al., 2011). Other differences higher than 2% were also
observed, mainly during short treatments which are the most sensitive
to small variations in heating–cooling conditions.
Extension of the kinetic models to other muscles than the SM was
tested on 10 mm meat cube cut from IS, LT, ST muscles heated in a
water bath (Oillic et al., 2011). The parameters of the model were the
same as on the SM muscle, whereas initial and equilibrium water con-
tents (X0 and Xeq) were specific to each muscle (Table 1, Fig. 1). The
average algebraic difference between predictions and measurements
was the same as that obtained on SM for LT and IS muscles but higher
on ST muscle. This difference can be reduced by adapting the model's
parameters.
4.2.2. Validations of model predictions on other samples or for other
cooking conditions
A first set of validations was performed on SM muscle using two
meat rectangular cuboid dimensions 30 × 30 × 120 mm and 70 ×
70 × 280 mm heated at 70 °C in a water bath (Fig. 5). For the same
initial water content, weight loss will tend towards the same value
after an infinite period of time, since it is calculated from Xeq which
only depends on temperature and not on sample dimensions. How-
ever, the kinetics of weight loss is greatly affected by sample dimen-
sions, and the 4-fold increase in one of the sample dimensions slowed
down the weight loss compared to what was observed on the corre-
sponding meat cubes. Thus, the difference in measured weight loss
between the 30 × 30 × 30 mm sample and the 30 × 30 × 120 mm
sample is − 15, − 24, − 10, − 9 and 6% after 4, 8, 16, 41 and 66 min
of treatment, respectively (Figs. 4 and 5). Similar but less pronounced
differences were measured between the 70 × 70 × 280 mm and the
70 × 70 × 70 mm samples. The model is able to take these variations
into account, since the average algebraic and absolute difference
between predicted and measured values on both the 30 ×
30 × 120 mm and the 70 × 70 × 280mm meat rectangular cuboids
was 0.3% and 1.6%, respectively. This ensures that model predictions
can be reliably used on samples of other shapes than cubes.
In a second stage, the combined model was validated in the oven
(Table 2). The “full steam” or “10% steam” injection conditions were
applied on samples of 3 different dimensions, i.e. a steak (15 ×
70 × 70 mm), a small roast (60 × 60 × 110 mm) and a big meat
piece (110 × 110 × 110 mm). Initial meat temperature varied from 0
to 10 °C, and final center temperature was 50–55 °C. Samples were
cooled either in room-temperature air or in ice-water until core tem-
perature was less than 40 °C, which stopped the expelling of juice due
to protein denaturation–contraction. Dry air conditions were also
applied on the 60 × 60 × 110 mm roast to explore the effect of

A B
60 60
Cooking loss (%) 10x10x10 mm Cooking loss (%) 30x30x30mm
50 50 90°C
90°C
40 40
70°C 70°C
30 30
20 20
10 50°C 10 50°C
Time (min) Time (min)
0 0
0 5 10 15 20 25 0 20 40 60 80

C D
60 60
Cooking loss (%) 50x50x50mm Cooking loss (%) 70x70x70mm
50 50
90°C 90°C
40 40
30 30
70°C 70°C
20 20
50°C
10 10 50°C
Time (min) Time (min)
0 0
0 20 40 60 80 100 0 30 60 90 120 150 180

Fig. 4. Cooking losses predicted by the combined heat transfer and kinetic models on SM meat cubes heated in the water bath (full lines) compared against measured losses (symbols □, Δ, ◊).
The side length of the cube is (A) 10 mm, (B) 30 mm, (C) 50 mm, and (D) 70 mm, respectively. Symbols correspond to the different water bath temperatures.
342 A. Kondjoyan et al. / Meat Science 95 (2013) 336–344

40 and separated evaporation losses from dripping losses by collecting

Cooking loss (%)
35
30
30x30x120mm
25
70x70x280mm
20
15
Parallelepipeds
10
70°C
5 evaporation and preventing crust formation. The heat treatment lasted
Time (min)
0
0 100 200 300 400
Fig. 5. Cooking losses predicted by the combined heat transfer and kinetic models on
SM meat rectangular cuboids heated in the water bath at 70 °C (full lines) compared
against measured losses (symbols □, Δ). The meat rectangular cuboid dimensions
are 30 × 30 × 120 mm and 70 × 70 × 280 mm respectively.
evaporation on cooking losses. The heat transfer coefficient values were
those given in Table 1, initial water content was X0 = 2.9 kg water/kg
DM, and the parameters of the kinetic models were Ea = 19,000,
A = 6.8 × 10−4 and B = 1.275.
Under the “full steam” or “10% steam” injection conditions, evapora-
tion is either inexistent or only concerns condensates or juice already ex-
pelled from the meat. The combined model was able to predict cooking
losses with algebraic and absolute differences between calculated and
measured weight losses of just 0.3 and 1.7%, respectively. On the small
roast, an increase in meat temperature due to an increasing steam injec-
tion during a 24-min cooking period led to a 17 to 19% variation in
cooking losses. Juice loss was limited to 1–1.5% after cooling in ice-
water but was 5.0–5.5% after air cooling. This additional loss was due
to juice expulsion stemming from protein denaturation–contraction
which continued during cooling until meat temperature fell below
40 °C (Fig. 3a) and not to water evaporation which was limited to the
juice already expelled from the meat (as checked by cooling the meat
wrapped or unwrapped). Extension of the kinetic models to other mus-
cles than the SM was tested by comparing model prediction to measured
weight loss on a 60 × 60 × 110 mm roast cut from the LT muscle.
Results proved that model was able to predict the cooking losses
measured on the LT muscle subjected to steam injection.
How does evaporation affect cooking losses during roasting? Goni
and Salvadori (2010) oven-roasted ST muscles in air at 180−200 °C
the juice that dripped from the muscle into a pan. Measured drip losses
represented about 50% of cooking losses (from 40 to 70% depending
on roasting conditions). However, this measure underestimates the
amount of juice actually expelled from the meat, as part of this juice
necessarily evaporates before dripping into the pan. We ran experi-
ments to analyze the effect of evaporation on weight loss when oven-
cooking a 60 × 60 × 110 mm roast subjected at 95 °C airflow and
cooled in ice-water. During this treatment, the temperature at the
meat surface remained below the boiling temperature, thus limiting
43 min until the temperature at the center of the meat reached 50 °C,
which led to a cooking loss of 11%. The value of the cooking loss predict-
ed by the combined model ranges from 11.6 to 16% depending on the
heat transfer value introduced into the model, which is between 20
and 30 Wm−2 K −1. Comparing measured against predicted cooking
losses leads to the conclusion that under the 95 °C air-flow, evaporation
did not bring any additional loss, and that evaporated water comes sole-
ly from the juice already expelled by protein denaturation and contrac-
tion. A second air experiment was performed in the oven by increasing
the air temperature up to 250 °C to promote water evaporation and
crust formation at the meat surface (Table 2). In this case, heat flux at
the meat surface could not be accurately determined as it is impossible
to reliably model the decrease in water activity and crust formation.
However, a good approximation of weight loss due to protein denatur-
ation–contraction can be obtained by assuming that the temperature at
the surface of the roast is equal to boiling temperature from the begin-
ning and throughout the heat treatment. This assumption neglects the
weight lost in the crust and assumes that the time needed to reach
100 °C just under the crust is very short. This is a reasonable assump-
tion, since the crust is only a few millimeters thick and represents a
very small fraction of the total volume of the meat piece. Moreover,
infrared measurements of the temperature at the surface of a meat
piece subjected to 200–250 °C air jets, where velocity was comparable
to that measured in the oven showed that surface temperature reached
boiling temperature in less than 5 s and subsequently continued to rise
towards the air temperature. These experiments, which are not detailed
here, employed the same procedures as our earlier experiments on
meat slices (Kondjoyan & Portanguen, 2008; Kondjoyan et al., 2010).
The assumption of a constant boiling point temperature just under the
surface of the meat (i.e. just under the crust) leads to a faster tempera-
ture rise at the center of the meat than during the pure steam treatment,
and thus shortens the treatment compared to a pure steam condition.

Table 2
Cooking losses predicted by the combined heat transfer and kinetic models on SM meat cuts heated in the oven compared against measured losses. The size of the meat pieces
varied from thin steaks up to big muscle cuts. Different air/steam conditions were applid to analyze the transition from wet air to dry air microenvironment.

Dimensions Oven Ti Cooking Cooling Cooling CL exp Calculated CL average Total diff Remark
(mm) condition (°C) duration duration (%) calc.-exp. (%)
(min) (min)

15 × 70 × 70 10% steam 0 4.33 Air 10 14.9 ± 0.4 8% cooking +5% cooling −1.9 Strong sensitivity to
injection sample thickness and
boundary conditions
60 × 60 × 110 Pure steam 10 24 Ice-water 20 18.8 ± 0.9 19.2% cooking 1.3% cooling +1.7
60 × 60 × 110 10% steam 10 24 Ice-water 20 16.3 ± 1.1 17.3% cooking 1.0% cooling +2.0
injection
60 × 60 × 110 10% steam 0 31 Air 40 25.3 ± 0.7 19.9% cooking +4.9% cooling −0.5
injection
1 2 1 2
60 × 60 × 110 Dry air 95 °C 10 43 Ice-water 20 11.1 ± 0.7 10.6% to 15.2% cooking +0.41 to 1.0% No crust
cooling 1
h = 20 Wm−2 K−1
2
h = 30 Wm−2 K−1
60 × 60 × 110 Dry air 10 21 Ice-water 20 23.1 ± 3.25 Evaporation and crusting
250 °C not modeled
110 × 110 × 110 Pure steam 10 65 Ice-water 20 22.0 ± 0.7 22.7% cooking +1.5% cooling +2.2
110 × 110 × 110 Pure steam 10 65 Ice-water 20 22.0 ± 0.7 22.7% cooking +1.5% cooling +2.2
110 × 110 × 110 10% steam 0 73 Air 70 31.9 ± 0.7 24.7% cooking +5.5% cooling −1.7
injection

(1) and (2) are two assumption on surface evaporation.

 A. Kondjoyan et al. / Meat Science 95 (2013) 336–344
This shorter cooking time, which is observed experimentally (Table 2),
leads to a similar cooking loss (20%) as in the pure steam treatment.
Comparing this value against the 23% weight loss measured during
the 250 °C air treatment suggests that even under these strongly evap-
orative conditions, more than 80% of the evaporated water came from
the juice already expelled at the surface of the meat by protein denatur-
ation and contraction. Further research is needed to confirm this result.
Present paper is focused on the ability of combined heat transfer to
kinetic models to predict the evolution of the average cooking loss of
beef meat samples. It can however be interesting to observe the variation
of the water concentration locally in the sample. Evolution of the water
concentration profile at mid-width, mid-height and in the length direc-
tion of the 60 × 60 × 110 mm roast cooked in the oven under the 10%
steam injection condition is presented in Fig. 6. A comparison with the
simulated results of Feyissa et al. (2013) prove that these profiles have
the same shape as the ones obtained by these authors using what they
consider as the most realistic value of meat porosity (10 −17 m2). A
more quantitative comparison between the results issued from the two
models will require applying exactly the same boundary conditions at
the geometry borders which was not the case here. Moreover, a real dis-
cussion on models performance will need an accurate measure of the
water concentration profiles. Feyissa et al. (2013) have measured the
local water content each 4 mm in the sample, which is not sufficient to
determine the slope of the water concentration profile close to the sur-
face where the variations are the more pronounced. This is part of future
researches which are beyond the scope of present paper.
5. Conclusion
Average meat temperature can be predicted in a wide range of
cooking conditions and sample dimensions by assuming that heat trans-
fer in the product is purely conductive and using adequate effective heat
transfer coefficient values. Accurate comparison between the simulated
and the measured temperature is difficult near the surface of the product
where thermocouple movements due to heat shrinkage lead to impor-
tant standard deviation values in the measurements. Heat shrinkage
affects the temperature at the center of the product, and thus shortens
the cooking time when it is determined from the temperature measured
at this location. Present results help in quantifying how differences in
meat shrinkage affect the cooking time when it is controlled from the
measurement of the temperature at the center of the meat product.
Despite its simplicity, the combined heat transfer and kinetic models
was able to predict the cooling losses of Semimembranosus cuts of very dif-
ferent dimensions with an average absolute difference between simula-
tions and the measurements of 6.8% of the cooling loss value during
water bath experiments, and of 8.7% during oven wet cooking treatments.
3,5
3 (2006a). Modelling coupled heat–water transfers during a decontamination treatment
X (kg water/kg DM)
2,5
t=0s
t=600s
2 surface temperature and water activity under fast transient air temperature conditions.
t=1200s
t=1800s
1,5
1 viande. Annales de Technologie Agricole, 4, 849–871.
0 20 40 60
Lenght (mm)
Fig. 6. Evolution of the calculated water concentration profile at mid-width, mid-height
and in the length direction of the 60 × 60 × 110 mm roast cooked in the oven under the
10% steam injection condition. Profiles have been simulated at 3 different cooking times:
600, 1200 and 1800 s.
343
During air cooling, part of the energy accumulated in the meat close to the
surface is able to heat the center of the product, increasing its tempera-
ture. Thus, juice expulsion due to protein denaturation–contraction con-
tinues until the temperature throughout the meat falls below 30–40 °C.
This period must absolutely be taken into account for a sound prediction
of cooking losses. Results obtained on small meat cubes derived from
other muscles than the SM tend to prove that the combined heat transfer
and kinetic models can also be applied to predict cooking losses from cuts
of LT, IS or ST meat, while it cannot be applied on some other muscles
such as the MA that have highly specific structure. During air cooking,
water evaporation at the meat surface is a further driver of cooking loss.
However, our findings lead to the conclusion that most of the evaporation
comes from juice already expelled by protein denaturation and contrac-
tion. Thus, weight loss by evaporation cannot just be added to the losses
obtained under wet conditions. Further research needs to be directed to-
wards analyzing and modeling crust formation.
Acknowledgments
This research was given financial support by the ProSafeBeef
project under the European Commission Sixth Framework Programme
(Food-CT-2006-36241), the French government (ANR-CiFRE), the CIV
and the ADIV. The authors thank Céline Bernard for providing important
technical assistance.
References
Baghe-Khandan, M. S., & Okos, M. R. (1981). Effect of cooking on the thermal conductivity
of whole and ground lean beef. Journal of Food Science, 46, 1302–1305.
Bouhrara, M., Clerjon, S., Damez, J. -L., Kondjoyan, A., & Bonny, J. -M. (2011). Dynamic
MRI and thermal simulation to interpret deformation and water transfer in meat
during heating. Journal of Agricultural and Food Chemistry, 59(4), 1229–1235.
Bouhrara, M., Clerjon, S., Damez, J. -L., Kondjoyan, A., & Bonny, J. -M. (2012). In-situ
imaging highlights local structural changes during heating: The case of meat. Journal
of Agricultural and Food Chemistry, 60(18), 4678–4687.
Dhall, A., Halder, A., & Datta, A. K. (2012). Multiphase and multicomponent transport with
phase change during meat cooking. Journal of Food Engineering, 113(2), 299–309.
Feyissa, A. H., Gernaey, K. V., & Adler-Nissen, J. (2013). 3D modelling of coupled mass
and heat transfer of a convection-oven roasting process. Meat Science, 93, 810–820.
Ghisalberti, L., & Kondjoyan, A. (1999). Convective heat transfer coefficients between
air flow and a short cylinder. Effect of air velocity and turbulence. Effect of body
shape, dimensions and position in the flow. Journal of Food Engineering, 42, 33–44.
Goni, S. M., & Salvadori, V. O. (2010). Prediction of cooking times and weight losses
during meat roasting. Journal of Food Engineering, 100, 1–11.
Goni, S. M., & Salvadori, V. O. (2012). Model-based multi-objective optimization of beef
roasting. Journal of Food Engineering, 111, 92–101.
Kondjoyan, A. (2006). A review on surface heat and mass transfer coefficients during
air chilling and storage of food products. International Journal of Refrigeration, 29,
863–875.
Kondjoyan, A., Chevolleau, S., Greve, E., Gatellier, P., Santé-Lhoutellier, V., Bruel, S.,
Touzet, C., Portanguen, S., & Debrauwer, L. (2010). Modelling of the formation of
heterocyclic amines in slices of longissimus thoracis and semimembranosus beef
muscles subjected to jets of hot air. Food Chemistry, 123(3), 659–668.
Kondjoyan, A., McCann, M., Rouaud, O., Havet, M., Foster, A., Swain, M., & Daudin, J. D.
of the surface of solid food products by a jet of hot air — I. Sensitivity analysis of
the model and first validations of product surface temperature under constant air
temperature conditions. Journal of Food Engineering, 76, 53–62.
Kondjoyan, A., McCann, M., Rouaud, O., Havet, M., Foster, A., Swain, M., & Daudin, J. D.
(2006b). Modelling coupled heat–water transfers during a decontamination treatment
of the surface of solid food products by a jet of hot air — II. Validations of product
Journal of Food Engineering, 76, 63–69.
Kondjoyan, A., & Portanguen, S. (2008). Prediction of surface and “under surface”
temperatures on poultry muscles and poultry skins subjected to jets of superheated
steam. Food Research International, 41(1), 16–30.
Laroche, M. (1978). Facteurs influencant les pertes d'eau pendant la cuisson de la
Lepetit, J. (2007). A theoretical approach to the relationships between collagen content,
collagen cross-links and meat tenderness. Meat Science, 76, 147–159.
Lepetit, J., Grajales, A., & Favier, R. (2000). Modelling the effect of sarcomere length on
collagen thermal shortening in cooked meat: Consequences on meat toughness. Meat
Science, 54, 239–250.
Modzelewska-Kapitula, M., Dabrowska, E., Jankowska, B., Kwiatkowska, A., & Cierach,
M. (2012). The effect of muscle, cooking method and final internal temperature
on quality parameters of beef roast. Meat Science, 91, 195–202.
344 A. Kondjoyan et al. / Meat Science 95 (2013) 336–344

Oillic, S., Lemoine, E., Gros, J. B., & Kondjoyan, A. (2011). Kinetics analysis of cooking
losses from beef and other animal muscles heated in a water bath — Effect of sample
dimensions and prior freezing and ageing. Meat Science, 88(3), 338–346.
Oroszvari, B. K., Bayod, E., Sjöholm, I., & Tornberg, E. (2006). The mechanisms controlling
heat and mass transfer on frying of beef burgers. III. Mass transfer evolution during
frying. Journal of Food Engineering, 76, 169–178.
Palka, K., & Daun, H. (1999). Changes in texture, cooking losses, and myofibrillar structure
of bovine M-semitendinosus during heating. Meat Science, 51, 237–243.
Santos, M. V., Zaritzky, N., & Califano, A. (2008). Modeling heat transfer and inactivation
of Escherichia coli O157:H7 in precooked meat product in Argentina using the finite
element method. Meat Science, 79, 595–602.
Tornberg, E. (2005). Effects of heat on meat proteins — Implications on structure and
quality of meat products. Meat Science, 70, 493–508.
Tsai, S. J., Unklesbay, N., Unklesbay, K., & Clarke, A. (1998). Thermal properties of
restructured beef products at different isothermal temperatures. Journal of Food
Science, 63, 481–484.
Van der Sman, R. G. M. (2007). Moisture transport during cooking of meat: An analysis
based on Flory–Rehner theory. Meat Science, 76, 730–738.