Professional Documents
Culture Documents
www.elsevier.com/locate/dsw
Invited Review
Abstract
Project Management is a professional domain receiving growing attention during the last decades and now it is
considered a key concept by Management Sciences to understand and to develop organizations. Operational Research
has given essential scienti®c contributions to the success of Project Management not just through multiple models to
understand and to represent projects but also by the development of algorithms and aids to support the decisional role
of project manager. In this paper, the major contributions of OR are presented and discussed. Ó 2002 Elsevier Science
B.V. All rights reserved.
0377-2217/02/$ - see front matter Ó 2002 Elsevier Science B.V. All rights reserved.
PII: S 0 3 7 7 - 2 2 1 7 ( 0 1 ) 0 0 0 9 7 - 2
2 L.V. Tavares / European Journal of Operational Research 136 (2002) 1±18
under the most convenient conditions. Therefore, The objective of this review is presenting such
this process is carried out in terms of an overall perspective and therefore a systemic
(D) An organizational framework. approach will be adopted based on the three
(E) A set of resources. major types of questions required by any systemic
(F) A methodological support. study:
Any taxonomy about project management in- (I) How can a project be modelled?
cludes a very wide variety of cases and problems as (II) How can a project be evaluated?
they can be described by the six perspectives al- (III) How can a project be scheduled and moni-
ready presented ((A)±(F)). tored?
A wide range of case-studies has been dis- Major ®ndings on I, II and III are presented in the
cussed: following Sections 2, 3, 4, respectively. A ®nal
· Development of a new missile Polaris by the US section (Section 5) presents more advanced and
Navy (in US Navy, 1958). recent results.
· Maintenance shutdowns of a DUPONT factory
(in Kelley, 1961).
· Development of a new aircraft: CONCORDE 2. Project modelling
(1964±1972) (in Hayward, 1983).
· Development of a European spacecraft (GI- 2.1. General model
OTTO) to observe Halley's comet, March 86
(in Jenkins and Link, 1984). The general model (Battersby, 1967) developed
· Development of a new railways network in to represent projects is quite a basic concept in
Oporto, Portugal (in Tavares, 1984). OR: a directed and acyclic network. Actually, each
· Construction of a new re®nery in New Zealand, project can be modelled by:
including the cooperation of three engineering (a) A discrete and ®nite set of entities, A, usu-
teams in Yokohama (Japan), the Hague (The ally, called jobs or activities with A fAi : i
Netherlands) and Whangarci (New Zealand), 1; . . . ; N g.
(in Bishop and Gembey, 1985). (b) A set of precedence conditions, J, with,
· Preparation of the school education systems at J fJi : i 1; . . . ; N g where Ji is the set of activi-
the commencement of each academic year (in ties immediately preceding i. Ji can be de®ned by
Palmer, 1985a). Ji fk:
k 2 Ji0 \
k 2 Jm0 ; for any m 2 Ji0
· Planning of rural development centers in where Ji0 or Jm0 is the whole set of activities which
Pakistan (in Palmer, 1985b). have to be completed before starting i or before
· Development of a new Concert Hall for the starting m. Similarly, the set of activities which are
European Culture Capital, Porto 2001 (in the immediate successors of i, Ki , can be de®ned by
Tavares, 2000). Ki fk: i 2 Jk g.
Project Management, PM, has been considered (c) A discrete and ®nite set of attributes
as a speci®c domain of professional activity since fB1
i; . . . ;
Bp
ig with p P 1 de®ned for each ac-
the second world war because of the challenges tivity and describing its properties relevant for
coming from the economic development, the project management such as duration, cost, con-
complexity of new technologies and the signi®cant sumption required of each resource, etc.
methodologic advances to support PM which have (d) A discrete and ®nite set of criteria
been mainly oered by OR. fV1 ; . . . ; Vq g which should express the values and
Such OR contributions include an extremely the preferences of the project manager to compare
wide variety of methods, techniques, algorithms, alternative decisions concerning the management
programs, etc. explaining why most reviews focus of the project. The most common criteria are the
more on one or another sub-domain rather than total duration, the total cost, a cost±bene®t func-
on the whole picture of OR know-how serving tion, a measure of the project risk, the net present
PM. value (NPV).
L.V. Tavares / European Journal of Operational Research 136 (2002) 1±18 3
The available models to describe each of these the arc between A and B will have a duration equal
components have been improved by a large num- to 5.
ber of OR researchers. The improvement of network models for pro-
ject management has been pursued along seven
dierent lines:
2.2. Progress in network modelling (A) Construction of ``generalized networks''
according to Kaumann and Desbazeille (1964)
The representation of a project by a network in where some activities just occur with speci®c
terms of (a) and (b) can be done adopting the probabilities or in terms of the outcomes of pre-
usual assumption
a ``activity-on-arc'' (AoA) vious activities (probabilistic networks).
where each arc describes an activity and where (B) Construction of ``logical networks'' where
each node represents the completion of the activ- the occurrence of each activity is conditioned by
ities converging on it, but it also can be produced logical relationships between precedent activities
using the alternative hypotheses
b ``activity-on- (Battersby, 1967).
node'' (AoN) where each node represents an (C) Modelling of ``overlapping activities'', in
activity and each arc between two nodes describes terms of the time domain (more easily achieved
a precedence relationship between the activi- using ``activity-on-node'' networks, as it was
ties associated to such nodes (see an example in shown by the Method of Potentials and by the
Fig. 1). ``Precedence Diagramming Method'', Leashman
The adoption of a is more common in the OR and Kun, 1993) or in terms of the consumed re-
literature as it was used by the popular methods of sources expressed by progress lag constraints for
PERT/CPM, but an alternative method proposed activities carried out at each time (see Leashman
by Roy (1964) (The Method of Potentials, Roy, et al., 1990).
1964) has adopted b. In this case, each node rep- (D) Introduction of ``hammock'' activities.
resents an activity, each arc describes a precedence These activities are associated to the time span
relation and the duration of each arc is equal to occurring between events concerning other activi-
the minimal time span between the pair of starting ties. For instance, the use of an equipment between
times of two adjacent activities. In any case, the the start (or end) of an activity and the start (or
computation of the critical path is straightforward end) of another activity. The duration of these
but the second representation is more ¯exible to activities is equal to the dierence of times between
describe overlapping activities. For instance, if A the two speci®ed events. The construction of net-
precedes B and A has duration equal to 10 units works with these activities was studied by a few
but B can start when half of A is carried out, then authors (see Harhalakis, 1990).
Fig. 1. The network model of a project using the activity-on-arc (I) and the activity-on-node (II) notation.
4 L.V. Tavares / European Journal of Operational Research 136 (2002) 1±18
applied by Tavares (1984) to a real case-study. crease the importance of the problem of project
Another model (MACMODEL) is now available scheduling, which will be studied in following
as a decision-aid to support the process of multi- sections.
criteria evaluation of a project (Tavares, 1998).
This model helps the decision-maker:
(a) To construct the most appropriate value- 4. Project scheduling and monitoring
tree (see, for instance, Fig. 7). The adopted tree
can include (cost, time, etc.) and also the proba- 4.1. Project scheduling without restrictions on
bility of having outcomes much worse than those resources
expected (risk).
(b) To assess the relative importance (weight) of Project scheduling has been a major objective
each criteria. of most models and methods proposed to aid
(c) To carry out a sensitivity analysis on the planning and management of projects.
weights or on the data of alternative projects Initially, the study of project scheduling has
(TRIDENT analysis). As an example, the sub- been done considering just the duration and pre-
domain of each ranking of alternatives is presented cedence conditions and ignoring the resource re-
on the weights space
k1 ; k2 ; k3 with k1 k2 quirements. Two basic methods proposed to
k3 1) in Fig. 8. schedule a project assuming deterministic dura-
Also, the discussion of the bidding and of the tions are the well-known CPM ± Critical Path
negotiation problems is receiving innovative con- Method and the very much ignored MP ± Method
tributions by Elmaghraby (1990) and Daynand of Potential. They both determine the critical path
and Padman (1994). which gives the minimal total duration of the
Unfortunately, most of the work on project project, T, and the slacks for each node and ac-
evaluation has not considered the relationship tivity.
between the evaluation and the adopted schedule. Obviously, these methods provide a very useful
However, it is quite clear that indicators such as aid for the project manager to schedule each ac-
NPV or the risk of delay strongly depend on the tivity of the project, assuming a total duration, T 0 ,
schedule as early (late) starting times tend to be to be respected (equal to, or greater than, the
responsible for lower (higher) NPV and risk of minimal total duration) because they compute the
delay. Therefore, this is another reason to in- minimal (or maximal) starting and ®nishing times
for each activity. If T T 0 , the scheduling of the (D) Analytical bounds for E
T (Elmaghraby,
critical activities oers no choice to the manager 1967; Robillard and Trahan, 1976).
but this is not the case if T 0 > T . However, in the world of applications, PERT re-
However, most durations have a random na- sults are still the most popular ones.
ture and therefore PERT was proposed to deter-
mine the distribution of the total duration, T. This
method is based on the substitution of the network 4.2. Project scheduling with restrictions on resources
by the CPAD ± critical path assuming that each
activity has a ®xed duration equal to its mean 4.2.1. Formulation
(``critical path using average durations''). In general terms, this problem can be formu-
The mean and the variance of the CPAD is lated by a mathematical model where:
given by the sum of the means and of the variances (a) The decision variables are the scheduled
of its activities, respectively, and therefore these starting times of the activities.
results are considered the mean and the variance of (b) The constraints include the precedence
the total duration of the network. conditions and the maximal (and or minimal)
Unfortunately, this is an optimistic assump- bounds concerning the available resources.
tion as the real mean, E
T , is greater than or (c) The objective function describes the main
equal to such estimate. Thus, many authors have criteria such as minimization of the total duration,
studied: levelling of resources or minimization of the risk of
(A) Analytical approximations of the cumula- delay as well as maximization of the net present
tive distribution function of T, F
T (Alesarea value or of other cost±bene®t indicators. Quite
and Drezner, 1986). often, the objective function is a weighted average
(B) Upper or lower bounds of F
T (Dodin, of some of these criteria. These criteria can be
1985). expressed by deterministic measures or by sto-
(C) Monte-Carlo simulations to estimate F
t chastic functions de®ned in terms of expectations
(Ragsdale, 1989). or of extreme quantiles (risk analysis).
L.V. Tavares / European Journal of Operational Research 136 (2002) 1±18 9
A wide range of alternative models can be The former problem (single-mode) is easier but
adopted for each of these variables or functions the latter is more realistic and it includes the dis-
and therefore several taxonomies can be pro- crete assumption (multimode-model) and the con-
posed. tinuous case (continuous model).
The author believes that an appropriate classi- Obviously, the second assumption implies that
®cation can be based just on three dimensions as is the duration of each activity depends on the in-
shown in Fig. 9 (in Tavares, 1998). tensity of the implementation allocated to each
A few key alternative assumptions should be time unit.
underlined: Another option about the implementation of
(A) Implementation of the activities. The im- activities concerns the preemption assumption al-
plementation of the activities can be assumed ready mentioned: (non) preemptive activities can
having to follow just one pattern or, alternatively, (not) be interrupted and continued later on.
the intensity of its implementation on each time A more realistic formulation assumes the pre-
unit can be considered as a discrete or continuous emptive hypothesis but it allocates an additional
variable under speci®c bounds. ®xed charge every time an interruption occurs.
(B) Uncertainty and randomness. The formu- (a) Exact methods. These results explore the full
lated variables (duration or resources require- space of the scheduling alternatives. Usually they
ments) can be assumed to be deterministic are based on branch and bound procedures to
magnitudes or, alternatively, random or stochastic avoid full enumeration. Recent proposals were
variables. presented by Brucker et al. (1998) and Sprecher
(C) Static or dynamic decision-making. The and Drexl (1998).
process of decision-making concerning the sched- (b) Heuristics. These algorithms do not guar-
uling of activities and the allocation of resources to antee the obtention of the optimum but they tend
the implementation of activities can be considered to be faster (Boctor, 1990). Recently, several pro-
static or dynamic: static if the decision should be cedures based on tabu search, simulated annealing
made before starting the project without the ac- or genetic methods have been applied to the
ceptance of any later correction or change; dy- scheduling problems.
namic if the decision can be changed along the Recent contributions can be quoted (Ahn and
process of implementing the project. Erengue, 1998; Bianco et al., 1998).
The real problem faced by most project man- These methods have to be tested using the ex-
agers corresponds to the dynamic formulation and perimental approach based on generated sets of
the adaptive control of the scheduling decisions is networks.
a major skill for successful project management. Unfortunately, the developed methods have
important shortcomings:
4.2.2. The deterministic static single-mode problem (a) Usually, they have not been tested for pro-
A very large number of methods have been ject networks with a medium or large size
proposed to solve the problem of project sched-
N > 50.
uling under resources constraints but unfortu- (b) A numeric solution is produced for each set
nately most of them address the deterministic, of data providing little understanding about the
static, single-mode non-preemptive formulations. structure of the problem. No sensitivity analysis
Most of these contributions are based on a is available.
model de®ned in terms of Xi
t 0 (or 1) if the (c) Their performance tends to be unstable and
activity i is (or is not) carried out at time unit t. sensitive to the features of the project network.
Less explicit formulations are based on Yi
t which However, this relationship has not been studied.
can be equal to 1 (or to 0) if the ®nishing or
starting time of i is equal (or not equal) to t. 4.2.3. Deterministic continuous mode problem
The objective of minimizing the total duration In this case, the decisions concerning the im-
can be formulated in terms of the maximal plementation of each activity include its starting
t Xi
t, or of t Yi
t for the last activity to be time and also its intensity in terms of time.
completed. Therefore, the total resource consumed by each
The objective of levelling the consumption of activity and its duration are continuous variables.
resources can be expressed by the minimization of All the magnitudes are assumed to be determi-
the sum of squares of the resources allocated at nistically known.
each time unit, t, over the duration of the project. It seems that the ®rst contribution to study this
The precedence conditions can be formulated in problem was given by Fulkerson (1962), using a
terms of linear functions of t Xi
t or t Yi
t. network ¯ow model.
This problem has been studied by binary opti- Weglarz (1981) has studied this problem using
mization methods since late 1960s (Pritsker et al., Optimal Control Theory and assuming that the
1969; Davies, 1972; Patterson, 1984; Demeule- processing speed of each activity at time t is a
meester and Herroelen, 1997) and every year sev- continuous, non-decreasing function of the
eral algorithms are proposed claiming better amount resource allocated to the activity at that
results than the competitors. They belong to two instant of time. This means that also time is here
major groups: considered as a continuous variable. Unfortu-
L.V. Tavares / European Journal of Operational Research 136 (2002) 1±18 11
nately, it seems that this approach is not applicable 4.3. Project monitoring
to networks with a reasonable size (>10).
A powerful approach to solve this problem is The monitoring and control of projects has
based on Dynamic Programming (Elmaghraby, been heavily supported by three types of instru-
1993, 1995). The author has presented a general ments:
model based on the decomposition of the project (a) Impressive development of information
into a sequence of stages (Tavares, 1987, 1989) systems under several labels such as Management
and the optimal solution can be easily computed Information Systems or Executive Information
for each practical problem as it is shown for a Systems to produce updated pictures of how the
real case-study. Also, the theoretical optimal re- project is progressing in terms of completion of
source pro®le can be deduced for some speci®c activities, consumption of resources, delays, qual-
cases using the Calculus of Variations (see ity control, etc. (Drigani, 1989).
Fig. 10). (b) Multivariate data analysis of completed
This formulation also can be extended to the activities or of previous projects to learn how to
dynamic case because at each stage of DP one has improve and to correct initial estimates adopted
the time of occurrence of that stage as a state for the project evaluation and scheduling (Kelley,
variable and therefore the optimal decision is made 1982).
in terms of such state which means that the model (c) Decision Support Systems to assess the
can consider eventual delays or advances. progress of the project and to update the adopted
Recently, new models were proposed to study models for project management (Mitra, 1986).
this problem (the continuous allocation problem) For instance, PERT or scheduling models can
and are presented in the following Section 5.1. be updated in terms of the information given by
The stochastic assumption introduces addi- the systems mentioned in (a) and using the
tional diculties and then the experimental ap- knowledge produced by models included in (b).
proach has to be adopted.
Actually, the study of advanced determinis-
tic or stochastic scheduling methods has to be 5. New results
based on generated sets of network which ex-
plains why generation methods have become so 5.1. The continuous allocation model
important.
New developments on generation of net- This model solves the deterministic continuous
works and on scheduling models are presented in mode problem (Tavares, 1998) and follows the line
Section 5. of development started by Ferreira and Antunes
(1989) and Tavares (1990).
PH
This model (Fig. 11) is based on: objective function t0 Yt has to be maximized
(a) Variables and one obtains the total duration T given by
· Xit ! intensity of implementation of activity i at
time unit t with t 0; 1; . . . ; H where H is the X
H
T
H 1 Yt :
maximal time limit to complete the project with
t0
0 6 Xit 6 1.
· mit ! consumption or use of resource by the ac- Therefore, an objective like the maximization of
tivity i at time unit t if Xit 1. Therefore, the ac- the net present value (see Elmaghraby and Her-
tual consumption of resource by activity i at roelen, 1990) can be expressed by: Max Bf T PC
time unit t is given by (mit Xit . where B is the bene®t received immediately after
· Mi ! total resource required to complete i. completing the project.
· A Xit ! cumulative resource used until time unit This model includes a set of complementary
(t 1) P and including (t 1) which is de®ned by conditions
Xjt DXit 0 and Yt Dt 0 which
AXti ts01 Xis mis . can be relaxed by introducing the corresponding
· D Xit ! remaining amount of resource not yet Lagrangan terms in the objective function to be
used until the beginning of the time unit t: minimized
AXit Mi AXit .
" #
· Finally, XX X
X a Xjt DXit b Yt Dt ;
DXt DXit :
i;j t t
i
Fig. 12. An example of the latest starting time schedule for a network with 150 activities.
However, this assumption implies that the user The published models do not allow the gener-
will give the number of nodes which is usually ation of networks preserving their morphologic
unknown and even undetermined as it has been features (Demeulemeester and Herroelen, 1993;
proved that dierent graphs can represent the Agrawal et al., 1996) but, recently, a model was
same network if AoA is adopted. proposed using AoN and generating networks in
14 L.V. Tavares / European Journal of Operational Research 136 (2002) 1±18
Fig. 13. Example of the graphical representation of a network with 75 activities (AoN).
terms of their morphology described by Tavares
j ! i is de®ned by [m
i m
j] where m
et al. (1997): is the progressive level of.
· Size, I1 N . · Rate of decrease of the number of links in terms
· Length, of their length, I5 p. Assuming an exponential
decrease, one has S
v 1 S
v. p where S
M 1
I2 for N > 1 with 0 6 I2 6 1: is the number of non-redundant links with
N 1 length
and v 1; 2; . . . ; V with V being the
· Width, maximal length.
W
m 1 · Maximal length of the precedence links,
I3
m with m 1; . . . ; M; I6
V 1=
M 1 with 0 6 I6 6 1.
N M
A test set of 216 networks was generated with
where W
m is the number of activities with the N 10; . . . ; 1000. An example of a network with
progressive level m with m 1; . . . ; M. Also, 75 activities is presented in Fig. 13. AoN is adop-
this indicator can be de®ned in terms of the ted and any activity i with K
i ; is represented
maximal width, by a square instead of a circle to avoid the need to
MW 1 represent the links between such nodes and the end
MW max W
m; I30 : of the project network.
m N M
Obviously, one has 0 6 I3 or I30 6 1
· Number of non-redundant precedence links 5.3. The stochastic modelling of project risk
with length equal to one, n
1, I4 n
1
N =D N where D is the maximal n
1 and The most realistic and useful models of project
again 0 6 I4 6 1.The length of a precedence link management adopt the stochastic assumption to
L.V. Tavares / European Journal of Operational Research 136 (2002) 1±18 15
where a; b are appropriate constants and ei is a This research assumes that the project manager
random normal deviate with zero mean. is attempting to implement a pre-de®ned schedule,
fts
jg, and hence, in each generated instance, k, Therefore, an optimal a has to be determined
the starting time of each activity, j, tks
j is deter- with the objective of minimizing the project risk.
mined by An example of the present cost and of the total
duration computed for a large number of gener-
s
ated instances is presented in Figs. 15 and 16.
tks
j max t
j; max tks
i Dk
i
i2I
j These results illustrate the new type of deci-
sion-aids which can be oered by OR stochastic
for I
j 6 ; and where Dk
i is the duration of i for models to help project managers to make better
the instance k. decisions.
If I
j ;, one has tks
j ts
j.
A theorem on the ¯oat management is proved in
(Tavares et al., 1998) showing that the manager References
can adopt for each instance
Acko, R.L., 1970. A Concept of Corporate Planning. Wiley,
ts
i tE
i a tL
i tE
i New York.
Agrawal, M.K., Elmaghraby, S.E., Herroelen, W.S., 1996.
with 0 6 a 6 1 and where tE
i and tL
i are the DAGEN a generator of testsets for project activity nets.
European Journal of Operational Research 90, 376±382.
earliest and latest starting times of i.
Ahn, T., Erengue, S.S., 1998. The resource constrained project
A large a tends to increase the duration and to scheduling problem with multiple crashable nodes: A
reduce the present cost due to the discount factor heuristic procedure. European Journal of Operational
and a lower a has the opposite eect. Research 107 (2), 250±259.
L.V. Tavares / European Journal of Operational Research 136 (2002) 1±18 17
Alesarea, K.P., Drezner, E., 1986. A multivariate approach to Elmaghraby, S.E., 1990. Project bidding under deterministic
estimating the completion time for PERT networks. Journal and probabilistic activity durations. European Journal of
of the Operational Research Society 37 (8), 811±815. Operational Research 49 (1), 14±35.
Battersby, A., 1967. Network Analysis for Planning and Elmaghraby, S.E., 1993. Research allocation via dynamic
Scheduling. MacMillan, New York. programming in activity networks. European Journal of
Bein, W.W., Kamburovski, J.E., Stallman, M.M., 1992. Opti- Operational Research 64 (2), 199±215.
mal reduction of two terminal directed acyclic graph. SIAM Elmaghraby, S.E., 1995. Activity nets: A guided tour through
Journal Computing 21, 1112±1129. some recent developments. European Journal of Opera-
Bianco, L., Dell'Olmo, P., Speranza, M.G., 1998. Heuristic for tional Research 82, 383±408.
multimode scheduling problems with dedicated resources. Elmaghraby, S.E., Herroelen, W.S., 1990. The scheduling of
European Journal of Operational Research 107 (2), activities to maximize the net present value of projects.
260±271. European Journal of Operational Research 49 (1), 35±50.
Bishop, Gembey, 1985. Management information ¯ow on a Ferreira, J.A., Antunes, 1989. Modelos para a gest~ao de
mega project. International Journal of Project Management empreendimentos de Engenharia Civil. Instituto Superior
3 (1), 39±44. Tecnico (in Portuguese).
Blazewicz, J., Ecker, K.M., Schmidt, G., Weglarz, J., 1985. Fulkerson, D.R., 1962. Expected critical path links in PERT
Scheduling in Computer and Manufacturing Systems. networks. Operations Research 10, 808±817.
Springer, Berlin. Harhalakis, G., 1990. Special features of precedence network
Boctor, F.F., 1990. Some ecient multi-heuristic procedures for charts. European Journal of Operational Research 49 (1),
resource-constrained project scheduling. European Journal 50±60.
of Operational Research 49, 3±14. Hayward, K., 1983. Government and British Civil Aerospace.
Boctor, F.F., 1994. A new and ecient heuristic for scheduling Manchester University Press, Manchester.
projects with resource restrictions and multiple executions Jenkins, R.H., Link, D.C.R., 1984. GIOTTO: Europe's explor-
modes. In: Proceedings of the Fourth International Work- atory mission to the Comet Haley. Journal of the British
shop on PMS, Leuven. Interplanetary Society 37, 17±27.
Brucker, P., Knust, S., Schoo, A., Thiele, 1998. A branch and Kaumann, A., Desbazeille, G., 1964. La methode du chemin
bound algorithm for the resource-constrained project critique. Dumond (in French).
scheduling problem. European Journal of Operational Kelley, A.J. (Ed.), 1982. New Dimension of Project Manage-
Research 107 (2), 272±288. ment. Lexington Books, Lexington, MA.
Clark, C.E., 1962. The PERT model for the distribution of an Kelley, J.E., 1961. Critical path planning and scheduling:
activity time. Operations Research 10, 405±406. Mathematical bases. Operations Research 9, 246±320.
Davies, E.W., 1972. Project scheduling under resources con- Leashman, R.C., Kun, S., 1993. A revised critical path method
straints: Historical review and categorization of procedures. for networks including both overlap relationships and
AIIE Transactions 5, 297±313. variable duration activities. European Journal of Opera-
Daynand, N., Padman, R., 1994. The payment scheduling tional Research 64 (2), 229±248.
problem: A client's model. In: Proceedings of the Fourth Leashman, R.C., Dinceler, Kim, S., 1990. Research ± con-
International Workshop on PMS, Leuven. strained scheduling of projects of variable ± intensity
Dean, B., Mertel Jr., S., Roepke, L., 1969. Research project activities. IIE Transactions 22 (1), 31.40.
cost distribution and budget forecasting. IEEE Transactions Mitra, S.S., 1986. Decision Support Systems ± Tools and
on Engineering Management 16 (4), 176±191. Techniques. Wiley, New York.
Demeulemeester, E., Herroelen, W.S., 1993. A random Muller, J.H., Spinrad, J., 1989. Incremental modular decom-
activity network generator. Operations Research 41 (5), position. Journal of ACM 36, 1±19.
972±980. Palmer, C., 1985a. Planning of rural development centres in
Demeulemeester, E., Herroelen, W.S., 1997. A branch-and- Pakistan. International Journal of Project Management 3
bound procedure for the generalized resource constrained (2), 73±81.
project scheduling problem. Operations Research 45 (2), Palmer, C., 1985b. Preparation of the school education system
201±212. at the commencement of each academic year. International
Dodin, B.M., 1985. Bounding the project completion time in Journal of Project Management 3 (4), 208±217.
PERT networks. Operations Research 33 (4), 862±881. Patterson, J.J., 1984. A comparison of exact approaches for
Drigani, F., 1989. Computerized Project Control. Dekker, New solving the multiple constrained research project scheduling
York. problem. Management Science 30, 854±867.
Drucker, P., 1974. Management: Tasks, Responsibilities, Prac- Pritsker, A.A.B., Watters, L.J., Wolfe, P.M., 1969. Project
tices. Harper & Row, New York. scheduling with limited resources: A zero±one programming
Elmaghraby, S.E., 1967. On the expected duration of PERT approach. Management Science 16, 93±108.
type networks. Management Science 13 (5), 299±306. Rafael, F., 1990. Sistema de apoio a gest~ao e programacß~ao
Elmaghraby, S.E., 1977. Activity Networks: Project Planning ®nanceira de projectos. Investigacß~ao Operacional 10 (1), 97±
and Control by Network Models. Wiley, New York. 108 (in Portuguese).
18 L.V. Tavares / European Journal of Operational Research 136 (2002) 1±18
Ragsdale, C., 1989. The current state of network simulation in (Eds.), Advances in Project Scheduling. Elsevier, Amster-
project management theory and practice. OMEGA 17 (1), dam.
21±25. Tavares, L.V., 1990. A multi-stage non-deterministic model for
Robillard, P., Trahan, M., 1976. Expected completions time in project scheduling under resources constraints. European
PERT networks. Operations Research 24 (1), 177±182. Journal of Operational Research 49 (1), 92±101.
Roy, B., 1964. Les problemes d' ordonnancement. Dunod, Tavares, L.V., 1994. A stochastic model to control project
Paris. duration and expenditure. European Journal of Operational
Simon, M.A., 1977. The New Science of the Management Research 78 (2), 262±266.
Decision. Prentice-Hall, New York. Tavares, L.V., 1998. Advanced Models for Project Manage-
Speranza, M.G., Vercellis, C., 1993. Hierarchical models for ment. Kluwer Academic Publishers, Dordrecht.
multi-project planning and scheduling. European Journal of Tavares, L.V., 2000. A Multicriteria project assessment for the
Operational Research 64 (2), 312±325. new Concert Hall of Oporto 2001 ± Casa da M usica.
Sprecher, A., Drexl, A., 1998. Multi-mode resource-constrained CESUR, Lisbon.
project scheduling by a simple, general and powerful Tavares, L.V., Weglarz, J., 1990. Project management and
sequencing algorithms. European Journal of Operational scheduling: A permanent challenge for OR. European
Research 107 (2), 431±450. Journal of Operational Research 49, 1±2.
Tavares, L.V., 1984. The TRIDENT approach to rank alter- Tavares, L.V., Ferreira, J.A., Coelho, J.S., 1997. The risk of delay
native tenders for large engineering project. Foundations of of a project in terms of a morphology of its network. EURO
Control Engineering 9 (4), 181±191. XV±INFORMS XXXIV Conference, Barcelona, Spain.
Tavares, L.V., 1986. Stochastic planning and control of Tavares, L.V., Ferreira, J.A., Coelho, J.S., 1998. On the
program budgeting ± the model MACAO. In: Coelho, optimal management of project risk. European Journal of
J.D., Tavares, L.V. (Eds.), OR Models on Microcomputers. Operational Research 107, 451±469.
North-Holland, Amsterdam. US Navy, 1958. PERT Summary Reports, Phase 1 and Phase 2,
Tavares, L.V., 1987. Optimal resource pro®les for program Special Projects Oce, Bureau of Naval Weapons, Depart-
scheduling. European Journal of Operational Research 29, ment of Navy, Washington, USA.
83±90. Weglarz, J., 1981. Project scheduling with continuously divis-
Tavares, L.V., 1989. A multi-stage model for project scheduling ible doubly constrained resources. Management Science 27,
under resource constraints. In: Slowinski, R., Weglarz, J. 1140±1153.