Professional Documents
Culture Documents
net/publication/289816884
CITATIONS READS
0 310
5 authors, including:
Some of the authors of this publication are also working on these related projects:
OPEN-GAIN : Optimal Engineering Design for Dependable Water and Power Generation in Remote Areas Using Renewable Energies and Intelligent Automation View
project
All content following this page was uploaded by C. De Prada on 24 March 2020.
Resumen
A dynamic model representing the interface between the evaporation station and the first stage of the crystallization section of a
typical beet sugar factory is described. The simulation represents a genuine control problem that requires the coordination of the
continuous evaporation section with the downstream semi-batch crystallizers. The model is intended to be used as a benchmark for
trying out different optimal plantwide control strategies. The benchmark is particularly challenging due to the close connection between
the continuous and the batch operated section, with multiple interchanges of process streams and energy. The plantwide control sugar
factory case study is offered in the form of three variants of increasing complexity and realism in order to be of interest to a wide
audience, ranging from the food and specifically the sugar process specialists at one end of the spectrum, to the general process control
and automation community at the other.
Copyright © XXXX CEA. Publicado por Elsevier España, S.L. Todos los derechos reservados.
Keywords:
Food processing, process control, plantwide control, optimal control, sugar processing
1. Introducción between the ERP and Automation levels and deals with the
allocation of needed resources for complying with the plan. For
The management of complex, large-scale systems, possibly example, in those cases where there is batch type of equipment, is
geographically dispersed, in a rapidly changing dynamic also in charge of short term scheduling tasks on those units. The
environment, trying to take the best decisions, in a robust way, automation layer, on its turn, has is own piramidal structure that is
that may include the system self-configuration, in the presence of going to be discussed with some detail later in the article, but that
the unavoidable uncertainties, constitutes a formidable task. At basically consists of keeping the value of some selected variables
the present moment, however, the current advances in technology close to their references, guaranteeing good dynamic performance
resulting in the ubiquitous presence of information generating, of the controlled system. The issue of how to select the variables
processing and networking capabilities embedded at every level to control in a plant-wide setup has profound implications and
of the factory, allows, at least, to ambition this possibility. The there is a large literature dealing with it. The analysis of each
achievement of these goals requires nothing less than a new particular problem usually indicates the need of regulating some
industrial revolution, refered to with the tag of Industry 4.0 or process variables to fixed references for reasons having to do with
smart factories. It seems to demand a new, more encompassing the quality of production, the technological particularities of the
paradigm, in which the plant, the intelligent sensing, actuating individual units, or due to safety and environmental concerns.
and controlling devices, the role of the communication network But once this basic control structure has been selected, it makes
and of the involved software are considered and studied as an sense to choose the degrees of freedom that typically remains in
integrated whole. It is this context that terms such as such a way as to optimize the economic performance of the whole
cyberphysical systems, Internet of Things (IoT) and big data are plant. The described approach obviously depends on the
gaining considerable attention. availability of the adequate of technical infrastructure: sensor and
The smart factory concept will not be fully achieved overnight actuator devices, different kinds of specialized automation
but incrementally and building on the already existing important equipment and software deployed at the different levels: from
advances in process control and supervision platforms. In the PLCs dealing directly with the regulation and sequential control
most advanced process industries (PI), including food processing tasks, to SCADA type of supervisory systems. For more advanced
industries (FPI), the state of the art situation should be understood multivariable control and for optimization the existence of
in relation with the control integrated manufacturing model (CIM) accurate models of the processes is also a fundamental
that the ISA-95 standard tries to codify (falta fig. 1). The levels 0, requirement.
1, 2 are directly related with the regulation and automation of the The above described situation is not yet the current practice in
plant. Level 4 (ERP) deals with long term planning of production, every factory. The typical situation of FPI is even less favorable.
while the intermediate level 3 MES layer, not fully developed in For example, and to mention only two recognized shortcomings:
most current implementations, is in charge of bridging the gap the lack of enough sensors, especially those measuring quality
Mazaeda et al. / Revista Iberoamericana de Automática e Informática industrial 00 (2015) 1–4
related information and the absence of accurate enough first The juice delivered from the diffusion section is not a pure
principles models for the underlaying processes. solution of sucrose in water. It is a technical solution that also
This way of doing things, recommended by ISA-95 standard, contains a great number of other inorganic and organic
is sound but the advocates of the Industry 4.0 paradigm claim for compounds. The existence of these impurities or non-sugar
a more flexible and horizontal kind of approach in which the components, the exact content of which varies from factory to
existence of digital processing capabilities embedded in field factory, is an unavoidable technological inconvenience. The need
devices translates into a distribution of the decision making and to handle their presence and the need to reduce their negative
control of the different sub-plants or units. impact on the efficiency of the factory, decides in good measure,
The network of excellence entitled “Highly-complex and its design and operation. For example, there is a whole chemical
Networked Control Systems (HYCON2)” funded by European plant, the purification section, dedicated to increase the purity of
Union Seventh Framework Program has united the efforts of diffusion juice. But the purification section cannot remove all
academic and industrial partners to advance the research in the impurities and their unavoidable presence in crystallization has an
fields above briefly discussed over different application domains. important negative impact that has to be dealt with.
An important cornerstone of HYCON2 has been the use of As will be clear at the end of this section, the reduction of the
benchmarks or case studies for testing the proposed solutions and impurities adverse consequences determines from the design and
stimulating the development of best practices. Two benchmarks management of the individual units (crystallizers and centrifuges)
were officially adopted, one having to do with vehicles to the large scale topology of the department that is typically built
transportation over highways and their control. The other case around the use three crystallization-separation stages.
study proposed belongs to the FPI and constitutes a plantwide Each of these stages, although structurally and functionally
model of the relation of the evaporation and crystallization similar, has assigned a different role: the first one is concerned
sections of the beet sugar industry whose description is the with the production of the commercial sugar grain, while the
purpose of the present contribution. following two are in charge of the economic efficiency of the
It is important to understand that the sugar case study here factory, by recovering from their incoming syrups as much
proposed is not conceived as a kind of simple, toy example, to sucrose as possible to be recycled to the first stage.
specifically show the virtues of a given approach to solve some
Recovered
particular problem. This type of benchmark is legitimate and has sugar
the merit of being easily undertood and reproducible for the
research community it is aimed at. In the HYCON2 sugar case 1º Stage
study the approach has been different. Here the model represents water
Boiler
melter
steam
a simplified but rich enough version of a genuine problem to be
Feed licor I tank
found in sugar process factories and it was from the beginning an
Syrup or ...
Evaporation
Purification
thick juice
discussed nature of project, not to hide the intrinsic complexity of massecuite
water
Steam from
the problem. In any case, the authors were very conscious of the evaporators Rich syrup I White
sugar
Poor
heterogeneity of the expected audience: general knowledge of the syrup I
process industries, although not especially of food or sugar Beet end 2º Stage
factories, but with a clear automation and control background. So, Rich syrup II Sugar II
Benchmark model
to ease the comprehension, the case study has been offered in the Poor
syrup II
3º Stage
form of three versions of increasing complexity.
The paper is organized as follows: in section 2 a general Sugar III
are also included. Correspondingly, the term purity is applied to applied to the shell of a heat exchanger or calandria incorporated
the mass ratio of sucrose to impurities for the solutions, but when to the unit. When the right super-saturation is reached, the seed, a
reference is made to the whole massecuite the sugar grains mass population of sugar crystals with average size of around 10 µm, is
is added in the numerator. introduced. Crystallization then proceeds along the lengthier and
most important growing of the grain step. As this process unfolds,
(-) Purity the purity of the solution, now refered to as mother liquor,
σ*
Concentration
Condenser
ancillary equipment, namely open and closed tanks, heat Wvap_in
Lls Feed liquor tank
exchangers, PID controllers, etc. which can be conveniently
instantiated for describing the specific dimensions, geometry or
PF1 PF2 PF3
any other parameters of the specific process units. Components Load1 Load2 Load3
components used are a simplified variant of those developed for a time (s) time (s)
library of units of the sugar industry conceived specifically for the
creation of plant wide simulation models to train control room Figure 6: Cycle duration depends on syrup’s a) concentration b) purity
operators. The developed OO libraries were necessarily dynamic In fig. 6 it is shown how a higher concentration of feed liquor
and in this sense their purpose and realm of application is very implies a shorter cycle if all other conditions are kept fixed. These
different from other simulation tools which also target the same other conditions are the purity of the feed liquor, the
industry such as SUGARSTM [][] more adequate for design and concentration versus level curve that controls the flow of syrup
stationary analysis. into the chamber and the rate of water evaporation. The reason
behind the observed acceleration is that, for a higher
3.1.1 Variant I concentration of the feed, the concentrations stipulated by
concentration versus level curve that regulates the crystal growing
The first, simpler variant of the case study (fig. 5) concentrates phase are reached at earlier, so the syrup intake is also
in testing the correct scheduling of three crystallizers connected in augmented. Figure .b shows a similar effect on the duration of the
parallel so that they keep up with the syrup entering the batch of a higher feed syrup purity. This acceleration is
department, or put equivalently, to consume the right amount of attributable in this case to an increase in crystallization speed.
feed liquor from the corresponding tank so as to keep its level So, the cycle duration of each unit, and consequentially the
between safe limits. This task must be performed in the presence combined ability of the battery of the three crystallizers for
of the disturbances represented by the flow of syrup entering, its consuming the arriving syrup at a higher rate depend on its
concentration and its purity and acknowledging the restriction characteristics. But, obviously, this duration can also be
imposed by the limited availability of heating steam. influenced by modifying the operation of the unit. For example, a
higher water evaporation rate would have similar consequences as
the above discussed for the case of more concentrated feed syrup.
This can be achieved by increasing the temperature difference at
both sides of the heat exchanger, or equivalently by increasing the
pressure at which the condensation of the heating steam is carried
out in the primary of the calandria or the by lowering the vacuum
pressure inside the chamber and thus the boiling temperature or,
obviously, by a combination of both actions.
Mazaeda et al. / Revista Iberoamericana de Automática e Informática industrial 00 (2015) 1–4
Pcal PID recipe setpoint juice (W jc ), the crystals seed (W seed ), the evaporated water (W evap )
and of massecuite (W mc ) output. Notice that the existence and the
PF=1.2 magnitude of the mentioned flowrates depend, in general, of the
state that the unit’s program is in. The seed flowrate, for example,
PF=1.0
Pcal (kg/s)
is only active during the short period of seeding while the output
PF=0.8 of massecuite is different from zero only in the discharge state.
The stream W cris represents the sucrose that goes from the
(a) dissolved to the crystallized phase. Notice that all impurities
remain dissolved so we are disregarding phenomena like
PF=1.2 Super-saturation inclusions of mother liquor inside the crystal structure, for
PF=1.0 example.
PF=0.8 dm (1)
= w jc + wseed − wevap − wmc
dt
s
dmsuc (2)
= w jc _ suc − wcris − wmc _ sac
(b) dt
dmimp
Time (s) = w jc _ imp − wmc _ imp (3)
dt
Figure 7: Cycle duration can be influenced by a) P F factor b) with an impact dmcris
on super-saturation. = wseed + wcris − wmc _ cris (4)
dt
In the benchmark, it is recommended to keep the recipe based mw = m − msuc − mimp − mcris (5)
controller in place in the understanding that it gives a reasonable
dmvap
performance. To make the problem more flexible, an extra degree = wevap − wvap _ cond (6)
dt
of freedom for increasing or shortening the duration of the cycle
of each crystallizer, is provided. This is achieved by multiplying
the profile of the set point of the PID controlling the heating Vapor and massecuite share the whole volume (m3) of the unit
steam pressure that the unit’s program proposes, by a continuous according to the constitutive relation (7). In that expression V cham
factor P Fi {i=1, 2, 3}, which is centered at the value of one (fig. is the maximum volume destined to contain the massecuite, V dome
7.a). It has to be taken into account, however, that too much is the space reserved exclusively for vapor and V vap the volume
acceleration (P F > 1) can be detrimental to the quality of the which is effectively taken by the vapor phase along the strike.
crystals. A higher evaporation rate determines an increase of the
m (7)
supersaturation in the solution with the associated risk in Vvap = Vcham + Vdome −
widening the CSD (fig. 7.b). This comes about because the ρ mc
kinetics of sugar growth depends on super-saturation but also on At the beginning of the cycle, the sensible heat delivered to the
other factors such as the purity, for example. This might imply mass is used to increase its temperature. Once the boiling
that the the acceleration of the process must deal with a total mass temperature is reached, however, water starts to evaporate. From
of sugar crystals whose aggregate surface in not capable of that moment on, we assume that the relation (8) holds. The
incorporating the dissolved sucrose at the required rate. parameter k evap is pragmatically chosen to be high enough so as to
Another element that has to be taken into account is the guarantee the practical equality of the pressure in vapor space
limited availability of heating steam, a resource which is provided (P vap ) and the water equilibrium pressure in the solution (P eq ) but
by the evaporation station and that has to be shared by the three no to high as to provoke difficulties by increasing, unwarrantly,
consumers. This fact is represented in this reduced model by a the stiffness of the numerical problem [Luyben]. The pressure
static component that represent a real steam source in the sense P vap is obtained by applying the ideal gas equation of state,
that the pressure starts to drop, in this case linearly, when the considered to be valid due to the low pressures at which the
extracted flow grows beyond a certain value. process is conducted. On the other hand, P eq depends on the
The most important unit here is the crystallizer. We have used temperature but also on the presence of dissolved substances in
a simplified version where only the elements concerning the right the mother liquor and it is inferred from known boiling point
evolution of the consumption of steam and syrup and its elevation tables.
dependence on the characteristics of the juice and on the pressure
of heating steam are intended to be captured. (
wevap = k evap Peq − Pvap ) (8)
Some key elements of the the crystallizer’s model are
succinctly described next. Notice that the model has embedded nRTvap
the discrete finite state machine driving the stage by stage Pvap = (9)
Vvap
evolution of the control program of the unit according to the
chosen recipe. The first principles mathematical model of each
crystallizer consists of material balances preformed to the overall Temperature in the vapor space (T vap in ºC) is determined by
mass (m) and to each of its components: sucrose (m suc ), no-sugar (10). The specific enthapy h evap (J/kg) which is a function of both:
(m imp ), water (m w ) and, after the seeding stage have taken place, the temperature of the boiling mass and of the pressure. Notice
of the crystallized sugar (m cris ) as in (1-5). There is also the need that this steam is overheated due to the mentioned boiling point
of keeping track of the vapor mass on top (6). In both cases we operation effect. The W vap_cond term represents the flowrate of
are considering perfectly mixed contents. All the flowrates are steam leaving the unit to the barometric condenser. It has a
expressed in kg/s and represent those corresponding to the feed
Mazaeda et al. / Revista Iberoamericana de Automática e Informática industrial 00 (2015) 1–4
specific enthalpy h vap . The term C p (J/kg.ºC) is the specific heat factors conspiring for the reduction of this coefficient, namely the
capacity. reduction of the circulation of the mass inside the chamber as the
dmvap effective viscosity of the mass increases and the natural
dTvap wevap hevap − wvap _ cond hvap − hvap
= dt (10) convection provoked by the evaporation bubbling diminishes, a
dt mvap c p fact that cannot be completely compensated for by the existence
dm of the mechanical stirrer. So in expression (15) U is is considered
w jc h jc + Q − wevap hevap − wmc hmc − hmc to depend, first on the existence or not of evaporation in the
dhmc dt (11)
=
dt m chamber. For the sensible heating phase, the coefficient is given a
constant value U sen to be adjusted for each case. In the latent heat
The dynamic evolution of the massecuite specific enthalpy phase, the value of U is made to match a regression expression
(h mc ) is given in (11). In this case h jc is the specific enthalpy of given by (16) as proposed in [Rouillard] which gives the flux of
the incoming juice which depends on its temperature and steam for unit area of the exchanger J evap (kg/hm2) as function of
composition and Q (J/s) is the heat flow provided by the the concentration of the massecuite (B mc ), its purity (P mc ), the
exchanger. The latter is considered to comply with (12) pressure in the vapor space in the chamber (P vap ) and on the
depending on the overall heat transfer coefficient U (J/m2sºC), the primary of the exchanger (P cal ), the height of the exchanger tubes
total area available for the exchange A (m2) and the temperature (L tub ), and the height of the mass over the upper rim of those
difference at both sides of the exchanger wall. The term T mc is the tubes (L ec ) which directly proportional to the level attained by the
temperature of the massecuite obtained from h mc as a function of massecuite in the chamber.
the composition of the mother liquor and the amount of embedded Due to this evolution of U, and also to the changes in the
crystals assuming an ideal mix. The temperature T cal is profile of the setpoint for the pressures in the chamber and on the
considered to be the saturation temperature of the heating steam primary of the exchanger, the consumption of heating steam
entering the primary of the exchanger which corresponds to the varies along the cycle as shown in fig. 7.a. This profile determines
pressure P cal . We are assuming that there are not important energy a similar shape for the consumption of syrup as determined by the
losses to the enviromment and that the energy exchange between massecuite concentration control scheme already explained
the vapor and the massecuite phase in dominated by the fig.7.b. The evolution of these two flowrates, which describes
convection effect represent by W evap . completely the interface of the crystalizer with the providers of
heating steam and of the feeding liquor whose correct description
Q = U ⋅ A ⋅ ∆T = U ⋅ A ⋅ (Tcal − Tmc ) (12) is the main purpose of this variant of the benchmark, is decided
by the discussed evolution of the heat transfer coefficient and
dmcal ultimately by the characterisitics of the feeding liquour and the
= wcal _ in − wcal _ out (13) control policy implemented in the crystallizer.
dt
Q The obtention of the profiles in fig. 7 do not require a very
wcal _ out = (14)
lat ( Pcal ) exhaustive description of the crystal size distribution. The only
interest here is in modelling, as faithfully as possible, the overall
evolution of the mass of crystals m cris and correspondingly the
The determination of P cal is done by an equation of state
mass flowrate w cris representing crystallization.
similar to (9). The volume is calculated taking into consideration
the mass of vapor m cal (kg), considered as saturated for
3 ⋅ (s − 1)α + w
2
convenience, whose evolution is given by []. In this last equation, wcris = k cris ⋅ mcris seed
(17)
W cal_in is the flowrate of heating vapor entering the exchanger,
ultimately depending of the heating vapor pressure control loop,
while W cal_out is the flow rate of condensed water going out. The
latter is determined as a function of Q and the latent heat of Heating steam
condensation of steam at the pressure P cal as in [].
Wcal (kg/s)
PF=1.2
PFactor=1.2
U sen Peq < Pvap (a) P =1.0
PFactor=1.0
F
(15)
U = PF=0.8
PFactor=0.8
(a)
lat (Pcal )
min ku ⋅ J evap ⋅ , U max Peq ≥ Pvap (b)
∆T
Feed licor
15.92 − 0.165 ⋅ Bmc − 6.01 ⋅ Pvap + 0.0311 ⋅ Pmc + (16) PF=1.2
J evap = exp
Wsyp (kg/s)
+ 0.639 ⋅ Pcal − 0.321 ⋅ Ltub − 0.298 ⋅ Lec
In (17) w cris is made to depend, obviously, on the seed flow The second variant introduces a more realistic model of the
over the short period it is active but fundamentally on the already evaporation station, while keeping almost intact the already
growing mass of existing crystals and on the excess of super- explained representation of the crystallizers section. The
saturation over unity, raised to some power α. The simplified evaporation section is represented, however, for the sake of
model (17) is similar to the one proposed in [Lauret]. The simplicity, by only the last three evaporation effects. This reduced
parameters α and k cris where chosen to approximate the results of model is able now to represent the fact that the flow and
a validated model with a more detailed description of the concentration of the syrup at the output of evaporation and the
crystallization kinetics. The super-saturation is defined in (18), as availability of steam for the crystallizers are inextricably
the ratio between the actual sucrose concentration (C suc ) with interrelated with each other and with the heating steam demand of
respect to the saturation concentration in the impure, technical each crystallizer. The workings of the plant as a whole depends
solution that the mother liquor represents (C suc,i ). Notice that the on the new disturbances, namely, the flow, concentration and
latter depends on the saturation concentration of pure sucrose purity of the thin juice coming into the first effect.
solution (C suc_sat,p ) which is only a function of the temperature and In this variant, the plant wide problem includes more degrees
of a saturation coefficient y sat . The determination of y sat can be of freedom. There are two new continuous decision variables: the
approximated by an expression as the one shown (19), where set point of the PID deciding the concentration of the syrup at the
q imp/w is the mass ratio of impurities to water and A, B and C are input of the feed liquor tank (BxC), and the value of the heating
parameters that should be adjusted for each specific factory. pressure to the first effect (P vap ). There is now also the possibility
of dynamically deciding, for each crystallizer, the source of the
csuc C suc (18) heating steam: the first or the second evaporation effect. This
s= =
csuc _ sat , i y sat ⋅ C suc _ sat . p (Tmc ) implies the existence of three new discrete decision variables
(SW i ).
The model of each evaporation unit has been radically
y sat = A ⋅ qimp / W + B + (1 − B ) ⋅ e
C ⋅ q imp / W
(19)
simplified from the one described in [Merino]. In order to
perform the continuous evaporation of a part of the incoming
With the above, this somewhat cursory description of the most juice, each evaporation unit requires energy which is provided by
complex unit of this variant of the benchmark is concluded. Is is the latent heat of condensation of the steam fed at the primary of
important to highlight that expressions and tables for general an incorporated heat exchanger not too dissimilar to the one used
properties of utilities like water and steam and others specifically in the crystallizers. The heating steam condenses in the exterior
belonging to the sugar industry such as density, specific shell of the exchanger while the juice to concentrate ascends in
enthalpies, heat capacities, boiling point elevation, etc. for the interior of the tubes. Part of the water evaporates from the
product streams like massecuite, juice and sugar crystal as a boiling juice, and ascends to the top of the chamber; while the
function of their compositions among other factors such as remaining more concentrated juice goes over the upper rim of
temperature, had been taken from the excellent compilation due to tubes to be collected in the central down take and from there to
[Bubnik and co-authors]. There are of course other units, most the output of the unit. The level here is kept constant to keep the
notably the feed liqour tank and the very simplified real heating vapors sealed at the top of the chamber.
steam source. The model of the tank is quite simple. It is an open
tank that receives the flow of thick juice and then delivers it at the
request of each individual crystallizer. The model should describe
the time evolution of the overall mass and of the sucrose,
impurities and water components in a way not to dissimilar to the
one <lready described for the dystallizer but with the obvious
simplification of not considering crystallization. Similarly the
energy balance is a simplified version of (11) without the
crystallization effect introduced by the enthalpy of massecuite
(here not existent) and the assumption of zero heat losses. On the
other hand the heating steam supply source pressure (P vap_t ) is
modeled according to (20): if the total demanded flowrate of
steam is less than certain parameter (W knee ) the deliverd pressure
is considered fixed at P vap_t,0 ; but if that flow limit is exceeded,
P vap starts to decrease with a negative slope (slp).
3.1.2 Variant II
Mazaeda et al. / Revista Iberoamericana de Automática e Informática industrial 00 (2015) 1–4
(a) Qe (24)
Steam I to
crystallizers
Steam II to
crystallizers wcal _ e =
PvapI PvapII
c _ lat ( Pcal _ e )
PC SP
wvapI wvapII
Steam from Notice that the use of II effect steam is preferably from the
Boiler
wvap_in point of view of the reutilization of energy. But, of course, the
Pvap_in
Effect Effect Effect
steam pressure available now is lower and this fact might be not
Bx SP
Juice I II III Condenser
BxC compatible with the required rate of evaporation in the crystallizer
Wjuc
Purjuc LC LC
LC Concentrated in order to keep up with the amount of incoming syrup.
Syrup to crystallizers
Bxjuc
3.1.3 Variant III
(b)
Manipulated var.
The more visible addition that the third variant of the
Disturbances
Measurements
benchmark exhibits in relation to the previous two is the
Concentrated syrup
from evaporation station incorporation of some units downstream the battery of
crystallizers, the more important of which is the component
representing the array of centrifuges. As already discussed, the
Condenser centrifuges separates sugar crystals from mother liquor; and uses
Lls Feed licor tank
some amount of water to guarantee the required purity of the
commercial product. The rich syrup coming out of the centrifuges
PF1 Load1 PF2 Load2 PF3 Load3 is recycled to be mixed with the thick juice thus forming the feed
SPpcal1 SPpcal2 SPpcal3
liquor of the crystallizers. Here, the water flow can be considered
*
(recipe)
(recipe)
Steam I
(recipe)
as an important new disturbance input that complicates the
Steam I Steam I
SW1
Crys.
I SW2
Crys.
II SW3
Crys.
III handling of the plant wide situation. More water implies a greater
flow to recycle, with more purity but also more diluted. This flow
is introduced downstream the evaporation, so the extra water has
Steam II Steam II Steam II
to be handled entirely in the crystallizers, with a greater
evaporation effort. Note that there also exists the possibility of
adding an extra flow of sugar representing the recovered sucrose
Figure 8: Benchmark variant II flowsheet.
from the not modeled second a third stage of sugar end. Notice
The model of each evaporator effect has been kept as simple as that in the real factory, the rich syrup and the recovered sugar
possible. In this case, we assume, for example that the mass of crystals are fed to a previous unit, the melter, designed to carry
juice inside the chamber is constant, and this give the following out the dissolution of the recycled grains. This extra unit is not
simple static relation between the incoming (w juc_in ), the outgoing represented here to make things simpler.
juice (w juc_out ) and the evaporation flowrates (w e_vap ): Between the crystallizer and the centrifuges there is the need
to place two new open tanks. The first one, the strike receiver, has
w juc _ out = w juc _ in − we _ vap (21) the purpose of acting as buffer to accommodate the supply rate of
massecuite by the crystallizers with its consumption by the
centrifuges, so its level is allowed to move freely. The second
We ignore also the dynamic effects on the relative content of unit, the channel, is level controlled by a PID that manipulates the
dissolved sugar in the juice, considering the purity of the juice aperture of a valve at its input, and its main purpose is to keep
input as equal to the purity of the output. In what respect to the smooth supplying conditions to the downstream centrifuges. In
dry matter content, its dynamics is essential because it is directly the real factory, the centrifuges are organized as an array of
related to description of the ability of the unit to provide steam to several batch units, working with a cycle which is many times
its consumers: faster than the cycle of crystallizers. A very detailed description is
given in [Mazaeda]. The consumption of the centrifuge is thus
dCdm w juc _ in .Cdm _ juc _ in − w juc _ out ⋅ Cdm (22) regulated by coordinating the scheduling of these batch units,
=
dt m including the turning on and off of some of them. For the purpose
of the benchmarks this extra complexity has been hidden. The
The energy is also essential: array of batch centrifuges has been substituted by a single static
component, and the input of massecuite is supposed to be
dh w juc _ in .h juc _ in − w juc _ out ⋅ h − we _ evap ⋅ he _ evap + Qe (23) regulated by the aperture of a valve (Ap_Cent), that should be
=
dt m manipulated as part of the plant wide control scheme, to
ultimately keep the level of the strike-receiver between safe
The pressure in the vapor space is directly considered as equal limits.
to the equilibrium pressure in the juice taking into consideration (Here de description of the centrifuge static component)
the boiling point elevation effect as function of dry matter content The crystallizer model used in this III variant is compatible
and temperature. However the model is simplified by considering with the ones used in the two previous cases in what concerns to
the evaporated steam as saturated instead of overheated as is the interface with the feed liquor and heating stream provided at
really the case. the input. However, in this case, it has been considered useful to
The vapor which is drawn in from the heating source is give a more detailed account of the growing of the crystals using
determined as; the traditional Population Balance Equation (PBE) approach,
represented by the partial differential equation () which describes
Mazaeda et al. / Revista Iberoamericana de Automática e Informática industrial 00 (2015) 1–4
the temporal evolution of such and such, it has been here solved from any compatible OPC client, such as for example MatlabTM.
by the method of moments. The kinetics of crystallization process The full documentation includes the OPC tags of the variables of
has been also described with the more complex relation () in interest. This standalone simulation is also offered with HMI
function as recommended in [] and []. The PBE equation interface [Alves], from which it can be manually operated, in a
considers the possible existence of secondary nucleation, which is way which is similar to the traditional operator’s practice in sugar
here described by (). industries (fig. 10).
(falta ecuaciones de balance de población y momentos, etc)
The authors consider that this increase in the realism and
complexity of the model of the crystallizers could be useful to
encourage some control approaches to go deeper and attempt a
direct involvement in the control of each individual unit
crystallzation cycle.
Steam from
boiler wvap_in
Pvap_in
Effect Effect Effect
I II III BxSP
Juice Condenser
BxC
Wjuc Figure 10: HMI interface for stand-alone simulation.
LC
Purjuc LC LC
Concetrated syrup
to crystallizers
Bxjuc
Measurements w2mc
Poor syrup min J
Sugar u
decision variables (u), in what follows further classified as u C and The optimal plantwide control of the last department of a
u D to make separate reference to the continuous and dicretes factory, as is the strictly the case that has been described here,
cases, whatever parameters might exist (p), and perhaps also consists basically in trying to keep up with the intermediate
explicitily of the time (t). product flowrate arriving from the upstream units in the most
efficient way. So, in this partial factory (PF) case, the
mathematical programming problem should not take into account
the flowrate of juice to be processed as a decision variable since it
Table 1. Summary of proposed plant-wide optimal control problems.
is considered to be fixed elsewhere and so is treated as a
Both Partial Factory (PF) and Full Factory disturbance. The optimal solution of this type of problem
Cost index (FF) cases: typically consists in minimizing the cost of production. In our
particular case, this cost is considered to be dominated by the
∫w dt
J = vap _ in consumption of heating steam by the crystallizers. A lower
∫ w j dt
average flowrate of steam from the evaporators is beneficial not
Variables Restrictions only due to the obvious fact that it directly implies a lower
Continuous FPi V: I,II,III 0.8 ≤ FPi ≤ 1.2 economic cost. A lower consumption also means that more steam
decision is available for other consumers in the factory, or that that it can
B x SP, Pvap _ in V: II,III 69 ≤ Bx SP ≤ 76
variables be obtained from a downstream, lower pressure evaporation effect
(u C ) Apcent V: III 0 ≤ Apcent ≤ 100 with the associated extra savings due to a greater reutilization of
energy in the evaporation station. There are also benefits related
Added for FF case: Added for FF case: to the quality of the resulting products: if the amount of juice to
w j V: I ,II, III 0 ≤ wj process allows it, it is always better to proceed with the
Discrete Load i V: I, II, III Loadi ∈ {0,1} crystallization at a slower pace, working at correspondingly lower
super-saturations avoiding the risk of unwanted nucleation.
Swi ∈ {0,1}
decision
Swi V: II,III Notice that the variable W vap_in stands for the mass flowrate of
variables
(u D ) steam that comes out of the very simplified model of evaporation
and into the three crystallizers for the variant I. In the case of
Equality path variants II and III we can keep same interpretation or, even better,
constraints h( x , x, z , u , p, t ) = 0 to take it to make reference to the flowrate of live stream entering
the evaporation station.
Granting that the PF problem as stated previously corresponds
Inequality Lsl _ min ≤ Lsl (t ) ≤ Lsl _ max best to the real scenario, it could be also of interest to consider the
path V:I,II,III case study as a self contained, full factory (FF) plantwide problem
constraints smin,i (t ) ≤ si (t ) ≤ smax,i (t ) to attempt also the maximization of the juice processing capacity
while minimizing the costs incurred. The stardard approach for
Lmx _ min ≤ Lmx (t ) ≤ Lmx _ max V: III this type of multi-objective problems is to assign market prices to
products and to utility resources such as steam. In practice, the
economic rationality of the factory, expressed in the typical
Major Disturbances Range of variation relationships between the prices of products and the costs of the
disturbances utilities, results in an optimal operation that consists in increasing
w j V:I, II, III (PF case) 1 ≤ wsyp ≤ 10 the production until the restrictions imposed by those processing
Pur j 69 ≤ w juc ≤ 76 units that constitutes the bottleneck of the plant put a limit to the
V: I, II, III overall production. Here however, we have used the same
Bx j 91 ≤ Pur j ≤ 95 expression for the cost function of both problems as is summed
69 ≤ Bxsyp ≤ 76 up in table []. That common expression for the cost can be
w2mc interpreted as the specific consumption of steam (w steam_in ) needed
V: III 0.01 ≤ w2mc ≤ 0.02 per unit of the incoming juice to process (w j ). The difference lies
wsug 0 ≤ wsug ≤ 10 in the fact that in the PF case the optimizing plantwide control can
only modify the consumption of steam by manipulating the
Measured Pvap , wvap V: I continuous decision variables P Fi for each crystallizer, seeking the
variables minimum value, corresponding to the maximum cycle lenght that
w j , Pur j , Bx j , is still compatible with the assimilation of all the incoming juice,
provided that an appropiate short-time scheduling of the semi-
Lls , Lcris _ i , Bxmc _ i , Rdy _ i , V: I, II, III
batch units given by the Load i discrete commands is achived. On
Pvap _ i ; Pcal _ i , Tmc _ i , si the other hand for the FF case to the already mentioned decision
variables, the flowrate w j itself is added, so the same cost function
can now be minimized as before by decreasing the consumption
Pvap _ in , PvapI , PvapII , wvapI , wvapII V: II, III or by increasing the juice processed as it appears in the
denominator. Notice that in table [] w j stands for the w syp the
Lmx , wmc , wrs , Bxsl , Pursl V:III flowrate of thick syrup entering the buffer feed liquour tank for
variant I of the case study study, while for variants II and III the
flowrate of the more diluted thin juice (w juc ) entering the
evaporation station is considered instead. In all cases the subindex
Mazaeda et al. / Revista Iberoamericana de Automática e Informática industrial 00 (2015) 1–4
i make reference to each of the three individual crytallizers. The things more than what is strictily necessary. In any case,
term h() in eq.() and in table [], which acts as an equality path depending on the commitment of the user, the above sugestions
constraint for the mathematical problem, represents the dynamic can be altered both in the sense of having more realism, by taking
model of the whole plant, that is the set of DAEs describing each into account the intermittent and delayed availability of purity or
specific variant. On the other hand, for the rest of elements such the fact that supersaturation has to be estimated; or, on the
as the decision variables, the disturbances acting on the plant and contrary, to consider as directly measured any other variable in
the measured variables, the changes from a one variant to next are the model, which is not contemplated in table [] but that,
introduced incrementally. So, for example, in variants II, III the depending on the application, would allow the practitioner to
continuous decision variables related to the evaporation station as focus on the control solution he/she tries to test, avoiding wasting
P vap_in , the live steam coming to the first stage and the setpoint of time and efforts in some aspect, that could be relevant in a real
the Brix controller BxSp are added to the already exisiting for implementation but that is considered of secondary importance if
variant I. On its turn variant III adds to the previous ones, the the intention is to highlight the virtues of some particular
continuous decision Ap cent that controls the massecuite flowrate to approach to solve the plantwide problem. The measured variables
the centrifuges thus acting as a knob for keeping the level of the considered by default are the pressure (P vap ) and mass flowrate
strike receiver (L mx ) between limits. Same thing happens with (W vap ) of steam that only have sense in variant I and, for all the
path inequality constraints. There are those valid for the three three case studies: the mass flowrate (w j ), the purity (Pur j ) and
variants: the one corresponding to keeping the level of the Brix (Bx j ) of thin juice or thick syrup depending on the variant
standard liquour buffer tank at permissible values and those to and the level of feed liquour (L ls ), and for each individual
make sure that an undue acceleration of the crystallization cycle crystallizer: the level (L cris_i ), the dry content of the massecuite
does not make the super-saturation to go beyond the safe meta- (Bx_i), the pressure in the calandria (P cal_i ), the pressure in the
stable zone. For the case of variant III, the already mentioned vapor space on top of the massecuite (P vap_i ) and the average
inequality, concerning the operation of strike receiver tank, is temperature of the massecuite (T mc_i ), the super-sturation (s i ) and
added. a discrete signal indicating if the particulat crystallizer is ready to
The acting disturbance and their expected range of variation initiate the processing of the following batch (R dy_i ). In variant II
are also shown in table []. Here for the PF case, the flow rate of and III other measurements must be added, such as: the pressures
juice or syrup entering the plant must be taken into consideration and mass flowrates of the I and II evaporation (P vapI , P vapII , w vapI ,
as previously discussed. In every case the purity and Brix of the w vapII ) effects along with the live pressure at the input of the
entering stream must always be considered. For the variant III evaporation station (P vap_in ). For the III variant the level of the
there is an important disturbance to take into account which is strike receiver (L mx ), the flow rate of the reclycled rich syrup
expressed by the ratio (w2mc) of water to massecuite entering the from the centrifuges (w rs ), and the dry content (Bx sl ) and purity
centrifuges. Additionally, and to describe the problem as (Pur sl ) of the feed liquour tank are also provided.
faithfully as possible to the real plant, the possibility of adding to
the feeding liquour tank the stream of sugar (w sug ) recovered by 5. A brief account of possible plantwide control strategies
the second and third stages of sugar house department has been
contemplated as additional disturbance, taking into account the The implementation of the optimal problem stated in table []
fact that those exhausting stages had not been modelled. can be attempted from different angles. The standard way of
The range stipulated for all the disturbances are compatible tackling this kind of problems, is perhaps the hierarchical, two
with the ones typically found in a real factory. Obviously the steps approach known as real time optimization (RTO). This type
absolute values of the flowrates involved, have been of strategy makes use of two models for the same plant: one of a
proportionally reduced to match the scale down version of the static nature, typically non-linear, representing an economic-wise,
factory that the model represent. On the other hand the Brix of the accurate description of the plant and it is used in the upper layer
thin juice entering the evaporation in variants II and III, is higher for calculating the optimal values of the available degrees of
in the model than in the real case, since we are omitting the first freedom. This optimal references are then presented to a the lower
two evaporation effects of the typical evaporation department. layer equipped conventional model predictive control (MPC)
Notice in particular that the range of purity and input and output using a second, simpler, maybe linear dynamic model of the same
of evaporation station are the same since this department as not process, which is used at a much shorter timescale, with the help
influence over this variable. of which it seeks to follows the handed down references,
The optimal problem as stated in table [] should be considered minimizing both the deviation of the output from the calculated
as a reasonable starting point for approaching the plantwide optimal trajectory and the control effort, complying, at the same
problems. However, the user could decide to modify it as time, with the required restrictions, in the receding horizon
considered appropriate. For example, it could make perfect sense manner which is characterisitics of the MPC algotithm. The RTO
not to make full use of all the degrees of freedom available for a strategy is sound but presents also important challenges for a
particular problem e.g. to solve variant II, III of the benchmark susccesful implementation. In the first place, the previous cursory
without changing the source of heating by not considering the description omits the need of keeping the upper layer model up to
SW i , discrete commands. On the other hand, the measured date, by a process that requires the gathering of the necessary
variables reported in table [] are the ones that are typically measurements from the processs once an approximate steady state
available in the real factory except for the fact that purity data are situation has been reached, and the subsequent carrying out of a
not available on-line but, for example, every eight hours shift data reconciliation and parameter estimation steps to guarantee
from the factory labs. Notice also that the super-saturation the current validity of the model. Of course, from this procedure,
variable, which in the real case is not measured and should be it is clear that for truly optimal operation of this scheme the upper
inferred from variables such as purity and temperature, among model has to be structurally correct. There is also the risk of
others, is here considered as directly available to not complicate having to handle in the lower layer unfeasible references
Mazaeda et al. / Revista Iberoamericana de Automática e Informática industrial 00 (2015) 1–4
provoked by the existing of a possible mismatch between the two attempted. There is always the possibility of implementing and
models of the same process. plantwide RTO or EMPC scheme, emcompassing the whole
The existence of structural model-plant mistmatch is an factory and embedding into the problem all the economic
important issue since the RTO type of solution cannot assure in objetives, quality and environmental related requirements, to then
this case the convergence to the true plant optimum. Recently, let the mathematical programming tools give us the, perhaps only
there are been some very promising results in applying an locally valid, optimal values of the plant’s degrees of freedom.
modifier adaptation (MA) strategy [] in which, instead of This approach is of course plausible but is plagued with important
changing parameters with data gathered from the process as in the pragmatic challenges going from the creation and maintanace of
traditional approach, it is the optimization problem the one that precise enough models to the timely and accurate solution of non-
gets iteratively updated with information related to the gradients linear non-convex optimization problems.
of the function cost and of the restrictions obtained from the plant In this context, the traditional plantwide control approach,
and from a possibly structurally erroneous model. It is shown developed and enriched over the years with the combined efforts
under some relatively mild conditions that the MA strategy leads of the industry and academia is far for being superseded. In [] an
to the true plant extremum. interesting account of the subject is given and some useful
Other shortcoming of the original RTO scheme is related to guidelines which recommend an approach that combines top-
the use of static plant model for the economic decisions. This fact down with a bottom-up methodology is discussed. The top-down
rules out its inmediate applicability to intrinsically non-static analysis would include the recognition of the plantwide economic
plants such as batch units, but even in those plants which are objetives, of the available degrees of freedom and, among them,
designed and conducted in a continuous way, its usefulness is the identification of the manipulated variables that determines the
limited to only those occasions in which it can be considered as plant’s throughput. The bottom up part, on its part, would start at
being in a reasonably quiescent state, disregarding the opportunity the level of the individual process units and is concerned with the
of economically optimizing the relatively frequent changes from dynamic aspects including things such as the control on
one steady state to the next. Improvement on this limitation has inventories, pairing of controlled with manipulated variables, and
come in the form of dynamic-RTO (D-RTO) where the upper stabilization, among others. This mixed approach doesn’t rule out
layer uses a dynamic model. But the most radical althought the application of more advanced techniques such as RTO to
somewhat natural step forward in this regard has been to orchestrate the overall economic management combined with the
consolidate both layers of the RTO scheme into a single one. The appeal to techniques such as MPC or SOC to offer at the level of
resulting controller uses the receding horizon scheme of MPC, the individual units, and thus at a faster timescale, a response that
uses a full fledged dynamical non-linear model of the plant, is also sound in an economic sense.
equipped with all the required terminal and path equality and It is worthwhile to include, in the design of the plantwide
inequality constraints, and a free form stage cost to optimize that control, a bottom up perspective similar to the above described,
is directly related to the economic performance of the plant. But centered on the efficient management of the individual process
the conceptual simplification that this so called economic non- units, as an important step in the building of the overall solution.
linear MPC (E-NMPC) brings about is not free of some new This type of strategy has been the traditional way of designing
hurdles of its own. Now, for example, the optimization problem and operating the plant with a decentralized approach. However,
to solve at each sample time is much harder: being typically non- the level of energy integration of the contemporary factory,
linear and non-convex. Additionally, the mature stability results determined by the need of maintaining a competitive edge in
existing for the conventional, reference-following MPC are not global markets with very narrow benefits margins, makes the
longer applicable. Recently, however, important breakthroughs in completely decentralized management impossible or unattractive.
of E-NMPC theory, concerning stability and the economic But other alternatives, such as distributed or cooperative
optimality of the results for the finite horizon case have started to plantwide management strategies, offer the convenience of
appear [][]. approximating the optimal control of the overall plant with the
There are other interesting approaches in this optimal conceptual simplicity, the factory operator friendliness and
operation setup. It is particularly interesting the idea of converting computational scalability of the more traditional process unit
the optimization problem into a regulation one. This strategy centered approach. The ideas of distributed control are not new [],
requires the off-line identification of the necessary conditions of but have been the objective of recently renewed interest []. In a []
optimality (NCO) for the problem at hand, via theoretical analysis a possible distributed architecuture for the sugar case study has
or intensive model experimentation. Then, and provided that the been discussed.
structure of the solution remains unchanged, at least a reasonably Another level of complexity that needs to be dealt with is the
good sub-optimal operation could be achieved by enforcing the need of performing the scheduling of batch or semi-batch units.
identified NCOs using conventional PID controllers. A similar This type of units is particularly common in food processing
idea is the one proposed as self-optimizing control (SOC), which because by resorting to the execution of a fixed or cycle to cycle
basically identifies the best variables or combination of variables improved recipe, they increase the opportunity of obtaining the
to regulate to fixed setpoints, is such a way as to keep the unit much needed quality properties of products in a situation of
close to optimal operation in a way which is relatively insensitive poorly known models and the lack of quality related on-line
to the expected disturbances. measurements. In a general context, and taking as a guide the
The previous compilation of tools, necessarily incomplete, ISA-95 proposed hierarchy, the market driven long to medium
should not give the impression that the plantwide control of term programming of the production performed at the ERP level
process industries is a sorted out problem. Quite the contrary, the must be translated to a short term scheduling of the individual
effective control of a single moderately complex process unit is a units, a task that should be performed at the MES layer taking
task that demands a lot of planning and attention to details. This into account the production throughput requirement handed down
complexity is of course compounded if the whole plant control is from ERP but also the information pertaining to the availability of
Mazaeda et al. / Revista Iberoamericana de Automática e Informática industrial 00 (2015) 1–4
individual units, the cycle duration in relation with the recipe to Prediction horizon
(NP complete cycle for each crystallizer)
be applied, the holding capacity of intermediate buffer tanks and
the coordination with the continuous units. The short time Control horizon
scheduling, which had been chosen preferably in an optimal way, (Nc complete cycles
, Ej: Nc=2)
over a horizon whose duration depends on the charateristics of the
plant, is then delivered to the lower control related layers. The
typical uncertain context in which this kind of problems is carried
out, suggest that the idea of retrofitting information from the
lower layers concerning the actual performance of the planned Crys. I ...
Tload1_1 Tload1_2
scheme to the upper levels with the objective of re-launching the
scheduling optimization would be an interesting idea. In [] a very Load1_1 Load1_2
complete discussion of these issues is given, making strongly the
case for a tighter integration between the scheduling and control
layers, in the way that the here describied sugar case study
Crys. II
...
requires. Tload2_1 Tload2_2
Load2_1 Load2_2
EMPC Disturbances:
Wsyp
...
Process
state Bxsyp
Pursyp Crys. III
Objective Tload3_1 Tload3_2
Model
Function
Load3_1 Load3_2
PFi, Tload_i J
Optimal Figure 12: Receding horizon strategy for crystallizers’ scheduling
PF_i, Tload_i
Optimizer
The simulation study considers the partial factory case with
Restrictions
Plant the profiles of the disturbances for the putiry, Brix and flowrate of
(V: I)
the incoming juice shown in figure (). The variables to optimize
are the PF i multiplying factors deciding the steam consumption of
Figure 11: E-MPC control scheme. the respective crystallizer, which are all restricted to the interval
6. Optimal control implementation example [0.7,1.2], and the discrete decisions (Load i ) on when to start each
individual cycle. Due to the hybrid nature of the model, the
The objective here is to give an example so as to show the mathematical optimization to be carried out at each sample time
general zest of the optimal control problem and of the trade-offs would require the use of a time expensive mixed integer non
involved. To keep things simple, a centralized solution has been linear programming algorithm, a fact that would aggravate the
adopted as represented by fig. considering the variant I of the case scalability problems of the adopted centralized approach. There is
study. Other simplifying assumptions include the absence of a parametrization for these kind of switching models where the
mistmatch between the model and plant and the full access to the binary variables are represented instead by the continuous time
state of the latter. The centralized optimal control is made of a elapsed from the end of previous cycle to the start of the next
single layer and is of the receding horizon type with an economic (T loadi ) []. Adopting this strategy, the mathematical programming
related performance index (E-NMPC) of the type considered in problem become continuous and can be solved to local optimality
table (). by a standard SQP type algorithm. This pàrametrization, however,
requires that both the prediction and the control horizon to be
defined in terms of a fixed number of cycles as depicted in fig().
In any case, a centralized approach to the plantwide problems
implies a considerable numerical effort for the E-NMPC strategy,
so the prediction and control horizon has been chosen both to be
equal to one complete cycle of each crystallizer. The sample time
of 1800 seconds has been considered adequate.
Mazaeda et al. / Revista Iberoamericana de Automática e Informática industrial 00 (2015) 1–4
(a)
(a)
(b)
(c)
Figure 13: Disturbances on input syrup: (a) mass flowrate, (b) Brix, (c) purity.
(b)
Figure 15: Heating steam source. (a) Pressure (bar), (b) mass flowrate (kg/s).
(a)
The crystallizers share the same source of heating steam, so
the availability of this resource is a key element in the success of
the plantwide optimal control scheme. In fig.b the total flowrate
demanded by the three crystallizers is shown, while the effect that
this demand imposes on the common real steam source, as
reflected by the pressure which it is able to deliver, is shown in
fig().a. Notice that the steeply decreasing flowrate of steam along
the cycle suggests the stipulation of a crystallizers’ scheduling
(b) that seeks to avoid the overlapping of the peak demands by
providing enough separation between the load commands of the
different units. This requirement can be enforced in different
ways by an appropriate adjustment of the baseline mathematical
problem outlined in table (). The solution here adopted consists in
introducing an additional restriction demanding for the common
heating pressure (P vap ) to be above some specified value.
(c)
(a)
factors at the end of the exercise can be partially explained with Rouillard, E., E., A., 1985. A study of boiling parameters under conditions of
the above considerations. laminar non-Newtonian flow with particular reference to massecuite
boiling. PhD Thesis. University of Natal. South Africa.
Sager, S., 2009. Reformulation and Algorithms for the Optimization of
7. Conclusions Switching Decisions in Nonlinear Optimal Control. Journal of Process
Control vol. 19, 1238-1247.
A sugar factory case study describing a genuine problem has Saguy, I.S. et al., 2013. Challenges facing food engineering. Journal of Food
Engineering, 119(2), pp.332–342. Available at:
been described. The simulation has been offered by means of
http://dx.doi.org/10.1016/j.jfoodeng.2013.05.031.
three variants of different degrees of complexity and realism with Sam Saguy, I., 2015. Challenges and Opportunities in Food Engineering:
the purpose of reaching as wide an audience as possible. A Modelling, Virtualization, Open Innovation and Social Responsibility.
baseline plantwide optimal control has been put forward as a Journal of Food Engineering, 176, pp.2–8. Available at:
reasonable mathematical statement expressing the purpose of http://www.sciencedirect.com/science/article/pii/S0260877415003118.
using the available degrees of freedom for coordinating the Urbaniec, K., 2004. The evolution of evaporator stations in the beet-sugar
industry. Journal of Food Engineering, 61(4), pp.505–508.
continuous with the batch type of units, to process the incoming
product stream with the minimum consumption of the common Storn, R., Price, K., 1997. Differential Evolution- A simple and efficient
and limited resources. A relatively simple centralized strategy has Heuristics for Global Optimization over Continuous Spaces. Journal of
been implemented mainly to show in action the main trade-offs Global Optimization, vol. 11, 341-359.
involved and the plausibility of the problem, but with the
aspiration of stimulating in the audience the interest of using more
promising plantwide strategies.
Acknowledgements
References