Journal of Food Engineering 110 (2012) 269–277

Contents lists available at ScienceDirect

Journal of Food Engineering

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

Modelling of food and food processes

Gilles Trystram
UMR 1145, INRA, AgroParisTech, F91305 Massy, France

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

Article history: Modelling is an important task in food engineering and numerous approaches are possible. Based on the
Available online 7 May 2011 massy experience the paper discusses the challenge, interest and difficulties of modelling of food and
food processes. Some key points are specifically discussed mainly on the base of heat processing opera-
Keywords: tion. Heat and mass transfer and their consequences on structure evolution and on the building of the
Modelling chemical content of food are discussed. Different kind of modelling approaches are also discussed consid-
Food ering control, optimisation and design objective. The coming challenge is as a conclusion presented on
Food process
two main questions. The building of parameters that are necessary for simulation, mainly using molec-
ular modelling approaches and the challenges related with scales description and the way a continuous
understanding becomes probably possible in the future between all the scales.
Ó 2011 Elsevier Ltd. All rights reserved.

1. Introduction  Even more limiting, the lack of scientific knowledge on process–

properties relationships is punctually overcome in practice by
Modelling of food products and food processes is a complicated empiricisms leading to a non-explicit and fragmented knowl-
task due to the lack of knowledge concerning mechanisms, the dif- edge, embodying the know-how of the enterprise. This exper-
ficulty for performing experimentation and obtaining numerous tise is, by definition, scarcely available, and cannot be
reliable data and the uncertainties concerning food properties. straightforwardly integrated in mathematical models (Ioannou
For long time, food processing was mostly dedicated to product et al., 2006; Perrot et al., 2006).
safety, stabilization and operations scale up in industry, and pro-
cess engineers applied concepts from chemical engineering and fo- These food processes are usually dynamic in nature. The repre-
cused on time–temperature diagrams for predicting and limiting sentation by models consists of sets of ordinary and/or partial dif-
residual spores or micro-organisms in foods (Sablani et al., 2007). ferential and algebraic equations. The dynamic of the system are
Despite a very large number of scientific publications (Bimbenet also in numerous applications encapsulated in the form of empir-
et al., 2007), the extension of this approach to food processing and ical models (data driven models) or hybrid models (Trystram and
design encounters the following limits: Courtois, 1996) taking into account different sources of knowledge.
The complexity of these models ranges from the simple (e.g.
 Many unit operation models are developed but their applicabil- empirical kinetics) to the highly complex (e.g. fluid dynamics cou-
ity is not straightforward; pled with heat transfer and reactions). It depends on the aim of the
 numerous uncertainties are available both for raw materials study: state estimation and simulation, optimised control, struc-
properties, operation design parameters and reliability of oper- tural understanding, etc.
ating conditions (specifically without control implemented); A Systematic Approach for Food Engineering Systems (SAFES)
 presently, the accurate knowledge of food scientists has led to based on the theoretical frame of Irreversible Thermodynamics
specific models, valuable in a tiny domain, either of composition Processes (Fito et al., 2007) was also proposed. Nevertheless this
or of physico-chemical environment. Moreover, their concep- last approach is based on the hypothesis of identifiability of the
tual frame does not allow easy integrating of the results from system when it is not often valid. The difficulty keeps then entire
other existing models with those under development. For when the aim is to establish relationships between food composi-
instance, most processing aspects are covered by differential tion and structure, and the properties of realistic food models. It
equations of heat and mass transfer phenomena, whereas becomes a very challenging task, including the presence of non-
microbiological or chemical aspects mostly take into account identifiability problems.
simple kinetic equations; coupling those is sometimes possible The high complexity of products elaboration and of the con-
but not easy or general. cerned processes is the main reason of the development of model-
ling. Understanding of complex mechanism in interaction, ability
E-mail address: gilles.trystram@agroparistech.fr to describe phenomenon carried out during processing and

0260-8774/$ - see front matter Ó 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jfoodeng.2011.05.001
270 G. Trystram / Journal of Food Engineering 110 (2012) 269–277

storage, control applications or optimisation strategies are differ- are today available that permits to lower the impact of implemen-
ent ways of using modelling tools in food engineering. For a long tation. Some relevant papers discuss the interest of using generic
time numerous works are dedicated to such development and a software for this task (Fluent, Comsol for example). Identification
huge number of papers deals with model, in a rigorous way, or not. follows in order to calculate the appropriate values of parameters
Food process engineering has three inherent objectives: to and model validation is normally the last step. The figure illus-
understand the phenomena carried out during processing, to de- trates numerous objectives of modelling.
sign unit operations or processes and to control them. Considering
the development of tools available for chemical industries pur- 2.1. Heat and mass transport and transfers approaches
poses and for food industries, a wide difference appears. Even if
numerous models are proposed along time for food, the applica- Heat and mass transport and transfer are certainly one of the
tions of simulation tools for real situation are low. Lack of knowl- most studied topics for modelling. Even if the main purpose of
edge, lack of appropriated parameters for models are often the thermal operations is to give to the product some specific char-
discussed as responsible of the situation. In the case of food pro- acteristics such as: taste, colour, moisture content, dimensions,
cesses, the objectives are generally complex and numerous. thickness, structure and mechanical properties, most of the studies
Nevertheless, models are well known in Food Process Engineer- deals firstly with moisture and temperature evolution. For Baking,
ing. For a long time, they were only used for design purposes, or for for example, this is only recently that one becomes aware of the
interpretation of the phenomena carried out during processing. great interest of a complete understanding of the baking process
Numerous calculations are computed for the choice of exchange and transformation phenomena induced inside the product, not
surface, nature of materials, kind of unit operations, cleaning fre- only to control product quality, but also to consider the baking unit
quency and accurate design calculations. The preservation pro- operation as a useful tool for new product development. In this
cesses is certainly the most studied in order to reduce the way, modelling of baking process becomes a priority in order to fill
microbial content of food. Numerous applications are also dedi- the lack of data that on-site measurements are not yet able to as-
cated to formulation and design of the best recipe that is obtained sess and to reach the required higher level of knowledge. Mecha-
through the coupling of models and optimisation procedure (Re- nistic modelling is complex and ambitious but it has the
sponse Surface Optimisation applications). Each time the proposed advantage of being easily adjusted to different geometries, baking
tools are different and specific. conditions and product recipes. By solving heat and mass balances
The purpose of this article is to present, describe and illustrate that are involved during the process, operating conditions can be
methods for steady state or dynamic modelling of food and food related to a quantitative understanding of the product evolution.
processes. The article is mainly based on our experience, partially There are a significant number of deterministic models on heat
developed on baking and other thermal processing operations, and mass transfer in baking ovens in the cereal products area
and discuss the stakes for the development of tools associated with (Broyart and Trystram, 2003; Mondal and Datta, 2008; Paloheimo
modelling tasks in order to help and increase the filed of applica- et al., 1984; Therdthai et al., 2002; Wagner et al., 2007). The
tion of models in Food Engineering. description of Fourier and Fick laws, including coupling action be-
tween both is relatively simple and well performed, mainly assum-
ing that thermal parameters are constant. The use of time varying
2. Modelling of food and food processes, main approaches laws for thermal and physical parameters is not easy and in fact
not often proposed. Generally, these models predict only average
The purpose of a mathematical model is to capture relevant fea- temperature and moisture content of the product. The difficulty
tures (in a given context) of a complex object or process, based on of predicting changes of dough characteristics during cooking lies
existing theoretical understanding of the phenomena and available in the fact that heat and mass transfers in porous foodstuffs are
measurements. Current industrial applications usually rely on ex- very complex phenomena as well as the physic-chemical transfor-
tremely simplified, stationary models that cannot produce a realis- mation kinetics that they induce. Broyart (1998) proposes a com-
tic evaluation of transient effects on plant performance, quality plete description of heat and mass transfers in case of biscuit
and safety conditions and environmental impact. The modelling baking and more recently, the case of bread as a model of porous
and simulation research efforts should be directed towards main media illustrates well that complicated heat transport are per-
phenomenological aspects, coupling different scales, such as heat, formed during baking. Evapo-condensation implies an increase of
mass, momentum, population balance coupled with biochemical heat transport and changes significantly the rate of mass transport
reactions. It must be stated that the model has to be well related also (Wagner et al., 2007).
with the objective. Complexity of modelling is not obvious. Often Despite the proximity of chemical and food or bioengineering,
for simple studies, simple models are sufficient. For optimisation the use of general Purposes simulators developed first for the
for example, most of the models used are steady state ones. For chemical industries is poor in the food domain. Datta (2008) dis-
control purposes, of smart sensors design, dynamic model are ex- cuss the interest of development of food operation simulation tools
pected. Each model is dedicated to one specific goal and as a con- using this kind of general software. Many advantages are high-
sequence is a specific one. A good, and concise, definition of lighted, and it is true that Fluent, Comsol or other tools are very
modelling is the building of tools able to predict the future of the useful. Part of the applied mathematic questions is easily solved,
product or process with good and enough accuracy depending on and mixing physics with Comsol or Fluent is easy. Nevertheless,
the purpose of model building. it must be pointed out that in most of the case, the discussion of
Fig. 1 presents the main steps necessary for modelling. The most the assumptions becomes poor and the reliability of models is
important step concerns the design of the structure of the model; not obvious. Concerning the use of general simulations software,
choice of the variable, nature of the relationship, the test of identi- numerous reasons are discussed that limit the field of applications.
fiability of the structure represents the validation of the ability to The main one is related with the complicated raw material that
calculate, if necessary, the unknown parameters from a given set Food industries have to cope. The databases are not reliable enough
of experiments and recording of variables. Experimental, in a per- and the work for the reconstruction of parameters that are use by
fect world will be a consequence of the previous task. It must be theoretical models available on Tools (ASPEN or PROSIM for exam-
recognised that often, experimental design is performed before ple) is important. A general methodology for optimising the oper-
any other task. Model implementation arrives as a key step. Tools ating conditions of an industrial distillation plant was developed.
 G. Trystram / Journal of Food Engineering 110 (2012) 269–277 271

Fig. 1. General framework of modelling steps, from design to validation. Main objectives of modelling in the context of food and food process engineering.

From a nominal point, combining measurements and data recon- using Residence time distribution are used, and sophisticated 3D
ciliation for model validation permits to define initial conditions visualising systems performed on maquette are often necessary.
(Samir et al., 2009). Then simulation results obtained with PROSIM
Tools (Simulis Thermodynamics, Thermophysical properties calcu- 2.3. Heat and mass transfer coupling with (bio)chemical reactions
lation server, http://www.prosim.net) give a fine and reliable pic-
ture of the plant, such as complete profiles in columns, allowing A general figure that represents the coupling between heat and
to study in detail, including not measurable variables and conse- mass form one side and the consequences on reaction and struc-
quently to improve it. Some proposed improvements have been al- ture evolution on the other side is proposed Fig. 2. During a long
ready implemented leading to reduction of ethanol losses and to a time most of the works are mainly dedicated to heat and mass
better quality of the neutral spirit. The next step is to test the opti- transfer. Step by step mechanics of the product arrives as a key,
mal conditions obtained by simulation. mainly for soft matter food where rheological properties are a
key (foam, emulsion, gel). Models are improved and heat and mass
2.2. Heat and mass transfer coupling with CFD transfer describe the evolution of state variable that are key for
modelling the evolution of mechanics. Despite numerous pro-
The Computational fluid dynamics tool becomes very useful and gresses, this point is still poor and specifically for solid food prod-
numerous applications are done. A major interest is to identify uct. Reactions in food, during processing or conservation are
accurately the velocity field, in order, for example, to obtain an studied for a long time mainly because of preservation processes,
accurate heat or mass transfer description that permits the calcu- where microbiological destruction was modelled (generally on a
lation of products heating, drying ands its consequences. Pulverisa- simple way). More recently high-level models are developed for
tion drying is probably the most studied operation. For industrial chemical and biochemical reactions. Here again, state variables
baking of cereal products where tunnel ovens are used, a CFD ap- that are the most often used are moisture content and tempera-
proach was done (Mirade et al., 2004; Wong et al., 2007) to predict ture. Heat and mass transfer models describe these two variables.
the air temperature and velocity profiles inside the baking cham- It is the purpose of Fig. 2 to illustrate this relation where the com-
ber of an industrial indirect gas-fired tunnel oven used for biscuit plexity comes from the necessary feedback between evolution of
baking. Two three-dimensional CFD models (one not covering the
conveying band of biscuits and the other including it) are used to
describe the complex air circulation resulting from the mecha-
nisms of air input and exit at the ends of the oven and of air extrac-
tion through the different extraction points located along the oven
length. Comparison of numerical results with experimental mea-
surements shows a fairly close agreement in the qualitative predic-
tion and a few inaccuracies in the quantitative prediction of the air
temperature profiles within the baking chamber. Furthermore, the
comparison also reveals great differences in the air velocity pro-
files. It illustrates well the powerful of such tools, when the right
coupling between air movement and heat transfers are described.
The main difficulty of such approach is still related with experi-
mental validation, because a large amount of data is obtained from Fig. 2. Representation of the relation between heat and mass transfer and transport
simulation and only a few, not really reliable, with large uncertain- mechanism and coupling with the evolution of structure and biochemical reaction
ties are obtained from real measurements. Interesting approaches performed during processing and conservation.
272 G. Trystram / Journal of Food Engineering 110 (2012) 269–277

Fig. 3. Illustration of the scale where works are necessary in order to describe how transformation and conservation of food is performed.

reaction or mechanic and the thermodynamic of food evolution. It for example). This approach is powerful for studying at the same
is clear that heat and mass transfer is not a sufficient description of time engineering of the operation and formulation of the biscuit.
product evolution. Same approach was done for Coffee roasting, sponge cake baking,
Many factors have to be considered in food engineering, includ- acrylamide in frying etc. The challenge is great if food engineers
ing raw material, formulation, industrial food processing, storage want to face chemical food safety.
and domestic reheating conditions. Thus, new tools are required
to estimate and predict the dynamic elaboration of food properties 2.4. Heat and mass transfers coupling with mechanical mechanisms
during industrial processing, storage and domestic operations. The
coupling between the reaction pathways that are involved during According to Fig. 2, one of the most promising field concerns the
heating of products: Maillard reactions, oxidation (for example of changes in food product structure resulting from the process evo-
polyunsaturated fatty acids), isomerisation are important to study lution. More recently, the improvement of observation and analyt-
and so complex that without any models it is difficult to propose ical methods (imaging techniques, magnetic and electronic beams)
interpretation of the phenomenon. The chemical reactivity is also has allowed investigating the different structural scales and the
important. Moreover, most of the kinetic data published in the lit- interactions between chemical species, mainly between small
erature and dealing with chemical reactions involved in the elabo- and macromolecules. The molecular mobility concept has allowed
ration of nutritional and sensorial characteristics of food, have to describe the joint influence of physical variables such as mois-
been set up in liquid model systems, not representative of solid ture and temperature, and to represent qualitatively the influence
food products, and with constant and controlled environmental of processes according to state diagrams. However, neither
variables (temperature, water activity, pH. . .), conditions which changes in state nor their dynamics and spatial distribution have
are very different from ‘‘realistic’’ industrial processing conditions. been thoroughly studied. The Bread baking illustrates well all the
It is a great challenge to face the coupling between heat and mass challenges. During baking, the process involves many coupled
transfer and reaction performed in food. Reaction engineering ap- transport phenomena that take place in the dough simultaneously
proaches becomes to be studied and developed for food products. and interdependently, water and gas vapourises at the cell/dough
In our own approaches, in frying (Achir et al., 2008) or baking interface, the cell volume increases provided that the dough film
(Broyart et al., 1995, Ait Ameur et al., 2006) successful results are retains gases and is deformable, biochemical reactions occur, such
obtained that permits to define new recipe or new operating con- as starch gelatinization and protein denaturation, and crust forma-
ditions for the operations. It is clearly an important trends for food tion. In front of the complexity and the number of coupled phe-
Engineering. In case of Ait Ameur works, for example (Ait Ameur nomena involved during bread baking, and in order to better
et al., 2006), the contaminants performed through Maillard reac- understand the baking process, different models of baking have
tions are modelled as influenced by biscuit temperature and mois- been proposed with different basic assumptions recently reviewed
ture content (assumes to be homogenous in biscuit). Classical (Mondal and Datta, 2008). There are many modes of transfer dur-
Arrhenius equations are used for simple reactions schemes. The ing baking. Mass transfer involves mainly water and CO2 appearing
ability to predict contaminants production is demonstrated as pos- in two phases, liquid or gaseous. Heat transfer occurs in all phases
sible even taking into account the recipe influence (kind of sugar by conduction but is also coupled to mass transfer. Moreover,
 G. Trystram / Journal of Food Engineering 110 (2012) 269–277 273

transport properties vary with temperature and composition Such works are necessary if we want to have significant pro-
evolutions. gresses in the field of food engineering. The main limitation is still
Deformation of bread is driven by pressure variation in the bub- the ability to perform experiment with real time recording of phe-
bles, which depends on transport phenomena. The increase of tem- nomenon well connected with structure evolution.
perature has two effects on pressure, thermal expansion and
increase of the gaseous mass contained in the bubbles, by vapouri-
3. Modelling and associated tools
sation of water vapour, and desolubilisation of CO2 in the case of
pre-fermented dough. Such expansion is limited by the dough
3.1. Identifiability
mechanical properties. After heating up, dough stiffens due to
starch gelatinization, protein denaturation and drying. This can lead
Modelling is not a one step ahead methodology. Far from this,
ultimately to crust formation that may stop further deformation.
food modelling implies good skills in: physics (e.g. thermodynam-
Furthermore deformation acts on transport properties through
ics, transfers...), biology (e.g. biochemistry, microbiology...), instru-
modifications in dimension and porosity and also changes the
mentation (e.g. sensors...), computer (e.g. programming...), applied
mechanisms of transport themselves. Dough films, for example,
mathematics,
can break under pressure gradient and mechanical constraints,
In general, the mathematical part of the work consists mainly in
leading to increasing connectivity between adjacent bubbles and
the analytical solving of the system of equation describing the
permeability, and permitting the gaseous phase to migrate by
model. Since, because an increase of the complexity is expected,
convection.
more and more models imply non-measurable variables and
The proposed first principle based model involve the main phe-
parameters not available in the literature. The identification of
nomena occurring during baking and considered as follows:
such parameters is done using optimisation routines. Since models
are generally non-linear and thus numerically solved, non-linear
(a) Surface temperature increases due to convective exchange
iterative optimisation routines are preferred. Two important ques-
with air and radiative exchange with the oven walls. This
tions has to be studied
leads to water evapouration at the surface of the product;
(b) temperature internal gradients ensure diffusive heat trans-
 Is the model structurally identifiable?
port from the surfaces to the core;
 Which measurements should I add to improve parameter
(c) liquid water migrates under moisture gradients from the
identifiability?
core to the surface;
(d) pressure in the bubbles increases, due to thermal gas expan-
Tools become available in order to study structure identifiabil-
sion, water vapourisation and CO2 desolubilisation;
ity (Vajda et al., 1989; Walter and Pronzato, 1997; Dochain et al.,
(e) pressure increase drives the deformation of the product;
1995).
(f) under partial water pressure gradients, water migrates by
Applications for drying of rice were proposed for a diffusion-
the mechanism of transport by evapo-condensation-
convection drying model and a compartmental-convection drying
diffusion;
model. The major conclusion is that the heat transfer coefficient
(g) with the increase of pressure and structure changes related
h in such models can only be identified if the product temperature
to the baking process, the porosity opens;
is measured. This is a very interesting point since this kind of mea-
(h) under total pressure gradient, while the bubbles communi-
surement is very difficult to obtain without any bias. For instance,
cate, the gaseous phase migrates from the core to the
putting a micro-thermocouple in the centre of the product is
surface.
known to lead to biased measures (due to the resulting heat short-
(i) The crust plays a crucial role implying mechanical con-
cut). On the other hand, h can be a priori fixed from literature as a
straints when the moisture content of the outer bread layer
given constant or as a K-related parameter (using Colburn analogy
becomes too low.
for instance). This also means that such models should not be ex-
pected to simulate precisely the product temperature. Hence, since
A model was established based on four conservative variables:
quality models strongly rely on this prediction, increased errors
energy, masses of water, of CO2 and of air. For each one, a conser-
may appear in the quality estimate. A minor conclusion is that
vation law can be established. All the variables were expressed rel-
measuring a single and global apparent moisture content of the
atively to dry matter. Xw (resp. Xw.liq) for example are total water
product is sufficient to identify all mass transfer coefficients. This
(resp. liquid water) contents by mass unit of dry matter, the vapour
was heavily expected since most authors adjust their (unknown)
moisture content was then Xw.vap = Xw  Xw.liq. Local thermody-
parameters on the basis of weight measures (Laroche and Courtois,
namic equilibrium is assumed. The two phases have the same tem-
2005; Lemaire et al., 1997).
perature at any spatial location. For water, it is assumed that liquid
water in the dough is in equilibrium with water vapour in the bub-
bles at any spatial location in the porous medium. That is sorption 3.2. Identification
equilibrium. Local repartition of water between liquid (Xw.liq) and
vapour (Xw.vap = Xw  Xw.liq) phases, if local total moisture content Numerous identification methods are proposed in the litera-
Xw, porosity e and temperature T are known is calculated. It is as- ture and supported by numerous algorithms available on tool
sumed that CO2 is solubilised in the water constitutive of the boxes (Matlab for example). Modelling is at least a two-step
dough at the beginning of baking and that during baking it is des- job, in which both parts have the same importance. After the
olubilised with increasing temperature. Air is assumed to be an choice and the design of the model structure, it becomes neces-
ideal gas and its solubility in water is neglected. At the beginning sary to identify the values of unknown parameters. Despite if
of baking, it is assumed that the bubbles are closed and contain numerous data are available in literature, mainly reliable data
only water vapour and CO2. After the bubbles had opened, air from are found for physical parameters, but considering reaction engi-
the product outside can migrate by diffusion in the bubbles. For neering applications it is obvious that lack of data is essential
structure building two situations are considered, the closed poros- and probably is one of the main differences with chemical engi-
ity where vapour-condensation-diffusion occurs and the open neering (for example thermodynamical models for food are still
porosity where Darcy transport is dominant. not reliable and an important Knowhow is necessary to used
274 G. Trystram / Journal of Food Engineering 110 (2012) 269–277

them). The identification procedure is an essential one and of applications are relevant, in numerous cases, they are not (Olmos,
course several methods have been developed. Most of them 2003, Banga et al., 2003). The main point is due to the extrapola-
are related to the model structures that are previously specifi- tion of the result, and the real applications are still poor. It is
cally chosen (Response Surface Methodology for example). Iden- recognised that the best way to develop optimisation applications
tification is a general problem for modelling, but the dynamic is linked to the quality of the model (reliability and accuracy).
models have specific requirements. One of the interests is to take into account the dynamic of the
Two main ways are encountered for identification. In the first unit operation. First principle models are used, and Black
case, the unknown set of parameters is estimated only from mea- Box dynamic models too (like Artificial Neural networks for
surements on the process. Estimation procedures are then used, example). In a large number of applications, hybrid models are
and numerous algorithms based on least squares method are avail- developed, and used for optimisation purposes.
able even if the estimator introduces a bias. On the other hand, if it Applications to food unit operations, for example, are despite pro-
is possible to simulate the model, the model-based method can be posed: micro filtration (Perrot et al., 1998), baking, drying (Noronha
used (Walter and Pronzato, 1997). A criterion is evaluated as a et al., 1996; Banga et al., 1991; Olmos et al., 2002), Thermal process-
function of the comparison between reality of experiments and ing (Terajima and Nonaka, 1996; Chen and Ramaswamy, 2002;
model simulation. If a difference is observed, the model parameters Balsa-Canto et al., 2002), fruits preservation (Morimoto et al.,
are adapted, until the criterion is satisfied. This problem is exactly 1997), texturation (de Blank et al., 1997).
the same as all optimisation problems. The adaptation loop can be Process optimisation is concerned with the selection of the pro-
performed off-line or on-line depending upon the use of the model cess operating conditions that maximise or minimise a function
or the specific problems that could be encountered. For example, in that relies technical and/or economic parameters called optimisa-
the case of adaptive control, implemented with a model or in the tion criterion. The optimal operating conditions are usually com-
case of diagnosis applications, on-line identification is performed. puted off-line and then applied to the process without change.
In most cases, off-line identification is chosen. The main advantage (Lee et al., 1999) Very simple controllers like the PID could success-
of the model-based methods is in the large number of possibility in fully accomplish the control for this approach. The quality of the
the choice of adaptation algorithms. Both linear and non-linear results is highly dependant of the quality of the model, but also
algorithms are available. close related with the way the optimisation problem is established.
In literature, the choice of the criterion is often neglected, even The choice of the criteria, of the constraints, the numbers of vari-
if it is an important one and if it has to be clear that the results (the ables that are considered are of main importance. The choice of
set of parameters) is only valid for the chosen criterion. Another the numeric method is important too.
equation will imply a different set of parameters. Before to use Application based on a first principle model was made for Bis-
an adaptation algorithm, it is important to choose a criterion to cuit baking (Olmos, 2003). Continuous industrial Baking is an inter-
be minimised during the optimisation procedure. Often, the crite- esting operation for optimisation purposes. First of all, the number
rion is defined as a simple quadratic function of the error between of actuators is high and the main question is to establish the profile
modelling and process outputs, or states. The criterion is for exam- of several key variables along the oven. Secondly, the optimisation
ple (where ym is the measurement and y the model value of vari- has to consider several unit operations in one oven. Indeed, drying,
able y): baking, flavouring, colouring and texturation occur during the tra-
Z vel in the oven. Energy and duration are two very important crite-
t max
ria. The proposed approaches in the bibliography are mainly
E¼ ðym  yÞ2 dt
0 dedicated to bread baking process. The objectives are minimisation
of baking duration, of weight losses, of energy. Christensen et al.
Numerous weighting methods of the measurements are pro-
(1984) and Mälkki et al. (1984) propose an approach where the
posed. If the small values are important (in drying operation
optimum is selected among all the experimented behaviour.
for example), the use of logarithm is better. If numerous mea-
Paloheimo et al. (1984) develop a linear model (a RSM type model)
surements uncertainties are present, it is better to introduce
and proposes on this basis an improvement of the oven. More re-
weight of the measurement based on measurement biases. It is
cently, Therdthai et al. (2002) establish an optimal profile of bak-
important to emphasise that this point is a critical one often
ing, based on an average set point temperature value, distributed
neglected.
in the space of the oven. A relation permits to establish a link be-
tween these set points and the product quality variables. A model
3.3. Optimisation (recipe and operations) (linear multi regression model) is proposed in order to relate
weight losses and operating variable. By that way an optimisation
Optimisation is an important stakes for the food industries, be- is proposed.
cause of its wide scope of applications: economic, quality, envi- The use of a dynamic model of the biscuit baking process per-
ronment etc. Probably due to the complexity of most of the mits to develop original optimisation procedure. The profiles of
food unit operations and processes, the number of relevant appli- operating conditions are calculated in order to minimise the baking
cations is quite low compared to other process industries, such as time, or the total used energy. The objectives of the baking, from
chemistry (Olmos et al., 2003). The search of optimal set point the product point of view are keep as constants as possible. The re-
values is performed using several approaches mainly depending sults permit to save time and energy, and the trade off between the
on the complexity of the considered unit operation, and in fact two criteria is established too. It appears, too, that the design of the
the model. As for recipe optimisation, the first, and main ap- oven could be modified from the optimisation results, for example
proach is based on experimental design and response surface if the objective is to process using the shorter baking duration as
methodology applications (RSM). Nevertheless more accurate possible. The example of biscuit baking is a good illustration of
and reliable results are obtained when more appropriated meth- numerous food processes situation. The operating variables are
ods are used. Looking at the scientific literature, more than 32% distributed along the oven, the number of possible operating vari-
of the papers based on optimisation propose approaches without ables is important, and the behaviour is a dynamic, multi variables
any models. The other papers (58%) are based on empirical mod- and non-linear one. The number of possible applications of such
els, like RSM approaches. Than only a few number of papers are approach is wide, notably, when the control variables are distrib-
optimisation first principle models based. In some case these uted in space.
 G. Trystram / Journal of Food Engineering 110 (2012) 269–277 275

3.4. Control and related applications purposes 4.1. Molecular modelling

Control is one of the most complicated situations, mainly be- Twenty years ago, the first review on molecular modelling in
cause transient behaviour has to be faced. The lack of under- food research (Kumosinski et al., 1991a,b) proposed a modelling
standing is higher as for steady state modelling applications. that was mainly inspired by biology to solve complex structure
Nevertheless, numerous methods based on modelling are pro- relationships such as the influence of steric factors on encapsula-
posed. Part of the method imply the off line calculation of con- tion. Currently physicists (Vogiatzisa et al., 2010) are able to solve
trol algorithm from simulation based on models. Part of other such problems with an explicit model that combine objects of dif-
method, certainly the most interesting one imply the on line ferent scales. New promising simulations propose an atomistic to
implementation of models. The condition is than to be able to continuum simulation to reconstruct transport properties such as
run the simulation and associated algorithms in real time. Even heat transfer in nanostructured solids (Wagnera et al., 2008).
if most of the food operation are not critical for time constant Transport properties such as diffusion coefficients and their activa-
consideration, easy to run models are expected. Different ap- tion with temperature in polymers and liquids have been success-
proaches are proposed and seem to be valid. From existing first fully investigated with coarse-grained simulation and long-term
principle models, more and more reduction model approaches molecular dynamics simulation up to 40 ls (Durand et al., 2010).
are proposed (European CAFÉ programme in progress). The idea By accepting to forgive additional atomistic details, simulations
is to start from a complex model and to reduce its complexity in up to few milliseconds can be used to calculate transport proper-
such a way that the simulation becomes easy and run quickly as ties in nanostructured possibly random matrices (Vitrac and
possible. Theoretical methods are available (Leducq et al., 2006) Hayert, 2007). When atomistic details are important, different
for example in refrigeration processes). Other approaches pro- techniques combining a generalised Flory-Huggins approximation
pose to develop black Box models, directly from data or indi- of pair molecular interactions and a model of the entropy of mixing
rectly as a model of the theoretical model of operation. The can be used to predict excess chemical potentials in polymers (Gil-
most used method is based on Artificial Neural Networks. let et al., 2009) and in liquid mixtures simulating food (Gillet et al.,
Numerous applications are available and it becomes very easy 2010). Combined with molecular dynamics at mesoscopic scale,
to implement. The only condition is to have a database with en- similar approaches can be used to predict wetting properties and
ough experiments that permit identification of a reliable model. adsorption isotherms (Noro et al., 2003). To simulate the micro-
It must be take into account that this kind of modelling ap- scopic coupling between transport and chemical or biochemical
proach is only available in the range of variation of operating reactions both molecular and Langevin dynamics have been pro-
condition. The generic is generally poor. New in progress ap- posed (Dobrzynski et al., 2007; Alves et al., 2006). They have been
proaches proposes to design the model structure from the applied within an outstanding multiscale simulation to explain the
understanding of mechanisms and to describe the relationship neurotransmission at a neural synapse. The powerful of such tools
between key variables using different tools. Dynamic Bayesian is impressive. At least any key direction appears to be important,
networks appears efficient, fro example in cheese ageing process that deals with parameters reconstruction. Results are available
(Baudrit et al., 2010). that permits without any experiment to establish a reliable value
Probably promising tools becomes step by step available that for physical parameter (diffusion coefficient, activity coefficient,
permits to help for such modelling applications. for example). It is probably one of the most interesting areas of re-
A totally different modelling approach was developed from 10 search for the next 10 years.
years. The idea, for control or decision support application is not
to try to model the operation, but the interaction between oper- 4.2. Scale reconstruction
ation and people in charge of the manual control. The main
observation that found the approach was first based on the lack It is obvious that numerous works are discussing the impor-
of sensors and the fact that many peoples in charge of food tance of scale for food. Between molecular and macroscopic scale
operation control on plant use their senses to evaluate the prod- observation tools are available at each scale, and models becomes
uct evolution, and then take some decision in order to improve also available. Nevertheless it is not obvious to have a tool able to
the process efficiency. The other point was the lack of under- skip from one scale to another one. The stakes are great to be able
standing of the process and the assumption that people under- to analyse, using models, the different understanding that are per-
stood better as we are able to put in mathematical law. formed at each scale and to bridge the gap between scales. Fig. 3
Several applications are developed with success: baking (Perrot presents the complexity and the scales where some works are
et al.), dry sausage process (Curt et al., 2006), foaming (Allais done. The arrows represent some links that are challenges for re-
et al., 2007), Cheese ripening (Perrot et al., 2004), etc. In most search on Food modelling. Molecular modelling permits now at
of the case rule based fuzzy tools are used as modelling ap- least to obtain the physical parameters that are necessary for mes-
proaches, but it is obvious from our experience that numerous oscopic description. Even if data are still missing for mechanics,
tools are able to help for implementation of such models for more models propose integration at this mesoscopic scale. Valida-
decision support system design. New incoming tools from the tion using tomography tools is easier also. Development of reaction
complex system set of theory are proposed that seems to be rel- scheme modelling permits also a better integration and opens the
evant and useful even for food operations and processes (Baudrit door to toxicology and nutritional better understanding and
et al., 2010). description. Human biology models, mainly the ones that are un-
der development for GI tack, mouth, stomach are new field for
researchers and even if different approaches are under studies,
4. Stakes for modelling in food engineering progresses are going well. Global integration is not open today
but is on the way and probably some applications will start that
The purpose of the article is to discuss and illustrate the difficul- permit to describe the whole complexity for simple product, soon.
ties and the promising benefit of using model in the food engineer- On Fig. 3 the arrows are doubled sided because, certainly one of
ing area. Without any idea of being exhaustive, some key the challenge in the future will be to perform the way from con-
challenges can be discussed. Two points are proposed as a final dis- sumer expectation, or physiological effects, to the process and then
cussion. Both are related with scales consideration. to the recipe and the choice of Ingredients and raw materials. The
276 G. Trystram / Journal of Food Engineering 110 (2012) 269–277

challenge has to face and to build tools able to help us to deal with Fito, P., LeMaguer, M., Betoret, N., Fito, P.J., 2007. Advanced food process engineering
to model real foods and processes: The SAFES methodology. Journal of Food
uncertainties (Van Impe et al., 2001).
Engineering 83, 390–404.
Even if the challenge, for a scientific point of view is great to try Gillet, G., Vitrac, O., Desobry, S., 2009. Prediction of solute partition coefficients
to model at the finest accuracy as possible all the scale presented between polyolefins and alcohols using a generalized Flory–Huggins approach.
Fig. 3, it must be pointed out that for engineering considerations, Industrial and Engineering Chemistry Research 48 (11), 5285–5301.
Gillet, G., Vitrac, O., Desobry, S., 2010. Prediction of partition coefficients of plastic
a model has to fit the right level and the right accuracy correspond- additives between packaging materials and food simulants. Industrial and
ing to the way and the objective the model has to be used. Proba- Engineering Chemistry Research 49 (16), 7263–7280.
bly, despite a larger number of modelling applications, it is the Ioannou, I., Mauris, G., Trystram, G., Perrot, N., 2006. Back-propagation of
imprecision in a cheese ripening fuzzy model based on human sensory
main reasons why models are still not s used in Food engineering, evaluations. Fuzzy Sets and Systems 157, 1179–1187.
and specific works has to be done to develop appropriated models Kumosinski, T.F., Brown, E.M., Farrell Jr., H.M., 1991a. Molecular modeling in food
able to help specific engineering questions in the food domain. research: applications. Trends in Food Science and Technology 2, 190–193.
Kumosinski, T.F., Brown, E.M., Farrell Jr., H.M., 1991b. Molecular modeling in food
Modelling is certainly one of the main areas for research in food research: technology and techniques. Trends in Food Science and Technology 2,
engineering in the coming years. 110–115.
Laroche, B., Courtois, F., 2005. Testing structural identifiability of non-linear drying
model for rice. In: Proceedings of the 44th IEEE Conference on Decision and
Acknowledgments Control and the European Control Conference, December 12–15.
Leducq, D., Guilpart, J., Trystram, G., 2006. Non-linear predictive control of a vapour
Numerous works presented here and that permit to discuss the compression cycle. International Journal of Refrigeration 29 (5), 761–772.
Lee, J.H., Lee, I.Y., Kim, M.K., Park, Y.H., 1999. Optimal pH control of batch processes
modelling of food processes are performed in collaboration with for production of curdlan by Agrobacterium species. Journal of Industrial
colleagues: Profs B. Broyart, D. Flick, F. Courtois, C. Trelea, and Microbiology and Biotechnology 23, 143–148.
Doctor T. Lucas, O. Vitrac, C. Bonazzi, N. Perrot. Lemaire, R., Courtois, F., Trystram, G., Nicolaï, B., De Baerdemaeker, J. (Eds.), 1997.
Application of Identifiability Test to Drying Models. Presented at the COST-
915 Copernicus CIPA-CT94–0120 workshop on Food Quality Modelling,
References June 4–7. Office for Official Publications of the European Commission,
Luxembourg.
Mälkki, Y., Seibel, W., Skjöldebrand, C., Rask, Ö., 1984. Optimization of the baking
Achir, N., Vitrac, O., Trystram, G., 2008. Simulation and ability to control the surface
process and its influence on bread quality. In: Zeuthen, P. (Ed.), Thermal
thermal history and reactions during deep fat frying. Chemical Engineering and
Processing and Quality of Foods. Applied Science, London, pp. 55–361.
Processing 47, 1953–1967.
Mirade, P.S., Daudin, J.D., Ducept, F., Trystram, G., Clément, J., 2004. Characterization
Aït Ameur, L., Trystram, G., Birlouez-Aragon, I., 2006. Accumulation of 5-
and CFD modelling of air temperature and velocity profiles in an industrial
hydroxymethyl-2-furfural in cookies during the backing process: validation of
biscuit baking tunnel oven. Food Research International 37 (10), 1031–1039.
an extraction method. Food Chemistry 98 (4), 790–796.
Mondal, A., Datta, A.K., 2008. Bread baking: a review. Journal of Food Engineering 86
Allais, I., Perrot, N., Curt, C., Trystram, G., 2007. Modelling the operator know-how to
(4), 465–474.
control sensory quality in traditional processes. Journal of Food Engineering 83
Morimoto, T., Purwanto, W., Suzuki, J., Hashimoto, Y., 1997. Optimization of heat
(2), 156–166.
treatment for fruit during storage using neural networks and genetic
Alves, R., Antunes, F., Salvador, A., 2006. Tools for kinetic modeling of biochemical
algorithms. Computers and Electronics in Agriculture 19 (1), 87–101.
networks. Nature Biotechnology 24, 667–672.
Noro, M.G., Paul, P.K.C., Warren, P.B., 2003. Linking atomistic and mesoscale
Balsa-Canto, E., Alonso, A.A., Banga, J.R., 2002. A novel, efficient and reliable method
simulations of water soluble polymers. Journal of the American Chemical
for thermal process design and optimization (Part I: Theory). Journal of Food
Society 125, 7190–7191.
Engineering 52, 227–234.
Noronha, J., Van Loey, A., Hendickx, M., Tobback, P., 1996. Simultaneous
Banga, J.R., Perez, M.R.I., Gallardo, J.M., Casares, J.J., 1991. Optimization of the
optimisation of surface quality during the sterilisation of packed foods using
thermal processing of conduction-heated canned foods: study of several
constant and variable retort temperature profiles. Journal of Food Engineering
objective functions. Journal of Food Engineering 14 (1), 25–51.
30 (3-4), 283–297.
Banga, J.R., Balsa-Canto, E., Moles, C.G., Alonso, A.A., 2003. Improving food
Olmos, Tréléa, I.C., Courtois, F., Bonazzi, C., Trystram, G., 2002. Dynamic optimal
processing using modern optimization methods. Trends in Food Science and
control of batch rice during process. Drying Technology 20 (7), 1319–1345.
Technology 14 (4), 131–144.
Paloheimo, M., Mälkki, Y., Kaijaluouto, S., 1984. Optimization of baking process:
Baudrit, C., Sicard, M., Wuillemin, P.H., Perrot, N., 2010. Towards a global modelling
combined heat transfer. In: Zeuthen, P. et al. (Eds.), Thermal Processing and
of the Camembert-type cheese ripening process by coupling heterogeneous
Quality of Foods. Applied Science, London, pp. 487–492.
knowledge with dynamic Bayesian networks. Journal of Food Engineering 98
Perrot, N., Agioux, L., Ioannou, I., Mauris, G., Corrieu, G., Trystram, G., 2004. Decision
(3), 283–293.
support system design using the operator skill to control cheese ripening.
Bimbenet, J.J., Schubert, H., Trystram, G., 2007. Advances in research in food process
Journal of Food Engineering 64, 321–333.
engineering as presented at ICEF9. Journal of Food Engineering 78, 390–404.
Perrot, N., Ioannou, I., Allais, I., Curt, C., Hossenlopp, J., Trystram, G., 2006. Fuzzy
Broyart, B., 1998. Modélisation des phénomènes de transferts et des modifications
concepts applied to food product quality control: a review. Fuzzy Sets and
de qualité induites lors de la cuisson d’un biscuit sec en continu. Thèse Génie
Systems 157, 1145–1154.
des Procédés. Ecole Nationale Supérieure des Industries Agricoles et
Sablani, S., Datta, A.K., Rahman, M.S., Mujumdar, A.S., 2007. In: Mujumdar, A., Datta,
Alimentaires, Massy, France.
A., Sablany, S., Rahman, S. (Eds.), Handbook of Food and Bioprocess Modelling
Broyart, B., Trystram, G., 2003. Modelling of heat and mass transfer phenomena and
Techniques. CRC Press, Bocca Raton Fla.
quality changes during continuous baking using both deductive and inductive
Samir, K., Patrick, L., Xavier, J., Martine D., 2009. Optimisation of a neutral spirit
(neural network) modelling principles. Food and Bioproducts Processing 81,
production plant. Récents Progrès en Génie des Procédés, Numéro 98–2009, 2-
316–326.
910239-72-1, Ed. SFGP, Paris, France.
Chen, C.R., Ramaswamy, H.S., 2002. Modelling and optimization of variable retort
Terajima, Y., Nonaka, Y., 1996. Retort temperature profile for optimum quality
temperature (VRT) thermal processing using coupled neural networks and
during conduction-heating of foods in retortable pouches. Journal of Food
genetic algorithms. Journal of Food Engineering 53 (3), 209–220.
Science 61 (4).
Christensen, A., Blomqvist, I., Skjöldebrand, C., 1984. Optimization of the baking
Therdthai, N., Zhhou, W., Adamczac, T., 2002. Optimisation of the temperature
process (convection ovens). In: Zeuthen, P., et al. (Eds.), Thermal Processing and
profile in bread baking. Journal of Food Engineering 55 (1), 41–48.
Quality of Foods. Applied Science. London, pp. 482–486.
Trystram, G., Courtois, F., 1996. Food process modelling and simulation. In: Mittal,
Datta, A.K., 2008. Status of physics-based models in the design of food products,
G., Dekker, M. (Eds.), Computerized Control Systems in the Food Industry.
processes, and equipment. Comprehensive Reviews in Food Science and Food
Marcel Dekker, New York, pp. 55–85.
Safety 7 (1), 121–129.
Vajda, K.R., Godfrey, H., Ratitz, 1989. Similarity transformation approach to
De Blank, H., Hendrix, E., Litjens, M., Van Maaren, H., 1997. On-line control and
identifiability analysis of nonlinear compartmental models. Mathematical
optimisation of the pelleting process of animal. Journal of the Science of Food
Biosciences 93 (2), 217–248.
and Agriculture 74 (1), 13–19.
Van Impe, J.F., Bernaerts, K., Geeraerd, A.H., Poschet, F., Versyck, K.J., 2001.
Dobrzynski, M., Rodríguez, J.V., Kaandorp, K.A., Blom, J.G., 2007. Computational
Modelling and prediction in an uncertain environment. In: Tijskens, L.M.N.,
methods for diffusion-influenced biochemical reactions. Bioinformatics 23,
Hertog, M.L.A.T.M., Nicolaï, B.M. (Eds.), Food Process Modelling. CRC Press, Boca
1969–1977.
Raton, USA.
Dochain, D., Vanrolleghem, P.A., Van Daele, M., 1995. Structural identifiability of
Vitrac, O., Hayert, M., 2007. Effect of the distribution of sorption sites on transport
biokinetic models of activated sludge respiration. Water Research 29 (11),
diffusivities: a contribution to the transport of medium-weight-molecules in
2571–2578.
polymeric materials. Chemical Engineering Science 62 (9), 2503–2521.
Durand, M., Meyer, H., Benzerara, O., Baschnagel, J., Vitrac, O., 2010. Molecular
Vogiatzisa, G.G., Voyiatzisa, E., Theodorou, D.N., 2010. Monte Carlo simulations of a
dynamics simulations of the chain dynamics in monodisperse oligomer melts
coarse grained model for an athermal all-polystyrene nanocomposite system.
and of the oligomer tracer diffusion in an entangled polymer matrix. Journal of
European Polymer Journal. doi:10.1016/j.eurpolymj.2010.09.01.
Chemical Physics 132 (19).
 G. Trystram / Journal of Food Engineering 110 (2012) 269–277 277

Wagner, M., Lucas, T., Le Ray, D., Trystram, G., 2007. Water transport in bread during Walter, E., Pronzato, L., 1997. Identification of Parametric Models from
backing. Journal of Food Engineering 78 (4), 1167–1173. Experimental Data, Vol. XVIII. Springer-Verlag, Heidelberg, 413 pp.
Wagnera, G.J., Jonesb, R.E., Templetona, J.A., Parksc, M.L., 2008. An atomistic-to- Wong, S.Y., Zhou, W.B., Hua, J.S., 2007. CFD modelling of an industrial continuous
continuum coupling method for heat transfer in solid (2008). Computer bread-baking process involving U-movement. Journal of Food Engineering 78,
Methods in Applied Mechanics and Engineering 197 (41–42), 3351–3365. 888–896.