Nof S.Y. Springer Handbook of Automation

Подождите немного. Документ загружается.

Sensors and Sensor Networks 20.1 Sensors 335

Fig. 20.1 Cutaway view of an LVDT. Current is driven

through the primary coil at A, causing an induction current

to be generated through the secondary coils at B

electromotive force (EMF) in any closed circuit is equal

to the time rate of change of the magnetic ﬂux through

the circuit. Quantitatively, the law takes the following

form

E =−

dΦ

(20.5)

where E is the electromotive force (EMF)andΦ

is the

magnetic ﬂux through the circuit.

Besides these types of sensors, thermocouples mea-

sure the temperature difference between two points

rather than absolute temperature. In traditional appli-

cations, one of the junctions (the cold junction) is

maintained at a reference temperature, while the other

end is attached to a probe. Having available a cold junc-

tion at a known temperature is simply not convenient

for most directly connected control instruments. They

incorporate into their circuits an artiﬁcial cold junc-

tion using some other thermally sensitive device, such

as a thermistor or diode, to measure the temperature

of the input connections at the instrument, with special

care being taken to minimize any temperature gradient

between terminals. Hence, the voltage from a known

cold junction can be simulated, and the appropriate

correction applied. Photodiodes, phototransistors, and

photo-interrupters are sensors that use the photoelectric

effect. Other types of sensors are listed in Table 20.1.

20.1.2 Position, Velocity,

and Acceleration Sensors

Sensors can also be classiﬁed by the physical phenom-

ena measured, such as position, velocity, acceleration,

heat, pressure, ﬂow rate, sound, etc. This classiﬁcation

of sensors is brieﬂy explained below.

Position Sensors

A position sensor is any device that enables position

measurement. Position sensors include limit switches

or proximity sensors that detect whether or not some-

thing is close to or has reached a limit of travel. Position

sensors also include potentiometers that measures ro-

tary or linear position. The linear variable differential

transformer (LVDT) is an example of the potentiome-

ters for measuring linear displacement, while resolvers

and opticalencoders measure the rotaryposition of a ro-

tating shaft. The LVDT and resolver function much

like a transformer. The optical encoder produces digi-

tal signals that convert motion into a sequence of digital

pulses. In fact, there also exist optical encoders for mea-

suring linear motion.

Some position sensors are classiﬁed by their

measuring techniques. Sonars measure distance with

sonic/ultrasonic waves and radar utilizes electronic/

radio to detect or measure the distance between two ob-

jects. Many other sensors are used to measure position

or distance.

Velocity Sensors

Speed measurement can be obtained by taking consec-

utive position measurements at known time intervals

and computing the derivative of the position values.

A tachometer is an example of a velocity sensor that

does this for a rotating shaft. The typical dynamic time

constant of a tachometer is in the range 10–100μs.

A tachometer is a passive analog sensor that pro-

vides an output voltage proportional to the velocity of

a shaft. There is no need for an external reference or

excitation voltage. Traditionally tachometers have been

used for velocity measurement and control only, but

all modern tachometers have quadratic outputs which

are used for velocity, position, and direction measure-

ments, making them effectively functional as position

sensors.

Acceleration Sensors

An acceleration sensor or accelerometer is a sensor

designed to measure continuous mechanical vibration

such as aerodynamic ﬂutter and transitory vibration

such as shock waves, blasts or impacts. Accelerom-

eters are normally mechanically attached or bonded

to an object or structure for which acceleration is to

be measured. The accelerometer detects acceleration

along one axis and is insensitive to motion in orthog-

Part C 20.1

336 Part C Automation Design: Theory, Elements, and Methods

onal directions. Strain gages or piezoelectric elements

constitute the sensing element of an accelerometer,

converting vibration into a voltage signal. The de-

sign of an accelerometer is based on the inertial

effects associated with a mass connected to a moving

object.

Detailed information and technical working pro-

cesses about position, velocity, and acceleration sensors

can be found in many references [20.5,6].

20.1.3 Miscellaneous Sensors

There are other groups of sensors to measure physical

quantities such as force, strain, temperature, pressure,

and ﬂow. Force sensors are represented by a load cell

that is used to measure a force. The load cell consists of

several strain gagesconnected toa bridgecircuit toyield

a voltage proportional to the load. Temperature sensors

are devices that indirectly measure quantities such as

pressure, volume, electrical resistance, and strain and

then convert the values using the physical relationship

between the quantity and temperature; for example:

(a) a bimetallic strip composed of two metal layers with

different coefﬁcients of thermal expansion utilizes the

differencein thethermal expansion of thetwometal lay-

ers, (b) a resistance temperature sensor constructed of

metallic wire wound around a ceramic or glass core and

hermetically sealed utilizes the resistance change of the

metallic wire with temperature, and (c) a thermocouple

constructed by connecting two dissimilar metals in con-

a) b)

c) d)

Fig. 20.2a–d Various sensors: (a) absolute encoder, (b) photoresis-

tor,

Robots)

tact produces a voltage proportional to the temperature

of the junction [20.7,8].

Flow Sensors

A ﬂow sensor is a device for sensing the rate of ﬂuid

ﬂow. In general, a ﬂow sensor is the sensing element

used in a ﬂow meter to record the ﬂow of ﬂuids. Some

ﬂowsensors have a vanethat is pushed by the ﬂuid (e.g.,

a potentiometer), while other ﬂow sensors are based on

heat transfer caused by the moving medium.

Ultrasonic Sensors

Ultrasonic sensors or transducer generates high-

frequency sound waves and evaluate the echo received

back by the sensor. An ultrasonic sensor computes the

time interval between sending the signal and receiving

the echo to determine the distance to an object. Radar

or sonar works on a principal similar to that of the ul-

trasonic sensor. Some sensors are depicted in Fig.20.2.

However, there are many other groups of sensors not

listed in this section. With the advent of semiconduc-

tor electronics and manufacturing technology, sensors

have become miniaturized and accurate, and brought

into existence micro/nanosensors.

Vision Sensors

Another widely used sensor is a vision sensor. A vision

sensor is typically used embedded in a vision system.

A vision system can be used to measure shape, ori-

entation, area, defects, differences between parts, etc.

Vision technology has improved signiﬁcantly over the

last decade in that they have become rather standard

smart sensing components in most factory automation

systems for part inspection and location detection. In

general, a vision system consists of a vision camera,

an image processing computer, and a lighting system.

The basic principle of operation of a vision system is

that it forms an image by measuring the light reﬂected

from objects, and the sensor head analyzes the output

voltage from the light intensity received. The sensor

head consists of an array of photosensitive, photodiodes

or charge-coupled devices (CCD). Currently, various

signal-processing techniques for the reﬂected signals

are applied for many industrial applications to provide

accurate outputs, as illustrated in Fig.20.3.

20.1.4 Micro- and Nanosensors

A microsensoris a miniature electronic device function-

ing similar to existing large-scale sensors. With recent

micro-electromechanical system (MEMS) technology,

Part C 20.1

Sensors and Sensor Networks 20.1 Sensors 337

a) b)

c) d)

Fig. 20.3a–d Types of vision sensor applications: (a) automated low-volume/high-variety production, (b) vision sensors

for error-proof oil cap assembly,

◦

inspection, (d) inspection of two-dimensional (2-D)

matrix-marked codes (courtesy of Cognex Corp.)

Dot 2001

demo scale

Wec 1999

smart rock

Spec 2003

mote on a chip

Mica2 2002

Mica 2002Rene 2000

Fig. 20.4 Evolution of smart wireless microsensors (courtesy of Crossbow Technology Inc.)

microsensors are integrated with signal-processing cir-

cuits, analog-to-digital (A/D)converters,programmable

memory, and a microprocessor, a so-called smart

microsensor [20.9,10]. Current smart microsensors

contain an antenna for radio signal transmission. Wire-

less microsensors are now commercially available and

are evolving with more powerful functionalities, as il-

lustrated in Fig.20.4.

In general, a wireless microsensor consists of

a sensing unit, a processing unit, a power unit, and com-

munication elements. The sensing unit is an electrical

part detecting the physical variable from the envi-

ronment. The processing unit (a tiny microprocessor)

performs signal-processing functions, i.e., integrating

Node Node

Node

Processing unit

(processor,

memory)

Receiver Transmitter Amplifier

Sensing unit

(sensor, ADC)

Power unit

Fig. 20.5 Wireless micronode model. Each node has a sensing

module (analog-to-digital converter (ADC)), processing unit, and

communication elements

Part C 20.1

338 Part C Automation Design: Theory, Elements, and Methods

Fig. 20.6 Three-dimensional (3-D)model of three types of

single-walled carbon nanotubes, like those used to make

certain nanosensors (created by Michael Ströck on Febru-

ary 1, 2006)



data and computation required in the processing of

information. The communication elements consist of

a receiver, a transmitter, and an ampliﬁer if needed.

The power unit provides energy source with other units

(Fig.20.5). Basically, all individual sensor nodesare op-

erated by a limited battery, but a base-station node as

a ﬁnal data collecting center can be modeled with an

unlimited energy source.

Under the microscale, nanosensors are used in

chemical and biological sensory applications to de-

liver information about nanoparticles. As an example,

nanotubes are used to sense various properties of

gaseous molecules, as depicted in Fig.20.6. In develop-

ing and commercializing nanosensors, developers still

need to overcome high costs of production and reliabil-

ity challenges. In the near future, there is tremendous

(7,10) nanatube

(chiral)

(10,10) nanatube

(armchair)

(0,10) nanatube

(zig-zag)

room to enhance the technology and implement various

nanosensors in real-life applications.

20.2 Sensor Networks

20.2.1 Sensor Network Systems

Before the advent of microminiaturization technology,

single-sensor systems had an important role in a variety

of practical applications because they were relatively

easy to construct and analyze. Single-sensor systems,

however, were the only solution when there was critical

limitation of implementation space. Moreover, single-

sensor systems for recently emerging applications have

various limitations and disadvantages:

•

They have limited applications and uses; for in-

stance, if a systemshould measure severalvariables,

e.g., temperature, pressure, and ﬂow rate, at the

same time in the application, single-sensor systems

are insufﬁcient.

•

They cannot tolerate a variety of failures which may

take place unexpectedly.

•

A single sensor cannot guarantee timely delivery

of accurate information all of the time because it

is inevitably affected by noise and other uncertain

disruptions.

These limitations are critical when a system requires

highly reliable and timely information. Therefore,

single-sensor systems are not suitable when robust and

accurate information is required in the application.

To overcome the critical disadvantages of single-

sensor systems in most applications, multisensor net-

work systems whichrequire replicatedsensory informa-

tion have been studied, along with their communication

network technologies. Replicated sensor systems are

applicable not only because microfabrication technol-

ogy enables production of various microsensors at

low manufacturing cost, but also because microsen-

sors can be embedded in a system with replicated

deployment. These redundantly deployed sensors en-

able a system to improve accuracy and tolerate sensor

failure, i.e., distributed microsensor arrays and net-

works (DMSA/DMSN) are built from collections of

spatially scattered microsensor nodes. Each node has

the ability to measure the local physical variable within

its accuracy limit, process the raw sensory data, and

cooperate with its neighboring nodes.

Sensors incorporated with dedicated signal-process-

ing functions are called intelligent, or smart, sensors.

The main roles of dedicated signal processing functions

are to enhance design ﬂexibility and realize new sens-

ing functions. Additional roles are to reduce loads on

central processing units and signaltransmission lines by

Part C 20.2

Sensors and Sensor Networks 20.2 Sensor Networks 339

distributing information processing to the lower layers

of the system [20.10].

A set of microsensors deployed close to each other

to measure the same physical quantity of interest is

called a cluster. Sensors in a cluster can be either of

the same or different type to form a distributed sen-

sor network (DSN). A DSN can be utilized in a widely

distributed sensor system and implemented as a locally

concentrated conﬁguration with a high density.

20.2.2 Multisensor Data Fusion Methods

There are three major ways in which multiple sensors

interact [20.11, 12]: (1) complementary, when sensors

do not depend on each other directly, but are combined

to give a more complete image of the phenomena being

studied; (2) competitive, when sensors provide indepen-

dent measurement of the same information regarding

a physical phenomenon; and (3) cooperative,whensen-

sors combine data from independent sensors to derive

information that would be unavailable from the individ-

ual sensors.

In orderto combineinformation collectedfrom each

sensor, various multisensory data fusionmethods can be

applied. Multisensor data fusion is the process of com-

bining observations from a number of different sensors

to provide a robust and complete description of an envi-

ronment or process of interest. Most current data fusion

methods employ probabilistic descriptions of observa-

tions and processes and use Bayes’ rule to combine this

information [20.13,14].

Bayes’ Rule

Bayes’ rule lies at the heart of most data fusion meth-

ods. In general, Bayes’ rule provides a means to make

inferences about an object or environment of interest

described by a state x, given an observation z. Based

on the rule of conditional probabilities, Bayes’ rule is

obtained as

P(x|z) =

P(z|x)P(x)

P(z)

(20.6)

The conditional probability P(z|x) serves the role of

a sensor model. The probability is constructed by ﬁx-

ing the value of x = x and then asking what probability

density P(z|x = x)onx is inferred. The multisensory

form of Bayes’ rule requires conditional independence

P(z

,...,z

|x) = P(z

|x)...P(z

|x)

i=1

P(z

) . (20.7)

The recursive form of Bayes’ rule is

P(x|Z

) =

P(z

|x)P(x|Z

k−1

)



k−1



(20.8)

From this equation, one needs to compute and store

only the posterior density P(x|Z

k−1

), which contains

a complete summary of all past information.

Probabilistic Grids

Probabilistic grids are the means to implement the

Bayesian data fusion technique to problems in map-

ping [20.15] and tracking [20.16]. Practically, a grid

of likelihoods on the states x

is produced in the form

P(z = z|x

) =Λ(x

). It is then trivial to apply Bayes’

rule to update the property value at each grid cell as

) =CΛ(x

)P(x

) ∀i, j , (20.9)

where C is a normalizing constantobtained by summing

posterior probabilities to 1 at node ij only. Compu-

tationally, this is a simple pointwise multiplication of

two grids. Grid-based fusion is appropriate to situations

where the domain size and dimension are modest. In

such cases, grid-based methods provide straightforward

and effective fusion algorithms. Monte Carlo and par-

ticle ﬁltering methods can be considered as grid-based

methods, where the grid cells themselves are samples of

the underlying probability density for the state.

The Kalman Filter

The Kalman ﬁlter is a recursive linear estimator that

successively calculates an estimate for a continuous-

valued state on the basis of periodic observations of

the state. The Kalman ﬁlter may be considered a spe-

ciﬁc instance of the recursive Bayesian ﬁlter [20.17]for

the case where the probability densities on states are

Gaussian.

The Kalman ﬁlter algorithm produces estimates that

minimize mean-squared estimation error conditioned

on a given observation sequence and so is the condi-

tional mean

x(i|j)

 E[x(i)|z(1),...,z( j)]  E[x(i)|Z

(20.10)

The estimate variance is deﬁned as the mean-squared

error in this estimate

P(i|j)

 E



[x(i)−

x(i|j)][x(i)−

x(i|j)]





(20.11)

The estimate of the state at a time k, given all infor-

mation up to time k, is written

x(k|k). The estimate of

Part C 20.2

340 Part C Automation Design: Theory, Elements, and Methods

the state at a time k given only information up to time

k−1 iscalled a one-step-ahead prediction and is written

x(k|k −1).

The Kalman ﬁlter is appropriate to data fusion prob-

lems where the entity of interest is well deﬁned by

a continuous parametric state. Thus, it would be use-

ful to estimate position, attitude, and velocity of an

object, or the tracking of a simple geometric feature.

Kalman ﬁlters, however, are inappropriate for estimat-

ing properties such as spatial occupancy, discrete labels

or processes whose error characteristics are not easily

parameterized.

Sequential Monte Carlo Methods

The sequential Monte Carlo (SMC) ﬁltering method is

a simulation of the recursive Bayes update equations

using sample support values and weights to describe

the underlying probability distributions. SMC recur-

sion begins with an a posterior probability density

represented by a set of support values and weights

k−1

k−1|k−1

}

k−1

i=1

in the form



k−1



k−1



i=1

k−1



k−1

−x

k−1



(20.12)

Leaving the weights unchanged w

k−1

and allow-

ing the new support value x

to be drawn on the basis of

old support value x

k−1

, the prediction becomes



k−1



k−1



i=1

k−1



−x



(20.13)

The SMC observation update step is relatively straight-

forward and described in [20.13,18].

SMC methods are well suited to problems where

state-transition models and observation models are

highly nonlinear. However, they are inappropriate for

problems where the state space is of high dimension.

In addition, the number of samples required to model

a given density faithfully increases exponentially with

state-space dimension.

Interval Calculus

Interval representation of uncertainty has a number of

potential advantages over probabilistic techniques. An

interval to bound true parameter values provides a good

measure of uncertainty in situations where there is

a lack of probabilistic information, but in which sen-

sor and parameter error is known to be bounded. In this

technique, the uncertainty in a parameter x is simply de-

scribed by a statement that the true value of the state x is

known to be bounded between a and b,i.e.,x ∈[a, b].

There is no other additional probabilistic structure im-

plied. With a, b, c, d ∈

R, interval arithmetic is also

possible as

d([a, b], [c, d]) =max(|a−c|, |b−d|) .

(20.14)

Interval calculus methods are sometimes used for detec-

tion, but are not generally used in data fusion problems

because of the difﬁculties to get results converged to

anything value, and to encode dependencies between

variables.

Fuzzy Logic

Fuzzy logic has achieved widespread popularityfor rep-

resenting uncertainty in high-level data fusion tasks.

Fuzzy logic provides an idealtool for inexact reasoning,

particularly in rule-based systems. In the conventional

logic system, a membership function μ

(x) (also called

the characteristic function) is deﬁned. Then the fuzzy

membership function assigns a value between 0 and 1,

indicating the degree of membership of every x to the

set A. Composition rules for fuzzy sets follow the com-

position processes for normal crisp sets as

A∩ B

 μ

A∩B

(x) =min[μ

(x),μ

(x)], (20.15)

A∪ B  μ

A∪B

(x) =max[μ

(x),μ

(x)]. (20.16)

The relationship between fuzzy set theory and probabil-

ity is, however, still debated.

Evidential Reasoning

Evidential reasoning methods are qualitatively different

from either probabilistic methods or fuzzy set theory. In

evidential reasoning, belief mass cannot only be placed

on elements and sets, but also sets of sets, while in

probability theory a belief mass may be placed on any

element x

∈χ and on any subset A ⊆χ. The domain of

evidential reasoning is the power set 2

. Evidential rea-

soning methods play an important role in discrete data

fusion, attribute fusion, and situation assessment, where

information may be unknown or ambiguous.

Multisensory fusion methods and their models are

summarized in Table 20.2, and details can be founded

in [20.13,14].

In addition, multisensor integration or fusion is not

only the process of combining inputs from sensors with

information from other sensors, but also the logical pro-

cedure of inducing optimal output from multiple inputs

with one representative format [20.19]. In the fusion of

Part C 20.2

Sensors and Sensor Networks 20.2 Sensor Networks 341

Table 20.2 Multisensor data fusion methods [20.13]

Approach Method Fusion model and rule

Probabilistic modeling Bayes’ rule P(x|Z

) =

P(z

|x)P(x|Z

k−1

)



k−1



Probabilistic grids P

) =CΛ(x

)P(x

)

The Kalman ﬁlter

P(i|j)

 E



[x(i)−

x(i|j)][x(i)−

x(i|j)]





Sequential Monte Carlo methods





(

i=1

k−1



= z

= x





−x



Nonprobabilistic modeling Interval calculus d([a, b], [c, d]) =max(|a−c|, |b−d|)

Fuzzy logic

A∩B μ

A∩B

(x) =min[μ

(x),μ

(x)]

A∪B μ

A∪B

(x) =max[μ

(x),μ

(x)]

Evidential reasoning

={{occupied, empty},...

{occupied}, {empty}, 0}

large-size distributed sensor networks, an additional ad-

vantage of multisensor integration (MSI) is the ability

to obtain more fault-tolerant information. This fault tol-

erance is based on redundant sensory information that

compensates for faulty or erroneous readings of sen-

sors. There are several types of multisensor fusion and

integration methods, depending on the types of sensors

and their deployment [20.20]. This topic has received

increasing interest in recent years because of the sen-

sibility of networks built with many low-cost micro-

and nanosensors. As an example, a recent improve-

ment of the fault-tolerance sensor integration algorithm

(FTSIA)byLiu and Nof [20.21,22] enables it not only

to detect possibly faulty sensors and widely faulty sen-

sors, but also to generate a ﬁnal data interval estimate

from the correct sensors after removing the readings of

those faulty sensors.

20.2.3 Sensor Network Design

Considerations

Sensor networks are somewhat different from tra-

ditional operating networks because sensor nodes,

especially microsensors, are highlyprone to failure over

time. As sensor nodes weaken or even die, the topol-

ogy of the active sensor networks changes frequently.

Especially when mobility is introduced into the sen-

sor nodes, maintaining the robustness and discovering

topology consistently become challenging. Therefore,

the algorithms developed for sensor network commu-

nication and task administration should be ﬂexible and

stable against changes of network topology and work

properly under unexpected failure of sensors.

In addition, in order to be used in most applications,

DSN systems should be designed with application-

speciﬁc communicationalgorithms andtask administra-

tion protocols because microsensors and their network-

ing systems are extremely resource constrained. There-

fore, most research efforts have focused on application-

speciﬁc protocols with respect to energy consumption

and network parameters such as node density, ra-

dio transmission range, network coverage, latency, and

distribution. Current network protocols also use broad-

casting for communication, while traditional and ad hoc

networks use point-to-point communication. Hence, the

routing protocols, in general, should be designed by

considering crucial sensor network features as follows:

1. Fault tolerance: Over time, sensor nodes may fail or

be blocked due to lack of power, physical damage

or environmental interference. The failure of sensor

nodes, however, should not affect the overall oper-

ation of the sensor network. Thus, fault tolerance

or reliability is the ability to sustain sensor network

functionality despite likely problems.

2. Accuracy improvement: Redundancy of information

can reduce overall uncertainty and increase the ac-

curacy with which eventsare perceived.Since nodes

located close to each other are combining informa-

tion about the same event, fused data improve the

quality of the event information.

3. Timeliness: DSN can provide the processing paral-

lelism that may be needed to achieve an effective

integration process, either at the actual speed that

a single sensor could provide, or at even faster oper-

ation speed.

4. Network topology: A large number of nodes de-

ployed throughout the sensory ﬁeld should be

maintained by carefully designed topology because

any changes in sensor nodes and their deployments

Part C 20.2

342 Part C Automation Design: Theory, Elements, and Methods

affect the overall performance of DSN. Therefore,

a ﬂexible and simple topology is usually preferred.

5. Energy consumption: Since each wireless sensor

node is working with a limited power source, the

design of power-saving protocols and algorithms is

a signiﬁcant issue for providing longer lifetime of

sensor network systems.

6. Lower cost: Despite the use of redundancy, a dis-

tributed microsensor system obtains information

at lower cost than the equivalent information ex-

pected from a single sensor because it does not

require the additional cost of functions to obtain the

same reliability and accuracy. The current cost of

a microsensor node, e.g., dust mote [20.9], is still

expensive (US$25–172), but it is expected to be

less than US$1 in the near future, so that sensor

networks can be justiﬁed.

7. Scalability: The coverage area of a sensor network

system depends on the transmission range of each

node and the density of the deployed sensors. The

density of the deployed nodes should be carefully

designed to provide a topology appropriate for the

speciﬁc application.

To provide the optimal solution to meet these de-

sign criteria in the sensor network, researchers have

considered various protocols and algorithms. However,

none of these studies has been developed to improve all

Cluster

BS BS

Cluster

head

Cluster

head

a) b)

BS BS

c) d)

Fig. 20.7a–d Four different conﬁgurations of wireless sensor net-

works:

(a) single hop with clustering, (b) multihop with clustering,

ing. BS – base station

design factors because the design of a sensor network

system has typically been application speciﬁc.

A distributed network of microsensor arrays (MSA)

can yield more accurate and reliable results based on

built-in redundancy. Recent developments of ﬂexible

and robust protocols with improved fault tolerance will

not only meet essential requirements in distributed sys-

tems but will also provide advanced features needed

in speciﬁc applications. While MEMS sensor technol-

ogy hasadvancedsigniﬁcantly in recent years, scientists

now realize the need for design of effective MEMS sen-

sor communication networks and task administration.

20.2.4 Sensor Network Architectures

A well-designed distributed network of microsensor ar-

rays can yield more accurate and reliable results based

on built-inredundancy. Recent developments of ﬂexible

and robust protocols with improved fault tolerance will

not only meet essential requirements in distributed sys-

tems but will also provide advanced features needed in

speciﬁc applications. They can produce widely acces-

sible, reliable, and accurate information about physical

environments.

Various architectures have been proposed and de-

veloped to improve the performance of systems and

fault-tolerance functionality of complex networks de-

pending on their applications. General DSN structures

for multisensor systems were ﬁrst discussed by Wes-

son et al. [20.23]. Iyengar et al. [20.24]andNadig

et al. [20.25] improved and developed new architectures

for distributed sensor integration.

A network is a general graph G = (V, L), where V

is a set of nodes (or vertices) and L is a set of com-

municating links (or edges) between nodes. For a DSN,

a node means an intelligent sensing node consisting

of a computational processor and associated sensors,

and an edge is the connectivity of nodes. As shown

in Fig.20.7,aDSN consists of a set of sensor nodes,

a set of cluster-head (CH) nodes, and a communi-

cation network interconnecting the nodes [20.20, 24].

In general, one sensor node communicates with more

than one CH, and a set of nodes communicating with

a CH is called a cluster. A clustering architecture can

increase system capacity and enable better resource

allocation [20.26, 27]. Data are integrated in CH by

receiving required information from associated sen-

sors of the cluster. In the cluster, CHs can interact

not only with other CHs, but also with higher-level

CHs or a base station. A number of network con-

ﬁgurations have been developed to prolong network

Part C 20.2

Sensors and Sensor Networks 20.2 Sensor Networks 343

lifetime and reduce energy consumption in forward-

ing data. In order to minimize energy consumption,

routing schemes can be broadly classiﬁed into two cat-

egories: (1) clustering-based data forwarding scheme

(Fig.20.7a,b) and (2) multihop data forwarding scheme

without clustering (Fig.20.7b,d).

In recent years, with the advancement of wireless

mobile communication technologies, ad hoc wireless

sensor networks (AWSNs) have become important.

With this advancement, the above wired (microwired)

architectures remain relevant only where wireless com-

munication is physically prohibited; otherwise, wireless

architectures are considered superior. The architecture

of AWSN is fully ﬂexible and dynamic, that is, a mo-

bile ad hoc network represents a system of wireless

nodes that can freely reorganize into temporary net-

works as needed, allowing nodes to communicate in

areas with no existing infrastructure. Thus, interconnec-

tion between nodes can be dynamically changed, and

the network is set up only for a short period of com-

munication [20.28]. Now the AWSN with an optimal

ad hoc routing scheme has become an important design

concern.

In applications where there is no given pattern of

sensor deployment, such as battleﬁeld surveillance or

environmental monitoring, the AWSN approach can

provide efﬁcient sensor networking. Especially in dy-

namic network environments such as AWSN, three

main distributed services, i. e., lookup service, com-

position service, and dynamic adaptation service by

self-organizingsensor networks, arealso studiedto con-

trol the system (see, for instance, [20.29]).

In order to route information in an energy-efﬁcient

way, directed diffusion routing protocols based on the

localized computation model [20.30, 31] have been

studied for robust communication. The data consumer

will initiate requests for data with certain attributes.

Nodes will then diffuse the requests towards producers

via a sequence of local interactions. This process sets

up gradients in the network which channel the delivery

of data. Even though the network status is dynamic, the

impact of dynamics can be localized.

A mobile-agent-based DSN (MADSN) [20.32] uti-

lizes a formal concept of agent to reduce network

bandwidth requirements. A mobile agent is a ﬂoating

processor migrating from node to node in the DSN and

performing data processing autonomously. Each mo-

bile agent carries partially integrated data which will

be fused at the ﬁnal CH with other agents’ informa-

tion. To save time and energy consumption, as soon as

certain requirements of a network are satisﬁed in the

progress of its tour, the mobile agent returns to the base

station without having to visit other nodes on its route.

This logic reduces network load, overcoming network

latency, and improves fault-tolerance performance.

20.2.5 Sensor Network Protocols

Communication protocols for distributed microsensor

networks provide systems with better network capa-

bility and performance by creating efﬁcient paths and

accomplishing effective communication between the

sensor nodes [20.29,33,34].

The point-to-point protocol (PTP) is the simplest

communication protocol and transmits data to only one

of its neighbors, as illustrated in Fig. 20.8a. However,

PTP is not appropriate for a DSN because there is no

communication path in case of failure of nodes or links.

In the ﬂooding protocol (FP), the information sent

out by the sender node is addressed to all of its neigh-

bors, as shown in Fig.20.8b. This disseminates data

quickly in a network wherebandwidth is not limitedand

links are not loss-prone. However, since a node always

sends data to its neighbors, regardless of whether or not

the neighbor has already received the data from another

source, it leads to the implosion problem and wastes re-

sources by sending duplicate copies of data to the same

node.

The gossiping protocol (GP) [20.35, 36]isanal-

ternative to the classic ﬂooding protocol in which,

instead of indiscriminately sending information to all

its neighboring nodes, each sensor node only forwards

the data to one randomly selected neighbor, as depicted

in Fig.20.8c. While the GP distributes information

more slowly than FP, it dissipates resources, such as

energy, at a relatively lower rate. In addition, it is not as

robust relative to link failures as a broadcasting protocol

(BP), because a node can only rely on one other node to

resend the information for it in the case of link failure.

In order to solve the problem of implosion and

overlap, Heinzelman et al. [20.37] proposed the sensor

a) PTP b) FP c) GP

Fig. 20.8a–c Three basic communication protocols: (a) point-to-

point protocol (PTP),

(b) ﬂooding protocol (FP), and (c) gossiping

protocol (GP)

Part C 20.2

344 Part C Automation Design: Theory, Elements, and Methods

protocol for information via negotiation (SPIN). SPIN

nodes negotiate with each other before transmitting

data, which helps ensure that only useful transmission

of information will be executed. Nodes in the SPIN pro-

tocol use threetypes ofmessages to communicate:ADV

(new data advertisement), REQ (request for data), and

DATA (data message). Thus, SPIN protocol works in

three stages: ADV–REQ–DATA. The protocol begins

when a node advertises the new data that is ready to

be disseminated. It advertises by sending an ADV mes-

sage to its neighbors, naming thenew data (ADV stage).

Upon receiving an ADV, the neighboring node checks

to see whether it has already received or requested the

advertised data to avoidimplosion and the overlap prob-

lem. If not, it responds by sending a REQ message for

the missing data back to the sender (REQ stage). The

protocol completes when the initiator of the protocol re-

sponds to the REQ with a DATA message, containing

the missing data (DATA stage).

In a relatively large sensor network, a cluster-

ing architecture with a local cluster-head (CH)is

necessary. Heinzelman et al. [20.38] proposed the low-

energy adaptive clustering hierarchy (LEACH), which

is a clustering-based protocol that utilizes randomized

rotation oflocal clusterbase stations to evenly distribute

the energy load of sensors in DSN. Energy-minimizing

routing protocols have also been developed to extend

the lifetime of the sensing nodes in a wireless network;

for example, a minimum transmission energy (MTE)

routing protocol [20.39] chooses intermediate nodes

such that the sum of squared distances is minimized by

assuming a square-of-distance power loss between two

nodes. This protocol, however, results in unbalanced

termination of nodes with respect to the entire network.

Sensor level Middleware level Application level

• Data

dessamination

• Self-configuration

• Sensor rate

configuration

• Communication

scheduling

• Monitoring the

battery power

• Sensor state

control

• Cluster control

messages

• Resource context

• Cluster context

• Energy context

• Temporal context

Context

management

• Error handling

• Service differenciation

• Network control

• Event detection and

coordination

• Middleware component

control

Qos control

• Service interpreter

• Application

specification

• Qos requirements

• Data aggregation

• Analysis service

Services

• Analyze data

• Manage database

• Application

requirements

• Industrial application

classification

• Trade-offs

Application manager

Networked

sensors

Low-level

data abstraction

Fig. 20.9 Middleware architecture for facility sensor network applications (MidFSN) (after [20.36])

In recent years, a time-based network protocol

has been developed. The objective of a time-based

protocol is to ensure that, when any tasks keep the

resource idle for too long, their exclusive service by

the resource is disabled; that is, the time-based control

protocol is intended to provide a rational collabora-

tion rule among tasks and resources in the networked

system [20.40]. Here, slow sensors will delay timely

response and other sensors may need to consume ex-

tra energy. The patented fault tolerant time-out protocal

(FTTP) uses the basic concept of a time-out scheme

effectively in a microsensor communication control

(FTTP is a patent-pending protocol of the PRISM Cen-

ter at Purdue University, USA).

The design of industrial open protocols for mostly

wired communication known as ﬁeldbuses, such as De-

viceNet and ControlNet, have also been evolved to

provide open data exchange and a messaging frame-

work [20.41]. Further development for wireless has

been investigated in asset monitoring and maintenance

using an open communication protocol such as Zig-

Bee [20.42].

Wireless sensor network application designers also

require a middleware to deliver a general runtime

environment that inherits common requirements un-

der application speciﬁcations. In order to provide

robust functions to industrial applications under some-

what limited resource constraints and the dynamics

of the environment, appropriate middleware is also

required to contain embedded trade-offs between es-

sential quality-of-service (QoS) requirements from

applications. Typically, a sensor network middleware

has a layered architecture that is distributed among

the networked sensor systems. Based on traditional

Part C 20.2