Anda di halaman 1dari 10

prevention: a prospective, randomized study.

Allergy and Immunology 11: 149161.
Huang YS and Sinclair AJ (1998) Lipids in Infant
Nutrition. Champaign: AOACS Press.
Institute of Medicine (1998) Dietary Reference Intakes for
Thiamin, Riboavin, Niacin, Vitamin B
, Folate,
Vitamin B
, Pantothenic Acid, Biotin and Choline.
Washington, DC: National Academy Press.
Jakobsson I, Lothe L, Ley D and Borschel MW (2000)
Effectiveness of casein hydrolysate feeding in infants
with colic. Acta Pediatrica 89: 1821.
Kawasaki Y, Tazume S, Shimizu K et al. (2000) Inhibitory
effects of bovine lactoferrin on adherence of entero-
toxiengic E. coli to host cells. Bioscience, Biotechnology
and Biochemistry 64: 348354.
Renner E (1989) Micronutrients in Milk and Milk-Based
Products. London: Elsevier Applied Science.
Salle BL and Sawyer PR (1993) Nutrition of the Low Birth
Weight Infant. New York: Raven Press.
SanGiovanni JP, Berkey CS, Dwyer JT and Colditz GA
(2000) Dietary essential fatty acids, long-chain poly-
unsaturated fatty acids, and visual resolution acuity in
healthy fullterm infants: a systematic review. Early
Human Development 57: 165188.
Toba Y, Takada Y, Yamamura J et al. (2000) Milk basic
protein: a novel protective function of milk against
osteoporosis. Bone 27: 403408.
Tsang RC and Lucas A (1987) ESPGHAN Committee on
Nutrition of the Preterm Infant: nutrition and feeding of
preterm infants. Acta Pediatrica Scandinavica 336
(supplement): 314.
Tsang RC, Lucas A, Uauy R and Zlotkin S (eds.)
Nutritional Needs of the Preterm Infant. Baltimore:
Williams & Wilkins.
R Oliveira, New University of Lisbon, Portugal
P Georgieva, Bulgarian Academy of Sciences, Soa,
S Feyo de Azevedo, University of Porto, Portugal
Copyright 2002, Elsevier Science Ltd. All Rights Reserved
Industrial evolution in the second half of the twenti-
eth century was inuenced mostly by four types of
interrelated factors: progress in digital technology,
advances in science, evolution of societal require-
ments and demands and, particularly over the last 30
years, evolution of business concepts.
Developments in digital technology and in systems
theory led to major progress in sensor and informa-
tion technology and a revolution in the availability
of distributed control systems and open software
applications. New concepts, particularly knowledge-
based measurement and advanced control meth-
odologies, are slowly but steadily being brought into
the practice of process operation, performing online
and in real time.
Infectious Diseases of Dairy Animals see Diseases of Dairy Animals, Infectious: Brucellosis;
Leptospirosis; Listeriosis; Tuberculosis; Johne's Disease; Bluetongue; Foot-and-Mouth Disease; Hairy Heel
Warts; Salmonellosis.
Process Control
Societal and economic factors have driven evol-
ution in the same direction. The increasing concern
for health, safety and sustainability issues, market
quality requirements, economic pressure and the
evolution of company strategy from local to global
business concepts the so-called knowledge economy
concepts all together reected on plant investment
decisions, favouring process automation for cleaner
and safer operation, higher product quality and im-
proved process efciency and productivity.
Discussing plant automation froma technical point
of view means a discussion on instrumentation, con-
trol system congurations, data communications and
theoretical control structures.
This article deals with instrumentation, addressing
in particular both basic and advanced concepts con-
cerning sensors and the issue of how to integrate local
hardware for automatic control, usually dispersed
throughout the plant.
Basics of Plant Automation
There are well-established methodological steps
for the design and implementation of a control
structure: (1) designing and implementing an
appropriate, exible, control conguration; (2) per-
forming rst-level data acquisition and process
monitoring, including record keeping and rst-level
alarms; (3) performing data interpretation (im-
plementing second-level process monitoring); and
(4) designing and implementing optimization and
control solutions.
In plant-wide distributed control congurations, as
illustrated in Figure 1, the backbone of transmission
is all digital. Communication protocols (some open,
some proprietary) ensure data transmission for cen-
tralized data interpretation, for monitoring and for
plant scale optimization. Control at sector and unit
level is usually performed locally. Local-level signal
transmission has for many years been analog only;
initially, pneumatic signals and later electric (current
or voltage) signals. However, more and more, infor-
mation also ows digitally between (smart) sensors
and controllers.
At a local level, process control system instru-
mentation includes: (1) sensors for measurement
of process variables; (2) controllers, for imple-
menting a proper (digital) control structure

I f x>34
Y=a*b/( g+1000)-d/f
Elseif x<100
End i f
Do I =1 to N
G( i) =h( I,8)
End do

I f x>34
Y=a*b/( g+1000)- d/f
E lseif x<100
E nd if
Do I=1 to N
G( i) =h(I ,8)
E nd do

I f x>34
Y=a*b/( g+1000)-d/f
Elseif x<100
End if
Do I =1 to N
G( i) =h( I,8)
End do

I f x>34
Y=a*b/( g+1000)- d/f
E lseif x<100
E nd if
Do I=1 to N
G( i) =h(I ,8)
E nd do

If x>34 Y=a*b/ (g+1000)- d/f Elseif x<100 Y=0 Ret ur n End if

I f x>34
Y=a*b/( g+1000)- d/ f
E lseif x<100
E nd if
Do I=1 to N
G( i) =h(I ,8)
E nd do

If x>34 Y=a*b/(g+1000)- d/f Elseif x<100 Y=0 Ret ur n End if
I f x>34
Y=a*b/( g+1000)- d/f
E lseif x<100
E nd if
Do I=1 to N
G( i) =h(I ,8)
E nd do

I f x>34
Y=a*b/( g+1000)-d/f
Elseif x<100
End if
Do I =1 to N
G( i) =h( I,8)
End do
Local unit 1 Local unit N

Figure 1 State-of-the-art instrumentation for process control: devices, signal transmission and information ow. Continuous line,
network line (e.g. Fieldbus protocol/devices); dashed line, analog line (010V, 420mA) or digital serial line (RS232, RS485); PID,
proportional-integral derivative; LP, local process; FP, FieldPoints.
(e.g. proportional-integral derivative (PID), pre-
dictive or adaptive controller, etc.); (3) nal
control elements for manipulation of process in-
puts; and (4) support devices such as general signal
conditioners of both input and output signals,
electric (V/I and I/V) transducers, electric-to-pneu-
matic transducers and hold elements.
The range of instrumentation and control systems
is very large today. Nowadays industrial users are
available through the internet information sites of
instrumentation manufacturers with very detailed
information on all types of instrumentation and
control systems congurations, including data com-
munications, related to plant automation.
A relevant management decision in the automation
of older plants and most fall into this category is
how to step from existing local control solutions
to integrated distributed control. Investment in
a complete new solution is high and the tendency is to
try to adopt a solution that makes use of available
equipment. This is often hard to achieve and leads to
a nal conguration that mixes too many different
control equipment suppliers, with the related costs of
operation and maintenance.
Measurement Instrumentation
A key limitation to the application of process control
is the lack of appropriate sensors for many process
variables. This is so in dairy processes and therefore
measuring techniques for online use in the dairy
industry are subject to considerable research and
manufacturing interest. Basic concepts and the main
measurement techniques are reviewed in this section.
Basic Characteristics of Sensors
A sensor is composed of a sensing element and
a transmitter, as indicated schematically in Figure 2.
The sensor output, whether analog or digital, should
be a standard signal suitable to be supplied to
a controller.
Most transmitters respond rapidly. When the sen-
sor element response is also fast, then measurement
dynamics can be neglected in view of modelling
process dynamics. Such a case is indicated in Figure 3,
which shows the time response of a pH electrode to
a step change in pH from 4 to 7. The characteristic
rst-order time constant for the sensor dynamics
(the time required for the response change to reach
63%of the step change) is in the order of milliseconds
( 570 ms). There are, however, cases where the
measurement dynamics, particularly time lag, may
be signicant and ignoring it can lead to control
Sensors can exhibit linear or nonlinear behaviour.
This is related to and expressed by the relation
between the variation of the property value and that
of the transmitted signal. Sources of nonlinearity
usually lie on the sensing element. For a linear sensor,
the gain for a given calibration is constant and equal
to the ratio between the set span and the range of the
sensor output. Nowadays, with digital data acquisi-
tion, transducer nonlinearities are easily incorpor-
ated in the data interpretation software and cause no
practical difculties.
Specialized Sensors/Measurement Systems
The most important properties of dairy processes
subject to measurement that reect both process
operation and product quality are classied as ob-
jective or subjective. Examples of the former are pH,
temperature, ow rates, pressure and level. Sensors
for online measurement of such properties have
been available for a long time. Properties such as
taste, avour, colour and consistency are considered
to be subjective and difcult to measure.
Commercial sensors applied in the dairy industry
are subject to several quality constraints, such as
and ADC
phenomena (P):
pressure, flow,
density, etc.
Digital control
DAC and
Final control
Figure 2 Schematic representation of stages in a digital control system. ADC, analog-to-digital conversion; DAC, digital-to-analog
sanitary, safety and environmental requirements.
A trend in new sensor design technologies is the
increasing integration of the sensing elements into
silicon chip microcircuits. These new measurement
devices directly incorporate all circuitry needed to
self-compensate for environmental changes and yield
an output that is suitably amplied for transmission
to standard electronic controllers. These sensors
offer the advantage of small size, reduced price and
practically no mechanical parts to wear out.
Next, measurement techniques and instrumenta-
tion for the most important objective properties in
the dairy industry are considered.
pH pHmeasurements are of paramount importance
for quality control in milk fermentation and related
processes. For example, inadequate pHcan be a cause
of discoloration in cheese or fermented milks, excess
free whey and excess or inadequate tartness in fer-
mented products; pH changes are related to the
viscoelastic properties of yoghurt and they are also
correlated to the physiological state of bacteria in the
lactic acid fermentation. Also, the nal pH value is
normally a feature of the nal product: fermentation
converts lactose to lactic acid, causing a drop in pHto
a value in the range of 4.254.5; rapid cooling at the
correct level of lactic acid then stops bacterial action.
pH electrodes can be in direct contact with food,
if they meet sanitary requirements. In general terms,
pH measurements, particularly in conjunction with
electrical conductivity measurements, constitute an
important means for continuous, real-time process
Temperature In industrial applications the main
measuring devices for low temperatures (below

C) are based either on thermoelectric effects

(thermocouples) or on resistance changes (e.g.
platinum resistance temperature devices (RTD), ther-
mistors). Measurement characteristics, particularly
sensitivity and the degree of (non)linearity, favour
the use of platinum RTD; the most widely used is the
so-called PT100 device.
Thermal processing is a key stage in most dairy
process operations. As an example, during pasteur-
ization of yoghurt the temperature is raised to 85

C to destroy undesirable microorganisms and

denature the whey to improve viscosity and prevent
syneresis. Then the product is cooled to 4045

C so
that the sterile mix is ready for inoculation. Also,
freezing and refrigeration are used to prevent the
growth of unwanted microorganisms. In general,
these operations and process units require a reliable
and accurate temperature-monitoring and control
Level, density or interface level Liquid height, dens-
ity or interface level between two liquids can be
measured either by differential pressure (d/p cells)
sensors or by measurement of buoyancy force on
a displacer suspended in a liquid. An electronic
transmitter converts the output of the sensing element
to an appropriate analog or digital output signal.
Nowadays instrumentation companies offer the
dairy industry a variety of special transmitters for
level monitoring and control applications in invent-
ory tanks and clean-in-place (CIP) vessels.
Pressure In most process operations, particularly
when thermal processing is required, pressure regu-
lation is one of the essential control loops. Pressure
sensors are thus among the most commonly used
0 500
Time (ms)
=2.45 V
N=0.12 V
=1.48 V
=570 ms
1000 1500 2000 2500 3000 3500 4000
Figure 3 Time response of a pH electrode to a step change in pH (47).
online sensors. D/p cell transmitters for either gauge
or absolute pressure are the solution which is gener-
ally adopted.
Thermal conductivity Thermal conductivity ex-
presses the ability of a substance to conduct heat.
The most common thermal conductivity probes
consist of an assembly of an electric heated wire
and a temperature measurement system (based on
thermocouples or thermistors) fromwhere heat uxes
are measured and heat conduction is inferred.
Thermal conductivity measurements nd wide
application in dairy chemistry and biochemistry and
in food process engineering. The measurement of
thermal conductivity by line heat source probes may
be used for inline determination of the coagulation
time of milk for cheese curd and yoghurt production
and may be helpful for to automating these processes,
aiming at maximizing the yield of cheese and yoghurt.
Coagulation time may be detected by the sharp
increase in the temperature difference between probe
temperature and initial milk temperature.
Electrical conductivity Electrical conductivity (G)
expresses the ability of a substance or medium to
conduct electricity. It is employed for quality control
and further nds online use in identifying feature
points of fermentation states. For example, electrical
conductivity measurements allow the urease activity
and the acidication activity in lactic acid fermenta-
tions to be distinguished.
In a dairy plant where many fermentations are
performed, either simultaneously or sequentially, the
real-time prediction of fermentation completion time
is very important for scheduling raw material supply
and energy utilization. Data-driven models (e.g.
neural network-based) relating characteristic prop-
erties of the fermentation state (model outputs) to
pH and electrical conductivity (model inputs) can be
rstly identied (trained) and subsequently used to
monitor the fermentation process and to predict fer-
mentation times.
Viscosity The main items of equipment for vis-
cosity measurement are process viscometers. Shear
viscosity for Newtonian uids can be measured by
a capillary ow viscometer. Cone and plate visco-
meters are suitable to determine the shear viscosity of
time-independent non-Newtonian uids. From the
changes in viscosity, the stages of aggregation and gel
formation can be described and calculated. The vis-
cosity can nally be used as an objective measure for
the control of coagulation processes.
In industrial operations involving slurries, pulps,
grease or other similar media, consistency, rather
than viscosity, is measured by rotation and oscillation
rheometers. These devices allow, for example, the
continuous online monitoring of the transition of
milk uid into a viscoelastic gel structure.
NIR spectroscopy Infrared and near-infrared (NIR)
spectroscopy can be used to measure the levels of
water, fat and protein in liquid milk and related
products online (and ex situ). This is a commonly
used technique for quality control, but may also be
used for online closed-loop control.
Optical density Optical density (OD) is commonly
used to measure biomass concentration online and
ex situ. The measurement principle is based on the
individual or combined use of measurements of
transmission, reection or scattering of light. The
interpretation of the signal is complex, but it is nor-
mally linearly correlated to biomass concentration in
diluted solutions.
Final Control Elements
Final control elements (FCE) are devices, driven by
controller signals, used to manipulate process control
variables. In most cases control actions consist of
adjusting ow rates of process input or output
streams (solid, liquid or gas) or cooling and heating
uids. The most widely used FCE are ow regulator
Designing a valve involves taking decisions on
valve size, choice of body material, choice of type
of valve (signal-to-close or signal-to-open) and choice
of ow versus aperture characteristics (essentially
linear or equal percentage valves).
Valve sizing (computing the Cv parameter) and the
choice of valve ow characteristics require consider-
ation of hydrodynamic aspects, particularly pressure
drops along the piping. The choice of material de-
pends on the properties (corrosive, slurry, etc.) of the
process uid. For the dairy industry, stainless-steel
valves are most common. The decision on working
with normally open or normally closed valves derives
directly from answering the safety question of how
the valve should stay in an emergency due to energy
Regulator valves are typically driven by a pneu-
matic signal (range 315 psig or 0.020.1 MPa).
Signals are normally transmitted to the FCE as
analog current signals (420 mA), being converted
locally by a current-to-pneumatic transducer. Altern-
atively, a step motor, driven by a digital signal from
the controller, can actuate valves.
Onoff valves normally use electrical or pneumatic
actuators. They are used mainly for sequential control
and during start-up and shut-down procedures. Other
nal control elements, namely displacement devices
and pumps, are increasingly digitally actuated by step
Digital Control Equipment
Digital control instrumentation represents, the new
standard and indeed has introduced a major change
of mind-set with respect to control system structures
and solutions, with new procedures concerning
communications and calibration routines. Digital
control instruments offer the computing power re-
quired to implement advanced model-based mon-
itoring and control algorithms (software sensors and
predictive control) at local level and in real time. They
enable the implementation of plant-wide monitoring,
optimization and control solutions through the
available distributed control architecture.
Distributed Control Systems
A distributed control system (Figure 1) is simply an
arrangement whereby control devices and computer
processing power are distributed through a network
instead of being centralized.
The main devices in a distributed control system
are essentially microprocessor-based subsystems
such as programmable logic controllers (PLC), smart
sensors, supervisory and engineering stations and
other inputoutput (I/O) devices (e.g. FieldPoints,
device integrators, etc.). All devices in the network
must be integrated with proper hardware and soft-
ware for communications.
PLC are today's industrial standard for local digital
control. They are reliable special-purpose computers
for control in the industrial environment, consisting
of a set of I/O modules and a programmable central
processing unit. They can perform analog-to-digital
(AD) and digital-to-analog (DA) conversion and
have special-purpose digital I/O ports (PLCs were
originally designed mainly for event control). PLC
normally use proprietary programming languages.
Smart sensors are devices that through their digital
system can be connected to the network, communic-
ating bidirectionally with the other devices. In par-
ticular, they accept remote commands for remote
General-purpose computers (PC, workstations)
may be interconnected in the control system network
to carry out inferential measurement procedures,
high-level data analysis, supervisory duties or more
sophisticated dedicated control tasks.
FieldPoints are modular I/O devices that connect
a bank of analog and/or digital I/O modules to an
industrial network, being able to perform AD/DA
signal conversion.
The key feature of a distributed control system is
thus that the measurement and control tasks are dis-
tributed out into the eld. Integrated hierarchical
control congurations may be built where high-level
tasks, like process supervision and optimization, are
fully integrated with the low-level data acquisition
and local control tasks.
Communication Standards
Industrial communications refer to the networking
hardware and software, together with the respective
communications protocol.
A number of industrial network standards, de-
signed to meet different application requirements, are
available today: Ethernet, DeviceNet, Foundation
Fieldbus, PROFIBUS and controller area network
(CAN). Details of these industrial network standards
can be found in Table 1.
Some specify low-level sensorcontrolleractuator
communication protocols (like CAN and DeviceNet),
whereas others are specially oriented for distributed
control systems in the process industries (Fieldbus
and PROFIBUS). They differ on communications bus
specications, on velocity of data transfer, on com-
munications protocol used and on the communica-
tions model. For distributed control systems in large
factories, Fieldbus and PROFIBUS are the two most
important standards.
Ethernet is the most widely used local area net-
work (LAN) technology. An Ethernet LAN may use
coaxial cable, special grades of twisted pair wiring,
or bre optic cable. `Bus' and `star' wiring congu-
rations are supported. Ethernet devices compete
for access to the network using a protocol called
carrier sense multiple access with collision detection
(CSMA/CD). Ethernet conforms to the IEEE 802.3
specications and runs commonly under the high-
level transfer communication protocolinternet
protocol (TCPIP) (although many others are pos-
sible). The Fieldbus standard has adopted a second
alternative for the physical layer that is based on
Ethernet, thus providing a solution for factory-to-
factory communication.
Basics of Analog-to-Digital and
Digital-to-Analog Signal Conversion
All digital control systems contain a data acquisition
interface that performs AD/DA conversion. Such
tasks are commonly performed by standard compu-
ters with AD and DA cards, by FieldPoint modules
or by PLC.
It is worth examining basic aspects of the data
acquisition and control problem, namely the ow of
information fromthe `process property' to the `binary
word in the computer' and the feedback from the
`binary decision variable' to nal control element.
Data acquisition The data acquisition chain is rep-
resented schematically in Figure 2. The design stage
of the data acquisition system should start with
a qualitative analysis, addressing the following main
v The sensor should be chosen with the objective of
maximizing sensitivity for the desired measure-
ment interval. This means that calibration should
be such that the sensor measurement span should
match the measurement interval and be mapped
into the full range of the output signal.
v Industrial analog-to-digital conversions (ADCs)
are nowadays standard. A 12-bit AD converter is
generally sufcient. The input range is not a prob-
lem, assuming that the required signal condi-
tioners are available.
v Signal conditioning should be such that the sensor
output signal V
is transduced into a signal V
exhibits the same range as that of the ADC. This
situation maximizes the overall resolution of the
acquisition chain.
Still at the design stage, the quantitative analysis
must be performed in steps from the process to the
v The sensor normally delivers an analog electrical
signal (V
), which should be a known function of
the process property, P. Assuming, for simplicity,
a linear relationship, eqn [1] holds:
= k
P z
v This signal will generally undergo some form of
conditioning (transduction, amplication, attenu-
ation, etc.), after the sensor and before the ADC,
depending mainly on aspects related to compat-
ibility and range of transmission signals (V
). These types of transformations can usually
be adequately expressed by linear relationships of
the form:
= k
= k
Table 1 Industrial communications standards
Standard Description General features Application areas
DeviceNet Low-level network designed
to connect industrial devices
(sensors, actuators) to higher-
level devices (controllers)
Powered bus consisting of two
separate twisted-pair cables
Built on CAN protocol
Producerconsumer model for
data transfer
Mainly manufacturing
Digital network standard designed
specially for distributed process
control; expected to substitute for
the 420mA analog standard
H1-powered 31.25kbs
(standards ISA S50.02-1992;
IEC 61158-2) or high-speed
10/100Mb s
Ethernet (HSE)
Mainly process industries
Fieldbus (communication)
protocol (IEC 1158-2)
Family of communication standards.
Leading open Fieldbus system in
Europe; PA is mainly used in the
process industries
DP and FMS: RS485 serial line
with baud rates up to 12Mb s
PA: Fieldbus standard (IEC 1158-2)
Manufacturing and process
Ethernet EtherNet is an industrial standard that
denes only the physical layer. Some
industrial standards are built on top
of Ethernet
v Coaxial cable with BNC connectors
or telephone wiring with RJ45
connectors or bre-optic cable
(101000Mb s
(standard IEEE 802.3)
Mainly in LANs for PC-to-PC
v Ethernet does not dene itself as the
communication protocol. It runs
commonly under the TCPIP
CAN Designed originally for in-vehicle
automotive communications
CANbus (serial bus) with CANbus
communication protocol
Also for process industries
CAN, controller area network; LAN, local area network; PA, process automation; DP, decentralized periphery; FMS, Fieldbus message specications;
BNC, bayonet nut connector; TCPIP, transfer communication protocolinternet protocol.
v Considering an n-bit ADC, with an input range
, V
], the digital word D, corresponding to
, is given by:

The corresponding discretization error is given by
. For a 12-bit converter this
error is well inside industrial requirements, gen-
erally lower than all other errors in the chain.
v For the choice of the sampling time (t) in the
implementation, a simple rule of thumb is often
used in industrial applications, adopting a sam-
pling time value of about one-tenth of the process
characteristic time constant.
The implementation stage of local signal acquisition
corresponds to effectively programming the multi-
plexing, ADC, data reading (binary value D) and data
decoding (getting the property P from the binary
value D).
v For each scanning (multiplexing and data reading)
of input channels, performed at every time interval
t, programming of data decoding is performed by
successively computing the values from the binary
word D to the property P, by solving eqns [4], [3],
[2] and [1] with respect to V
, V
, V
and P,
= V
(D 0:5)
= V
( )=k
= V
( )=k
P = V
( )=k
v At this stage, process values are available in the
data acquisition application for all types of desired
actions, namely data interpretation, data plotting,
bookkeeping, analysis of alarms and related
actions, computing of control actions and output
of control commands.
Control action It is outside the scope of this article
to analyse control algorithms. Assuming that a
control decision has been taken, given by a binary
word C, such a command is transmitted from the
control system to the nal control element through
an elementary chain including the digital-to-analog
conversion (DAC) (with a hold element), and, in
the more general case where the nal control element
is a valve, a signal conditioner, a power buffer
amplier and a current-to-pneumatic signal trans-
ducer (Figure 2). For design purposes, assuming that
the DAC is set for a voltage output, eqn [9] represents
the DA conversion, where n is the number of con-
version bits of the DAC , V
is the reference voltage
(corresponding to an output interval of 0V
with V
= V
+ C=2
The industrial standard for DAconverters is normally
of 12 bit. Most common ranges of output signals are
010V or 420mA.
Finally, referring back to signal conditioning, care
must be taken to ensure that wiring is correct, that
instruments and source grounds are of good equi-
valent level (a comment that also applies to wiring in
AD lines) and that appropriate buffer ampliers are
used to protect control devices from high currents.
Advanced Topics
Two topics closely related to the state of the art in
both information technology and systems theory are
now discussed: how to make measurable what is not
so; and how to integrate information and manage
large-scale systems.
Software Sensors
In many cases, key process variables and character-
istic parameters, namely kinetic and transport para-
meters, are not available directly online and in real
time, either because they are really not measurable
or simply because measurements may be expensive
and/or unreliable.
Software sensors are algorithms for the online
computation of those state variables and parameters
that are not measurable in real time, from more easily
accessible (accurate and inexpensive) related meas-
urements. The concept is closely related to those of
inferential measuring and of state observers, widely
discussed in the specialized systems engineering lit-
erature. The design and implementation of software
sensors provide in some cases a suitable answer to
cope with the lack of instrumental sensors. It may
require the computational power of a dedicated
computer in the distributed control network,
a requirement that nowadays represents no technical
or economical problem.
A software sensor relies on a process (subsystem)
model that establishes the relationship between
measured and estimated properties. Hence, the key
for success (or failure) of a software sensor is the
availability of knowledge/information about the
relationship between measured and unmeasured
properties, i.e. the accuracy and robustness of the
underlying model. With respect to and in the context
of dairy industries, the kinetics in fermentation
tanks are the most difcult part of the process to
model. Traditionally, designing adaptive observation/
estimation algorithms, of which the most frequently
reported technique is the extended Kalman lter,
circumvents the problem. This type of method re-
quires a number of simplifying assumptions that are
not always acceptable. More recently, knowledge-
based software sensors have been reported. They rely
on articial intelligence (AI) models like articial
neural networks (ANN) and fuzzy or hybrid neuro-
fuzzy models, combined with mechanistic expres-
sions of process behaviour. Though models reported
in the literature are excellent in their ability to predict
based on minimum information, the requirements
of both know-how and computational power for
implementation are still substantial. This hinders
its industrial application on a wide scale, in the
short term.
Measurement of Subjective Properties
Properties such as taste, avour, colour and consist-
ency are subjective properties that are extremely
important for process operation in food industries. At
present operators use their human senses, i.e. smell/
aroma, feel, taste, as a gauge for the acceptability of
products. Measurement techniques for subjective
properties are reported to be under intensive invest-
igations. The electronic nose concept is maybe the
best-known example. The electronic nose mimics the
human nose to detect specic aromas and smell. This
technology is commercially available today. Image
analysis is now a well-developed technique. Proper-
ties such as material visual aspect can be inferred
fromthese techniques, providing newinformation for
further and new automatic processing.
Factory-to-Factory Automation
As a nal point, with the internet, it has become clear
that the information ow scheme of Figure 1 in an
automated factory may move down the hierarchy. As
this technology becomes reliable and secure, we may
speak of a factory-to-factory (worldwide) automated
information ow and distributed processing. Whilst
this does not represent a state-of-routine automation
solution, it represents a real management tool in the
context of prevailing business concepts.
Basic measurement concepts and control instru-
mentation applied in the dairy industry have been
reviewed in this article. The intention was to provide
an insight to the main principles and characteristics
of classical and more advanced sensor devices and
control equipment.
The measurement mechanisms considered range
from the commercially available sensors for import-
ant objective properties of dairy processes to more
sophisticated model-based software sensors and new
technological solutions for monitoring subjective
properties in a variety of food industries.
The discussion of digital control system instru-
mentation is also focused on core items such as nal
control elements, distributed control systems with
respective communications standards and signal
conversion required for data acquisition.
Though the topics considered are far from being
a complete overview of all research and industrial
developments in this area, the article provides struc-
tured information on the main instrumentation as-
pects routinely implemented or potentially applicable
in the dairy industry.
Dr Petia Georgieva is on leave from the Institute of
Control and Systems Research, Bulgarian Academy
of Sciences, Soa, Bulgaria, supported by EU
Research Project HPRN-CT-2000-00039. This work
was further nanced by the Portuguese Foundation
for Science and Technology within the activity of the
Research Unit Institute for Systems and Robotics
See also: Flow Equipment: Valves. Hazard Analysis
and Critical Control Points: Processing Plants.
Instrumentation and Process Control: Process
Control. Process and Plant Design.
Further Reading
Bentley JP (1995) Principles of Measurement Systems, 3rd
edn. New York: Longman House.
Campbell J (1984) The RS-232 Solution. Alameda: Sybex.
Carr-Brion K (1986) Moisture Sensors in Process Control.
New York: Elsevier Applied Science.
Cascetta F and Vigo P (1988) Flowmeters: A Compre-
hensive Survey and Guide to Selection. North Carolina:
Instrument Society of America.
Cubberly WH (1988) Comprehensive Dictionary of
Instrumentation and Control Reference Guides for
Instrumentation and Control. North Carolina: Instru-
ment Society of America.
Doz Y, Santos J and Williamson P (2001) From Global to
Metanational: How Companies Win in the Knowledge
Economy. Boston: Harvard Business School Press.
Feyo de Azevedo S, Chora o J, Goncalves MJ and Bento L
(1993) On-line monitoring of white sugar crystallization
through software sensors. International Sugar Journal
95: 483488.
Feyo de Azevedo S, Chora o J, Goncalves MJ and Bento L
(1994) On-line monitoring of white sugar crystallization
through software sensors. 2. International Sugar Journal
96: 1826.
Feyo de Azevedo S, Oliveira R and Sonnleitner B (2001)
New methodologies for multiphase bioreactors. 3. Data
acquisition, modelling and control. In: Cabral JM, Mota
M and Tramper J (eds.) Multiphase Bioreactor Design,
pp. 5383. New York: Taylor & Francis.
Fraser RE (2001) Process Measurement and Control:
Introduction to Sensors, Communication, Adjustment
and Control. Englewood Cliffs: Prentice-Hall.
Johnson CD (1997) Process Control Instrumentation
Technology. London: Prentice-Hall.
Krohn DA (1988) Fiber Optic Sensors: Fundamentals and
Applications. North Carolina: Instrument Society of
Liptak BG and Venczel K (eds.) (1982) Instrument
Engineers, Handbook: Process Measurement. Randor:
Chilton Book Co.
Morris AS (1996) The Essence of Measurement. London:
Oliveira R, Ferreira EC and Feyo de Azevedo S (2002)
Stability, dynamics of convergence and tuning of
observer-based kinetics estimators. Journal of Process
Control 12: 311323.
Twork J and Yacynych AM (eds.) (1990) Sensors in
Bioprocess Control. New York: Marcel Dekker.
Process Control
P Georgieva, Bulgarian Academy of Sciences, Soa,
R Oliveira, New University of Lisbon, Portugal
S Feyo de Azevedo, University of Porto, Portugal
Copyright 2002, Elsevier Science Ltd. All Rights Reserved
Over the past 20 years signicant changes have
occurred in the international food market which
have led to a signicant demand for competitive pro-
duction. The opening of frontiers in Europe, the
global market, wider availability and awareness of
information and increased purchasing power, together
with a rapidly changing lifestyle, have made customers
more sophisticated and demanding of product quality.
Also, the increasing concern and awareness about
implementing sustainability directives led to the
introduction of requirements and constraints for
operation, substantiated in new quality management
systems like ISO 9000 and other check systems based
on EU and US directives. Last but not least, advances
in digital technology have paved the way for bringing
into industrial practice new(computer-aided) methods
for competitive process operations.
The food industry had to adapt its whole chain of
production processes to these market and techno-
logical changes, having taken a large step towards
better-quality products and new, more efcient,
process operation routines in recent years. Selection
of process operation policies for increased pro-
ductivity, safer production and greater efciency with
fewer losses is a very important topic, which should
be periodically revised within this evolutionary mind-
set and practical reality.
This article gives an overview of process control
paradigms employed or potentially employable in
the dairy industry, ranging from well-established
classical control approaches to more advanced tech-
niques based on recent theoretical directions of
knowledge processing.
Specificity of the Dairy Industry
The dairy industry, like other industrial areas, has its
own specicity. Processing plants are composed of
a large number of interlinked process units, operating
mostly in batch and fed-batch modes, designed for
specic processing such as mixing of various in-
gredients, pasteurizing, homogenizing, ripening, a-
vouring, freezing and packaging. Each individual
piece of equipment normally has certain schedules for
performing different tasks such as reacting, mixing,
heating, holding and cooling. This whole structure
requires both local control at unit level and optimal
scheduling strategies at sector and plant levels.
In addition, the high variability of ingredients and
the lack of a complete scientic understanding of the
manufacturing processes render difcult the applica-
tion of traditional process control structures.
The role of biological know-how in improving
process operation was for many years the dominant
factor, and correctly so. Lactic fermentation serves
well as an illustration. This is the main biological
process that occurs in the production of products such
as yoghurt, butter, cheese and several other milk-
derived products. Lactic acid bacteria grow readily in
milk. Until pasteurization was developed in the late
nineteenth century it was hard to keep milk fresh.
From a biological point of view, deeper basic insight
into the metabolism and genetics of lactic acid bac-
teria enabled the modication of the characteristics of
these bacteria, thereby improving shelf-life, as well as