Nof S.Y. Springer Handbook of Automation

Подождите немного. Документ загружается.

Machining Lines Automation 35.5 Conclusion and Perspectives 615

Table 35.5 The ranges of places Q(i) for operations

Operation i 1 2 3 4 5 6 7 8 9 10

Q(i) [1,23] [3,23] [1,15] [3,23] [3,29] [3,29] [9,29] [16,29] [16,29] [16,30]

while minimizing the number of additional machines

and/or those that move from a workstation to another

in order to rebalance the line. This problem is similar

to the design problem considered in the previous sec-

tions. Therefore, the proposed MIP model (35.1–35.17)

can be easily adapted for this new problem.

The following modiﬁcations are made in the model

(35.1–35.17): a new objective function is considered

(35.1’) along with additional constraints (35.18)and

a new set of variables (35.19) which represents the gap

between the number of machines for each station in the

line before reconﬁguration (y

Old

), and their number in

the line reconﬁgured.

Minimize



k=1



+δ

−



(35.1’)



n=1

n ·y

−



n=1

n ·y

Old

=δ

−δ

−

∀k =1, 2,...,m

, (35.18)

,δ

−

≥0 , ∀k = 1, 2,...,m

(35.19)

35.5 Conclusion and Perspectives

Transfer lines are used inmany manufacturing domains,

especially in machining systems, to efﬁciently effectu-

ate high-quality and economical production. In today’s

competitive business environment, several manufactur-

ers have opted for transfer lines to beneﬁt from their

advantages, namely precision, quality, productivity, re-

duction of handling cost, etc. However, transfer lines

also present some drawbacks, such as requiring a large

investment. Normally, transfer lines are highly auto-

mated, but the level of automation depends on the type

of customer demand. Three types of transfer lines ex-

ist: dedicated, ﬂexible, and reconﬁgurable. Dedicated

lines are composed of workstations with multispindle

heads. Flexible transfer lines have several types of CNC

machines. Reconﬁgurable lines offer a mix of different

types of machines (special machines, CNC machines,

machining units, etc.) and can have different architec-

tures (simple line, U-line, parallel stations, etc.).

With increasing technological progress and de-

velopment of ever more sophisticated and efﬁcient

machining equipment, the problem of automated ma-

chining line design is exceptionally pertinent. Indeed,

the concepts for machining lines are continuously

improved through the development of new types of ar-

chitectures and machines. Unfortunately, there is a gap

between industrial cases and research problems treated.

In contrast with assembly systems, in the domain of

machining lines, the gap is often due to the lack

of collaboration between the industrial and academic

worlds.

In this chapter, written jointly by academic

(Ecole des Mines de Saint Etienne) and industrial

(PCI/SCEMM) partners, a general overview of the au-

tomated machining lines is presented. The principal

characteristics of these lines and a general methodol-

ogy for their design are introduced. This methodology

is valid independent of the type of the line: dedicated,

ﬂexible or reconﬁgurable.The goal isto help machining

line manufacturers to design efﬁcient lines and become

more competitive.

On receipt of a customer demand for a line which

includes a part description (plans, characteristics, etc.)

and the required output, the machining line manufac-

turer must be able to proposea complete solution within

a very short time interval. This preliminary solution

concerns the line architecture, number of machines,

and equipment, with a line cost evaluation. A major

difﬁculty deals with line balancing, which is a hard

combinatorial optimization problem. All types of trans-

fer lines are concerned. In this chapter, a short survey

of general line balancing approaches is given. The

methods developed for the balancing of the automated

machining lines are enumerated and commented. Then,

an industrial case study is presented. It illustrates and

highlights the importanceof the line balancing problem.

Afterwards, a mixed integer program for the considered

case is proposed. The presented model and approach

are useful from a practical perspective. They gener-

ate more appropriate preliminary solutions to customer

needs within a very short timeframe. From an academic

Part D 35.5

616 Part D Automation Design: Theory and Methods for Integration

point of view, this is a new formulation of the line bal-

ancing problem with sequence-dependent setup times

and parallel machines.

For future research work, beside the improvement

of the models and resolution algorithms, numerous re-

search perspectives have yet to be studied in this ﬁeld.

Among them, the combination of optimization methods

and discrete-event simulation seems promising. Simu-

lation is a powerful method to illustrate and study the

ﬂow of material on the line and to determine the effect

of its architecture on line reliability and performance.

Also, the development of interactive and iterative soft-

ware could provide useful decision-aid systems for

industry.

References

35.1 M.P. Groover: Automation, Production Systems and

Computer Integrated Manufacturing (Prentice Hall,

Eaglewood Cliffs 1987)

35.2 R.G. Askin, C.R. Standridge: Modeling and Anal-

ysis of Manufacturing Systems (Wiley, New York

1993)

35.3 K. Hitomi: Manufacturing System Engineering

(Taylor & Francis, London 1996)

35.4 A.I. Dashchenko (Ed.): Manufacturing Technologies

for Machines of the Future: 21st Century Technolo-

gies (Springer, Berlin, Heidelberg 2003)

35.5 A.I. Dashchenko (Ed.): Reconﬁgurable Manufactur-

ing Systems and Transformable Factories (Springer,

Berlin, Heidelberg 2006)

35.6 A. Dolgui, J.M. Proth: Les syst

emes de production

modernes (Hermes-Science, Paris 2006)

35.7 S.Y. Nof, W.E. Wilhelm, H.J. Warnecke: Industrial

Assembly (Chapman Hall, London 1997)

35.8 A. Kusiak: Modelling and Design of Flexible Manu-

facturing Systems (Elsevier, Amsterdam 1986)

35.9 Y. Koren, U. Heisel, F. Javane, T. Moriwaki,

G. Pritchow, H. Van Brussel, A.G. Ulsoy: Recon-

ﬁgurable manufacturing systems, CIRP Ann. 48(2),

527–598 (1999)

35.10 J. Villasenor, W.H. Mangione–Smith: Conﬁgurable

computing, Sci. Am. 276(6), 66–71 (1997)

35.11 G.W. Zhang, S.C. Zhang, Y.S. Xu: Research on

ﬂexible transfer line schematic design using hi-

erarchical process planning, J. Mater. Process.

Technol. 129, 629–633 (2002)

35.12 A. Dolgui, O. Guschinskaya, N. Guschinsky, G. Levin:

Decision making and support tools for design of

machining systems. In: Encyclopedia of Decision

Making and Decision Support Technologies,Vol.

1, ed. by F. Adam, P. Humphreys, (Idea Group,

Hershey 2008) pp. 155–164

35.13 M.E. Salveson: The assembly line balancing prob-

lem, J. Ind. Eng. 6(4), 18–25 (1955)

35.14 A. Dolgui (Ed.): Feature cluster on the balancing

ofassemblyandtransferlines,Eur.J.Op.Res.

168(3), 663–951 (2006)

35.15 I. Baybars: A survey of exact algorithms for the sim-

ple assembly line balancing problem, Manag. Sci.

32(8), 909–932 (1986)

35.16 S. Ghosh, R.J. Gagnon: A comprehensive literature

review and analysis of the design, balancing and

scheduling of assembly line systems, Int. J. Prod.

Res. 27, 637–670 (1989)

35.17 E. Erel, S.C. Sarin: A survey of the assembly line

balancing procedures, Prod. Plan. Control 9(5),

414–434 (1998)

35.18 B. Rekiek, A. Dolgui, A. Delchambre, A. Bratcu: State

of the art of assembly lines design optimisation,

Annu.Rev.Control26(2), 163–174 (2002)

35.19 N. Boysen, M. Fliedner, A. Scholl: A classiﬁcation

of assembly line balancing problems, Eur. J. Oper.

Res. 183(2), 674–693 (2007)

35.20 N. Boysen, M. Fliedner, A. Scholl: Assembly line

balancing: which model to use when?, Int. J. Prod.

Econ. 111,509–528(2008)

35.21 T.K. Bhattachajee, S. Sahu: Complexity of single

model assembly line balancing problems, Eng.

Costs Prod. Econ. 18, 203–214 (1990)

35.22 W.P. Helgeson, D.P. Birnie: Assembly line balancing

using the ranked positional weight technique, J.

Ind. Eng. 12, 394–398 (1961)

35.23 C.L. Moodie, H.H. Young: A heuristic method for

assembly line balancing for assumption of constant

or variable elements time, J. Ind. Eng. 16,23–29

(1965)

35.24 A.L. Arcus: COMSOAL: a computer method of se-

quencing operations for assembly lines, Int. J.

Prod. Res. 4(4), 259–277 (1966)

35.25 F.F. Boctor: A multiple-rule heuristic for assembly

line balancing, J. Oper. Res. Soc. 46, 62–69 (1995)

35.26 B. Rekiek, P. De Lit, A. Delchambre: Designing

mixed-product assembly lines, IEEE Trans. Robot.

Autom. 16(3), 414–434 (1998)

35.27 A. Scholl: Balancing and Sequencing of Assembly

Lines (Physica, Heidelberg 1999)

35.28 M. Amen: Heuristic methods for cost-oriented as-

sembly line balancing, A comparison on solution

quality and computing time, Int. J. Prod. Econ. 69,

255–264 (2001)

35.29 M. Amen: Heuristic methods for cost oriented as-

sembly line balancing, a survey, Int. J. Prod. Econ.

68, 1–14 (2000)

35.30 J. Bukchin, M. Tsur: Design of ﬂexible assembly line

to minimize equipment cost, IIE Trans. 32,585–598

(2000)

35.31 J. Bukchin, A. Rubinovitz: A weighted approach

for assembly line design with station parallel-

Part D 35

Machining Lines Automation References 617

ing and equipment selection, IIE Trans. 35,73–85

(2002)

35.32 C. Andrés, C. Miralles, R. Pastor: Balancing and

scheduling tasks in assembly lines with sequence-

dependent setup times, Eur. J. Oper. Res. 187(3),

1212–1223 (2008)

35.33 J. Szadkowski: Critical path concept for multi-tool

cutting processes optimization. In: Manufacturing

Systems Modeling, Management and Control: Pro-

ceedings of the IFAC Workshop, ed. by P. Kopacek

(Elsevier, Vienna 1997) pp. 393–398

35.34 A. Dolgui, N. Guschinski, G. Levin: On problem

of optimal design of transfer lines with paral-

lel and sequential operations, Proc. 7th IEEE Int.

Conf. Emerg. Technol. Fact. Autom. (ETFA’99), Vol. 1,

ed. by J.M. Fuertes (IEEE, Barcelona 1999) pp. 329–

334

35.35 S. Belmokhtar: Lignes d’usinage avec équipements

standards: modélisation, conﬁguration et opti-

misation. Ph.D. Thesis (Ecole des Mines de Saint

Etienne, Saint Etienne 2006), in French

35.36 S.Belmokhtar,A.Dolgui,N.Guschinsky,G.Levin:

An integer programming model for logical layout

design of modular machining lines, Comput. Ind.

Eng. 51(3), 502–518 (2006)

35.37 A.Dolgui,B.Finel,N.Guschinsky,G.Levin,F.Ver-

nadat: MIP approach to balancing transfer lines

with blocks of parallel operations, IIE Trans. 38,

869–882 (2006)

35.38 A. Dolgui, N. Guschinsky, G. Levin: A Special case of

transfer lines balancing by graph approach, Eur. J.

Oper. Res. 168(3), 732–746 (2006)

35.39 A. Dolgui, N. Guschinsky, G. Levin, J.M. Proth: Opti-

misation of multi-position machines and transfer

lines, Eur. J. Oper. Res. 185(3), 1375–1389 (2008)

35.40 A. Dolgui, I. Ihnatsenka: Branch and bound

algorithm for a transfer line design problem: sta-

tions with sequentially activated multi-spindle

heads, Eur. J. Op. Res. (2007) available online,

doi:10.1016/j.ejor.2008.03.028, (in press)

35.41 A. Dolgui, I. Ihnatsenka: Balancing modular trans-

fer lines with serial-parallel activation of spindle

heads at stations, Discret. Appl. Math. 157(1), 68–

89 (2009)

35.42 W.E. Wilhelm: A column-generation approach for

the assembly system design problem with tool

changes, Int. J. Flex. Manuf. Syst. 11, 177–205 (1999)

35.43 O. Guschinskaya, A. Dolgui, N. Guschinsky, G. Levin:

A heuristic multi-start decomposition approach for

optimal design of serial machining lines, Eur. J.

Oper. Res. 189(3), 902–913 (2008)

35.44 B. Finel: Structuration de lignes d’usinage: méth-

odes exactes et heuristiques. Ph.D. Thesis (Univer-

sité de Metz, Metz 2004), in French

35.45 A.Dolgui,B.Finel,N.Guschinsky,G.Levin,F.Ver-

nadat: An heuristic approach for transfer lines

balancing, J. Intell. Manuf. 16(2), 159–171 (2005)

35.46 B. Finel, A. Dolgui, F. Vernadat: A random search

and backtracking procedure for transfer line bal-

ancing, Int. J. Comput. Integr. Manuf. 21(4),

376–387 (2008)

35.47 M. Essaﬁ, X. Delorme, A. Dolgui: A heuristic method

for balancing machining lines with paralleling of

stations and sequence-dependent setup times,

Proc. Int. Workshop LT’2007 (Sousse 2007) pp. 349–

354

35.48 O. Guschinskaya: Outils d’aide à la décision

pour la conception en avant-projet des systèmes

d’usinage à boîtiers multibroches. Ph.D. Thesis

(Ecole des Mines de Saint Etienne, Saint Etienne

2007), in French

35.49 S. Masood: Line balancing and simulation of an

automated production transfer line, Assem. Au-

tom. 26(1), 69–74 (2006)

Part D 35

“This page left intentionally blank.”

619

Large-Scale C

36. Large-Scale Complex Systems

Florin-Gheorghe Filip, Kauko Leiviskä

Large-scale complex systems (LSS) have tradi-

tionally been characterized by large numbers of

variables, structure of interconnected subsys-

tems, and other features that complicate the

control models such as nonlinearities, time de-

lays, and uncertainties. The decomposition of LSS

into smaller, more manageable subsystems al-

lowed for implementing effective decentralization

and coordination mechanisms. The last decade re-

vealed new characteristic features of LSS such as

the networked structure, enhanced geographical

distribution and increased cooperation of subsys-

tems, evolutionary development, and higher risk

sensitivity. This chapter aims to present a balanced

review of several traditional well-established

methods and new approaches together with typ-

ical applications. First the hierarchical systems

approach is described and the transition from

coordinated control to collaborative schemes is

highlighted. Three subclasses of methods that

arewidelyutilizedinLSS – decentralized control,

simulation-based, and artiﬁcial-intelligence-

based schemes – are then reviewed. Several basic

aspects of decision support systems (DSS) that are

36.1 Background and Scope.......................... 620

36.1.1 Approaches ................................. 621

36.1.2 History........................................ 622

36.2 Methods and Applications ..................... 622

36.2.1 Hierarchical Systems Approach....... 622

36.2.2 Other Methods and Applications .... 626

36.3 Case Studies ......................................... 632

36.3.1 Case Study 1:

Pulp Mill Production Scheduling..... 632

36.3.2 Case Study 2: Decision Support

in Complex Disassembly Lines........ 633

36.3.3 Case Study 3: Time Delay Estimation

in Large-Scale Complex Systems..... 634

36.4 Emerging Trends .................................. 634

References .................................................. 635

meant to enable effective cooperation between

man and machine and among the humans

in charge with LSS management and control

are brieﬂy exposed. The chapter concludes by

presenting several technology trends in LSS.

There is not yet a universally accepted deﬁnition of

the large-scale complex systems (LSS) though the LSS

movement started more than 40 years ago. However, by

convention, one may say that a particular system is

a large and complex one if it possesses one or several

characteristic features. For example, according to To-

movic [36.1], the set of LSS characteristics includes the

structure of interconnected subsystems and the presence

of multiple objectives, which, sometimes, are vague and

even conﬂicting. A similar viewpoint is proposed by

Mahmoud, who describes a LSS as [36.2]:

A system which is composed of a number of smaller

constituents, which serveparticular functions, share

common resources, are governed by interrelated

goals and constraints and, consequently, require

more than one controllers.

Siljak [36.3] states that a LSS is characterized by

its high dimensions (large number of variables), con-

straints in the information infrastructure, and the

presence of uncertainties. At present there are soft-

ware products on the market which can be utilized

Part D 36

620 Part D Automation Design: Theory and Methods for Integration

to solve optimization problems with thousands of

variables. A good example is Solver.com [36.4].

Complications may still be caused by system non-

linearities, time delays, and different time constants,

and, especially over recent years, risk sensitivity

aspects.

36.1 Background and Scope

In real life one can encounter lots of natural, man-

made, and social entities that can be viewed as LSS.

From the early years of the LSS movement,theLSS

class has included several particular subclasses such

as: steelworks, petrochemical plants, power systems,

transportation networks, water systems, and societal

organizations [36.5–7]. Interest in designing effective

control schemes for such systems was primarily moti-

vated by the fact that even small improvements in the

LSS operations could lead to large savings and impor-

tant economic effects.

The structure of interconnected subsystems has ap-

parently been the characteristic feature of LSS to be

found in the vast majority of deﬁnitions. Several sub-

classes of interconnections can be noticed (Fig.36.1).

Resources

SSy

u = h(z)

SSy

s = g(s,z)

u = h(s)

Fig. 36.1a–c Interconnection patterns: (a) resource sharing, (b) di-

rect interconnection,

m = control variable, y = output variable, w = disturbance, u =

interconnection input, z = interconnection output, g(·) = stock dy-

namics function, h(·) =interconnection function

First there are the resource sharing interconnections

described by Findeisen [36.8], which can be identi-

ﬁed at the system level as remarked by Takatsu [36.9].

Also, at the system level, subsystems may be inter-

connected through their common objectives [36.8].

Subsystems may also be interconnected through buffer

units (tanks), which are meant to attenuate the ef-

fects of possible differences in the operation regimes

of plants which feed or drain the stock in the buffer.

This type of ﬂexible interconnection can frequently

be met in large industrial and related systems such

as reﬁneries, steelworks, and water systems [36.10].

The dynamics of the stock value s in the buffer unit

can be modeled by a differential equation. In some

cases buffering units are not allowed because of tech-

nological reasons; for example, electric power cannot

be stocked at all and reheated ingots in steelworks

must go immediately to rolling mills to be processed.

When there are no buffer units, the subsystems are

coupled through direct interconnections,attheprocess

level [36.9].

In the 1990s, integration of systems continued and

new paradigms such as the extended/networked/virtual

enterprise were articulated to reﬂect real-life develop-

ments. In this context, Mårtenson [36.11] remarked

that complex systems became even more complex.

She provided several arguments to support her remark:

ﬁrst, the ever larger number of interacting subsystems

that perform various functions and utilize technolo-

gies belonging to different domains such as mechanics,

electronics, and information and communication tech-

nologies (ICT); second, that experts from different

domains can encounter hard-to-solve communication

problems; and also, that people in charge of control and

maintenance tasks, who have to treat both routine and

emergence situations, possess uneven levels of skills

and training and might even belong todifferent cultures.

Nowadays, Nof et al. show that [36.12]:

There is the need to create the next generation man-

ufacturing systems with higher levels of ﬂexibility,

allowing these systems to respond as a component

of enterprise networks in a timely manner to highly

dynamic supply-and-demand networked markets.

Part D 36.1

Large-Scale Complex Systems 36.1 Background and Scope 621

Table 36.1 Summary of methods described in this chapter

Decomposition-coordination-based Mesarovic, Macko, Takahara [36.7]; Findeisen et al. [36.16];

methods Titli [36.17]; Jamshidi [36.6]; Brdys, Tatjewski [36.18]

Optimization-based methods Dourado [36.19]; Filip, et al. [36.20]; Filip et al. [36.21];

Guran et al. [36.22]; Peterson [36.23]; Tamura [36.24]

Decentralized control Aybar et al. [36.25]; Bakule [36.26]; Borrelli et al. [36.27];

Inalhan et al. [36.28]; Krishnamurthy et al. [36.29];

Langbort et al. [36.30];

Siljak et al. [36.31];

Siljak et al. [36.32]

Simulation-based methods Arisha and Yong [36.33]; Chong et al. [36.34]; Filip et al. [36.20];

Gupta et al. [36.35]; Julia and Valette [36.36]; Lee et al. [36.37];

Leivisk

a et al. [36.38,39]; Liu et al. [36.40];

Ramakrishnan et al. [36.41]; Ramakrishnan and Thakur [36.42];

Taylor [36.43]

Intelligent methods

• Fuzzy logic Ichtev [36.44]; Leivisk

a [36.45]; Leivisk

a and Yliniemi [36.46];

• Neural networks Arisha and Yong [36.33]; Azhar et al. [36.47]; Hussain [36.48];

Liu et al. [36.49]

• Genetic algorithms Dehghani et al. [36.50]; El Mdbouly et al. [36.51]; Liu et al. [36.40]

• Agent-based methods Akkiraju et al. [36.52]; Hadeli et al. [36.53]; Heo and Lee [36.54];

ık and La

zansk

y [36.55]; Park and Lim [36.56]; Parunak [36.57]

They also emphasize that e-Manufacturing is highly

dependent on the efﬁciency of collaborative human–

human and human–machine e-Work. See Chap.88 on

Collaborative e-Work, e-Business, and e-Service.In

general, there is a growing trend to understand the

design, management, and control aspects of complex

supersystems or systems of systems (SoS). Systems

of systems can be met in space exploration, mili-

tary and civil applications such as computer networks,

integrated education systems, and air transportation sys-

tems. There are several deﬁnitions of SoS,mostof

them being articulated in the context of particular ap-

plications; for example, Sage and Cuppan [36.13] state

that a SoS is not a monolithic entity and possesses

the majority of thefollowingcharacteristics: geographic

distribution,operational andmanagement independence

of its subsystems, emergent behavior, and evolutionary

development. All these developments obviously im-

ply ever more complex control and decision problems.

A particular case which has received a lot of attention

over recent years is large-scale critical infrastructures

(communication networks, the Internet, highways, wa-

ter systems, and power systems) that serve not only the

business sector but society in general [36.14, 15]. All

of these recent developments are likely to provide fresh

strong stimuli for new research in the LSS domain.

36.1.1 Approaches

The progresses made in information and commu-

nication technologies have enabled the designer to

overcome severaldifﬁcultieshe mighthave encountered

when approaching a LSS, in particular those caused

by a large number of variables and the low perfor-

mance (with respect to throughput and reliability) of

communication links. However, as Cassandras points

out [36.58]:

The complexity of systems designed nowadays is

mainly deﬁned by the fact that computational power

alone does not sufﬁce to overcome all difﬁculties

encountered in analyzing, planning and decision-

making in the presence of uncertainties.

A plethora of methods have been proposed over

the last four decades for managing and controlling

large-scale complex systems such as: decomposition,

hierarchical control and optimization, decentralized

control, model reduction, robust control, perturbation-

Part D 36.1

622 Part D Automation Design: Theory and Methods for Integration

based techniques, usage of artiﬁcial-intelligence-based

techniques, integrated problems of system optimiza-

tion and parameter estimation [36.59], and so on. Two

common ideas can be found in the vast majority of

approaches proposed so far:

a) Replacing the original problem with a set of sim-

pler ones which can be solved with the available

tools and accepting the satisfactory, near optimal

solutions

b) Exploiting the particular structure of each system to

the extent possible.

Table 36.1 presents a summary of the main methods

to be described in this chapter.

36.1.2 History

Though several ideas and methods for controlling LSSs

were proposed in the 1960s and even earlier, it is ac-

cepted by many authors that the book of Mesarovic

et al. published in 1970 [36.7] triggered the LSS move-

ment. The concepts revealed in that book, even though

they were strongly criticized in 1972 by Va ra iya [36.60]

(an authority among the pioneers of the LSS move-

ment), have inspired many academics and practitioners.

A series of books including those of Wismer [36.61],

Titli [36.17], Ho and Mitter [36.62], Sage [36.63],

Siljak [36.3,64], Singh [36.65], Findeisen et al. [36.16],

Jamshidi [36.6], Lunze [36.66], and Brdys and Tat-

jewski [36.18] followed on and contributed to the

consolidation of the LSS domain of research and paved

the way for practical applications.

In 1976, the ﬁrst International Federation of Au-

tomatic Control (IFAC) conference on Large-Scale

Systems: Theory and Applications washeldinUdine,

Italy. This was followed by a series of symposia which

were organized by the specialized Technical Commit-

tee of IFAC and took place in various cities in Europe

and Asia (Toulouse, Warsaw, Zurich, Berlin, Beijing,

London, Patras, Bucharest, Osaka, and Gdansk). The

scientiﬁc journal Large Scale Systems published by

North Holland played an important role in the devel-

opment of LSS domain, especially in the 1980s.

36.2 Methods and Applications

36.2.1 Hierarchical Systems Approach

The central idea of the hierarchical multilevel systems

(HMS) approach to LSS consists of replacing the orig-

inal system (and the associated control problem) with

a multilevel structure of smaller subsystems (and asso-

ciated less complicated problems). The subproblems at

the bottom of the hierarchy are deﬁned by the interven-

tions made by the higher-level subproblems, which in

turn utilize the feedback information they receive from

the solutions of the lower-level subproblems.

There are three main subclasses of hierarchies

which can be obtained in accordance with the complex-

ity of description, control task,andorganization [36.7].

Levels of Description

The ﬁrst step in analyzing an LSS and designing the

corresponding control scheme consists of model build-

ing. As Steward [36.67] points out, practical experience

witnessed there is a paradoxical law of systems.Ifthe

description of the plant is too complicated, then the de-

signer istempted to consider only a part ofthe systemor

a limited sets of aspects which characterize its behavior.

In this case it is very likely that the very ignored parts

and aspects have a crucial importance. Consequently it

emerges that more aspects should be considered, but

this may lead to a problem which is too complex to be

solved in due time. To solve the conﬂict between the

necessary simplicity (to allow for the usage of exist-

ing methods and tools with a reasonable consumption

of time and other computer resources) and the accept-

able precision (to avoid obtaining wrong or unreliable

results), the LSS can be represented by a family of mod-

els. These models reﬂect the behavior of the LSS as

viewedfrom various perspectives, called [36.7] levels of

description or strata,orlevels of inﬂuence [36.63,68].

The description levels are governed by independent

laws and principles and use different sets of descrip-

tive variables. The lower the level is, the more detailed

the description of a certain entity is. A unit placed on

the n-th level may be viewed as a subsystem at level

n−1. For example, the same manufacturing system can

be described from the top stratum in terms of economic

and ﬁnancial models, and, at the same time, by control

variables (states, controls, and disturbances) as viewed

from the middle stratum, or by physical and chem-

ical variables as viewed from the bottom description

level (Fig.36.2).

Part D 36.2

Large-Scale Complex Systems 36.2 Methods and Applications 623

Levels of Control

In order to act in due time even in emergency situ-

ations, when the available data are uncertain and the

decision consequences are not fully explored and eval-

uated, a hierarchy of specialized control functions can

be an effective solution as shown by Eckman and

Lefkowitz [36.70]. Several examples of sets of levels of

control are:

a) Regulation, optimization, and organization [36.71]

b) Direct control, supervisory control, optimization,

and coordination [36.72]

c) Stabilization, dynamic coordination, static opti-

mization, and dynamic optimization [36.8]

d) Measurement and regulation, production planning

and scheduling, and business planning [36.73].

The levels of control, also called layers by

Mesarovic et al. [36.7], can be the result of a time-scale

decomposition. They can be deﬁned on the basis of

time horizons taken into consideration, or the frequency

of disturbances which may show up in process vari-

ables, operation conditions, parameters, and structure

of the plant as stated by Schoefﬂer [36.68], as shown

in Fig.36.2.

Organization layers

(echelons)

General

management

Production

control

Unit process

control

Economic representations

Regulation

Disturbance frequency in

… process variables

… input variables

… plant structure

Optimization

Organization

Control models

Physical variables

Description levels

(strata)

Control levels

(layers)

Fig. 36.2 A hierarchical system approach applied to an industrial plant (after [36.69])

Levels of Organization

The hierarchies based on the complexity of orga-

nization were proposed in mid 1960s by Brosilow

et al. [36.74]andLasdon and Schoefﬂer [36.75]and

were formalized in detail by Mesarovic et al. [36.7].

The hierarchy with several levels of organization,

also called echelons by Mesarovic et al. [36.7], has

been, for many years, a natural solution for man-

agement of large-scale military, industrial, and social

systems, which are made up of several intercon-

nected subsystems when a centralized scheme can-

not be either technically possible or economically

acceptable.

The central idea of the multiechelon hierarchy is

to place the control/decision units, which might have

different objectives and information bases, on several

levels of a management and control pyramid. While

the multilayer systems implement the vertical divi-

sion of the control effort, the multiechelon systems

include also a horizontal division of work. Thus, on

the n-th organization level the i-th control unit, CU

has limited autonomy. It sends coordination signals

downwards to a well-deﬁned subset of control units

which are placed at the level n−1 and it receives co-

ordination signals from the corresponding unit placed

Part D 36.2

624 Part D Automation Design: Theory and Methods for Integration

Coordinator

Level 2

Level 1

SSy

Fig. 36.3 A simple two-level multilevel control system (CU =

control unit, SSy = controlled subsystem, H = interconnection

function, m = control variable, u = input interconnection variable,

z = output interconnection variable, y = output variable, w = dis-

turbance)

on level n +1. The unit on the top of the pyra-

midiscalledthesupremal coordinator and the units

to be found at the bottom level are called inﬁmal

units.

Manipulation

of Complex Mathematical Problems

To take advantage of possible beneﬁts of hierarchical

multilevel systems a systematic decomposition of the

original large-scale system and associated control prob-

lem is necessary. There are many situations when the

control problem maybe formulatedas (orreduced to)an

optimization problem (P1), which is, deﬁned in general

terms as

(P1) : extr

J(v); v ∈ V , (36.1)

where v is the decision variable (a scalar, or a vector),

V is the admissible variation domain (which can be

deﬁned by differential or difference equations and/or al-

gebraic inequations), and J is the performance measure

(which can be a function or a functional).

The decomposition methods are based on var-

ious combinations of several elementary manipula-

tions [36.76]. There are two main subsets of elementary

manipulations:

a) Transformations, which are meant to substitute the

original large-scale complex problem by a more ma-

nipulable one

b) Decompositions, which are meant to replace alarge-

scale problem by a number of smaller subproblems.

In the sequel several elementary manipulations

will be reviewed following the lines exposed by

Wilson [36.76].

The variable transformation replaces the original

problem (P1) by an equivalent one (P2) through the

utilization of a new variable y = f(x) and a new per-

formance measure Q(y) and admissible domain Y,so

that there is the inverse function v = f

−1

(v). The new

problem is deﬁned as

(P2) : extr

Q(y) ;

(y ∈Y) , (∀y) = f(v)[Q(y) = J(v)] .

(36.2)

The Lagrange transformation can simplify the admissi-

ble domain; for example, let the domain V be deﬁned

by complicated equalities and inequalities

V =



v :(v ∈ V

) , (g

(v) , g

−

(v) ≤0)



, (36.3)

where V

is a certain set, g

and g

−

represent equality

and inequality constraints, respectively.

A Lagrangian can be deﬁned

L(v, π, γ ) = J(v)+π, g

(v)−γ,g

−

(v), (36.4)

where π are the Lagrange multipliers, γ are the Kuhn–

Tucker multipliers, and ·, · is the scalar product.

If L possesses a saddle point, the solution of (P1)

is also the solution of the transformed problem (P2)

deﬁned as

(P2) : max

πγ

min

[L(v, π, γ )];v ∈ V

. (36.5)

The manipulation called evolving the problem is uti-

lized whennot allparameters are known or the priorities

and the constraints are subject to alteration in time. In

such situations, the problem is solved even under uncer-

tainties and then is reformulated to take into account the

accumulation ofnewinformation. The repetitive control

proposed by Findeisen et al. [36.16] is based on such

a transformation.

Having transformed the original problem intoa con-

venient form, a subset of smaller subproblems can be

obtained through decomposition as shown in the sequel.

The partitioning of the large-scale problem can be

applied if several subsets of independent variables can

Part D 36.2