You are on page 1of 5

TECHNICAL AR TICLE

A Multiobjective Linear Program be modeled (scheduled) using linear pro-


gramming (LP), which allows a decision-
maker to maintain work continuity,
for Scheduling Repetitive Projects increasing productivity and decreasing the
duration, thus controlling costs.
A contractor’s usual goal is to com-
plete a project in the shortest possible time
Dr. Kamal Al-Subhi Al-Harbi, Dr. Shokri Z. Selim, and Mazen Al-Sinan for the lowest possible cost. As illustrated
in figure 1, an individual activity’s cost
tends to increase with a decreasing activity
duration, up to a certain duration, which
ertain construction projects, repetitive projects. Unlike other types of is called the crashed duration. The

C such as pipelines, highways,


high-rise buildings, and bridges,
include repetitive construction
activities. We thus refer to these projects as
construction, where resources and tasks
change from stage to stage, in repetitive
projects, each stage consists of more or less
the same activities. Repetitive projects can
crashed duration is the fully-expedited or
minimal activity duration time that is tech-
nically possible. The crashed cost is
assumed to be the minimum cost required
to achieve the crashed duration.
Cost A linear programming model, named
the repetitive project model (RPM), uses a
single objective function to minimize the
direct cost for each feasible project dura-
CCi tion, while satisfying the constraints per-
ai Cost Slope taining to the production rate and conti-
nuity of work for each activity’s crew, time,
and stage buffer [9].
The activity’s flowline curve (see fig-
ure 2) represents the movement of an
activity’s crew from one stage to another,
CNi throughout the project’s stages. The slope
of each line represents the production rate
for each activity. Each activity, I, has a
Time
time buffer at each stage between the start
di Di
time of the activity and the start time of
Crashed Normal each preceding activity, H. This buffer
Duration Duration time equals the duration of the preceding
activity, plus any necessary lead time. The
Figure 1—Time Cost Curve horizontal distance between the two lines
at any stage represents the time buffer
Stage
between the two activities during the
stage.
Any two concurrent activities may
have a stage buffer of a specific number of
stages to meet practical and/or technologi-
cal purposes. This stage buffer has to be
H
identified by the planner for these activi-
Slope
(Production Rate) ties. The RPM finds, for each feasible proj-
I
ect duration, the start time of each activity
at the first and last stage, the duration of
Minimum Stage Buffer = M each activity (which is to be set equal at all
Stage X stages to maintain a constant production
Minimum Time Buffer rate for the crew working on the activity),
and finally, the project minimum direct
Time cost that can achieve this project-feasible
Start Start duration. The vertical distance between
Time Time of the lines in figure 2 represents the stage
of H I buffer between the two activities at a par-
ticular time.
Figure 2—Typical Flowline Curves of Repetitive Activities

Cost Engineering Vol. 38/No. 12 DECEMBER 1996 41


The RPM requires that an activity has cretizing the dynamics of a system of a MATHEMATICAL FORMULATION
the same duration during all stages. hypothetical cut-and-fill job on a section OF THE MOLP MODEL
However, in reality, a single activity may of highway.
have different durations during different Russell and Caselton presented a two- We have defined the variables and the
stages of the project, assuming constant state variable N-stage dynamic program- constants used in the mathematical for-
production rates. Selinger explicitly illus- ming formulation of the linear scheduling mulation as the following:
trated this idea in his work [11]. If the program [10]. Their formulation is based
assumption of constant activity times at on the conventional dynamic program- • N = number of repetitive stages in the
different stages is relaxed, then we should ming framework, and has the goal of short- project;
expect that the resulting linear program ening a project’s duration. Their method • P = number of activities in a stage;
will be longer than that of the repetitive accounts for several of the realities of • M = number of buffer stages;
project model. However, other features of repetitive construction, including general- • ai,s = cost slope of activity i in the sth
the RPM will remain. ized precedence relationships and the stage;
The purpose of this article is to extend ability to treat a variety of work-continuity • CNi,s = normal cost of activity i in the
the RPM to general repetitive projects, constraints. Their contributions were di- sth stage;
while considering several objective func- rected toward the goal of having linear • TSi,s = start time of activity i in stage
tions. We selected two common objec- scheduling become a practical, computer- s;
tives: project cost and project duration. based tool for repetitive construction proj- • Di,s = normal duration of activity i in
ects. The model tries to lessen the overall stage s;
project duration. • di,s = crashed duration of activity i in
LITERATURE REVIEW stage s;
• yi,s = number of days that activity i in
Approaches for scheduling repetitive A MULTIOBJECTIVE LINEAR the s stage can be shortened; and
construction projects can be classified into PROGRAM MODEL (MOLP) • TLh,i = lead time between activity i
three categories: and preceding activity h.
We are presenting a multiobjective
• traditional network techniques such as linear program (MOLP) for activity net- The objective functions can be illus-
the critical path method (CPM) [8]; works that satisfies the following require- trated by the following equations.
• graphical approaches, such as the line ments: N P
of balance [1, 2], the vertical produc- Minimize Z1 = ∑ ∑ai,s yi,s
tion method [7], and the linear sched- • maintains a constant production rate s=1 i=1
uling method [3, 6]; and for each activity’s crew; N P
• optimization approaches like the • maintains continuity of work for each +∑ ∑CNi,s.
dynamic programming model [10] and activity’s crew, considering that an s=1 i=1
the linear programming model [9]. activity may have different durations (equation 1)
in different stages of the project;
The use of the CPM for scheduling • allows for a time buffer between activ- Minimize Z2 = TSP,N + DP,N - yP,N.
repetitive projects has two disadvantages. ities in the same stage; and (equation 2)
First, it requires a large number of activi- • allows for a stage buffer between con-
ties to represent the project, making it dif- current activities. The first objective function gives the
ficult to visualize the project and to total cost of the project. The first term in
process it on a computer. Second, it does The MOLP model is an effective Z1 is the cost of crashing the project activ-
not guarantee the continuity of work need- method to analyze conflicting objectives ities, while the second term is the normal
ed to minimize cost, since each activity is in project scheduling. We have decided to cost. Since CNi,s is constant, it can be
started as soon as all of its preceding activ- consider the relationship between the omitted. In this case, the first objective
ities are finished. time (duration) and the cost of a project’s function (Z1) can be rewritten as
The graphical approaches plot repeti- schedule. The MOLP model generates a N P
tive activities as lines with slopes equal to set of efficient solutions (time and cost) Minimize Z1 = ∑ ∑ai,syi,s.
their production rates. The slopes can be that satisfy the constraints of the model. s=1 i=1 (equation 3)
constant or variable. The axes are distance To clarify the concept of an efficient
versus time. Graphical approaches do not solution in multiobjective mathematical The second objective function, Z2, is
guarantee the minimal direct cost for a programming, consider a MOLP with two the total project duration. Both of these
project, since they try to finish the project objectives, say Z1 and Z1. A solution where objectives are conflicting, since minimizing
as early as possible, under certain assump- Z1 = z1* and Z2 = z2* is considered to be a project’s duration requires additional cost,
tions of production rates. efficient if there is no other pair of values and vice versa. Efficient solution points of
Handa used the optimal control theo- z1 and z2 that satisfies z1 ≤ z1* and z2 ≤ the objective functions will be obtained so
ry to view the construction/production z2*. that the project constraints are satisfied.
process as a dynamic system that evolves These equations are subject to the fol-
over time [5]. The work is based on dis- lowing constraints.

42 Cost Engineering Vol. 38/No. 12 DECEMBER 1996


Constraints 1 and 2 Constraint 4 Constraints 5 and 6 are for each pair
TSi,s - TSh,s + yh,s > Dh,s + TLh,i yi,s < Di,s - di,s of concurrent activities i and j that need a
minimum number, M, of stages as a buffer
for for between them for practical and/or techno-
i = 1, 2, . . . , P and activity h is a prede- i = 1, 2, . . . , P and logical purposes. They constrain the time
cessor of activity i. s = 1, 2, . . . , N. that the crew working on activity j begins
(constraint 1) (constraint 4) at the first stage to be greater than or equal
to the time the crew working on activity i
TSi,(s + 1) - TSi,s + yi,s > Di,s Constraint 4 limits the number of starts working on stage M + 1. Also, they
days that an activity can be shortened in constrain the time that the crew working
for stage s to be less than or equal to the dif- on activity i start at stage N to be less than
i = 1, 2, . . . , P and ference in the activity’s normal and or equal to the time that the crew working
s = 1, 2, . . . , (N - 1). crashed durations, at the same stage. on activity j starts working on stage N - M.
(constraint 2) It should be noted that if the two activities
i and j have unequal durations, then activ-
Constraint 1 maintains a specified Constraints 5 and 6 ity i should be the activity with the shorter
time buffer between the starting time of M M
duration, to allow the two activities to fin-
activity i and each activity h that precedes TSj,s - TSi,s + ∑yi,s > ∑Di,s ish in a shorter time.
it, for each stage of the project. This is s=1 s=1 There are a number of computer soft-
achieved by constraining the time buffer for ware programs available to help work out a
between the start of activities i and h at concurrent activities i,j = 1, 2, . . . , P and problem in the form of a model. These
stage s, (TSi,s - TSh,s), to be greater than or s = 1, 2, . . . , M with a minimum stage programs are known as matrix generators
equal to the time needed. This time is buffer of M stages between them. [13]. The solution for the above MOLP
equal to the normal duration of activity h (constraint 5) could be obtained by using a program
in stage s, minus the number of days activ- called ADBASF [12].
ity h is shortened in stage s, plus the addi- N-M-1 N-M-1
tional lead time, if there is any [9]. TSj,N - TSj,1 + ∑yj,s < ∑Dj,s
Constraint 2 maintains the order for SAMPLE PROJECT AND ANALYSIS
s=1 s=1
each activity between different stages. This for
constraint is needed because each activity all activities i,j with a minimum stage buffer A sample project has been adopted
might have different durations during of M stages between them. [4, 9] that includes modifications that take
each stage. (constraint 6) into account having different durations for

Table 1—Cost and Duration Data for the Project Example


Constraint 3 Stages 1 and 3 Stage 4 Stages 2 and 5
N -1 N -1 Activity Activity D d a D d a D d a
TSi,N - TSi,1 + ∑ yi,s = ∑Di,s Abbrev. (days) (days) ($/day) (days) (days) ($/day) (days) (days) ($/day)
s=1 s=1
Locate & clear LC 1 1 - 1 1 - 1 1 -
for
Excavate EX 4 1 50 6 2 100 3 1 50
i = 1, 2, . . . , P. String pipe ST 2 1 50 4 2 100 3 1 50
(constraint 3) Lay pipe LA 5 1 150 6 2 150 6 2 150
Test TE 2 2 - 2 1 100 2 2 -
This constraint maintains the conti- Backfill BF 1 1 - 4 2 100 1 1 -
nuity of crew work on each activity D = normal duration/d = crashed duration/a = cost slope
between stages. To see this, we rewrite
constraint 3 as
Excavate
N -1 (2)
TSi,N = TSi,1 + ∑(Di,s - yi,s) i = 1, 2, . . . , P.
s=1
(rewritten constraint 3) Locate and
Lay Pipe Test Backfill
Clear
(4) (5) (6)
(1)
The second term on the right side is the
sum of the actual durations of activity i in
all but the last stage. This gives the com-
pletion time of activity i in stage N - 1, String Pipe
(3)
which is the start time of the same activity
in the last stage.
Figure 3—Typical Stage Network for the Project Example

Cost Engineering Vol. 38/No. 12 DECEMBER 1996 43


some activities in different stages. The Table 2—Sample Project Duration and Additional Crashing Cost
sample project is the relocation of 5 miles
(8 km) of natural gas pipeline. The project Project Duration Additional Crashing Cost Additional Crashing Cost ($)
(Days) ($) (Without Buffer Constraints)
is divided into five stages: each stage rep-
resents 1 mile (1.6 km). Table 1 presents
the project’s activities, their abbreviations, 40 0 -
normal durations, crashed durations, and 39 50 -
cost slopes at different stages of the project. 38 100 0
Figure 3 shows an activity on node net- 37 200 50
work for a typical stage of the project. 36 300 100
The mathematical formulation of the 35 400 200
34 500 300
sample project was developed first. It con-
33 600 400
sists of two objective functions, 83 con-
32 700 500
straints, and 47 variables. They are omitted 31 800 600
here due to space considerations. The 30 950 700
ADBASF computer program was used to 29 1100 850
solve the resulting two-objective linear 28 1250 1000
program. Some of the important findings 27 1400 1150
are described below. 26 1600 1300
The output file listed in table 2 pro- 25 1800 1450
vided efficient solutions to the problem. 24 2000 1600
The feasible project duration ranges from 23 2200 1800
22 2400 2000
40 days, which is the normal duration, to
21 2600 2200
15 days, which is the project’s crashed 20 2800 2400
duration. The additional costs correspond- 19 3000 2600
ing to each duration are also shown. 18 3300 2800
Table 2 also shows feasible project 17 3650 3100
durations versus the additional cost 16 4000 3450
required to minimize the project’s dura- 15 4350 3800
tion without stage-buffer constraints (con- 14 - 4200
straints 5 and 6). The feasible project dura-
tion is 2 days longer when time-buffer con- Table 3—Starting Time of Project Activities
straints are considered for the normal Project Duration = 40 Days Project Duration = 15 Days
durations. This increase is needed to meet Stage Number Stage Number
the additional stage-buffer constraints. Activity 1 2 3 4 5 1 2 3 4 5
The output also provides the start LC 0 1 2 3 4 0 1 2 3 4
time for each activity, in each stage. EX 3 7 10 14 20 2 3 4 6 8
Considering the stage-buffer constraints, ST 1 3 6 8 12 1 3 3 4 6
table 3 shows the start time of the activity LA 7 12 18 23 29 3 4 6 8 10
TE 27 29 31 33 35 4 6 8 10 12
when the project duration is 40 days (nor-
BF 32 33 34 35 39 9 10 11 12 14
mal duration) and when the project dura-
tion is 15 days (crashed duration).
projects. The computational efficiency of 2. Carr, R.I., and W.L. Meyer. Planning
the model needs to be studied further. A Construction of Repetitive Building

T
his article presents a mathemat- software program with a user-friendly inter- Units. Journal of the Construction
ical formulation of activity net- face needs to be developed to simplify the Division/ASCE 100 (no. 3 1974):
works that may have repetitive implementation. The repetitive project 403-412.
structure. The model uses mul- model discussed in this article can be used 3. Chrzanowski, E.N., and D.W.
tiobjective linear programming and to help minimize costs on construction Johnston. Application of Linear
attempts to minimize two conflicting projects that include repetitive actions. Scheduling. Journal of the Con-
objectives, namely time and cost. The struction Division/ASCE 100 (no. 3
mathematical formulation of the MOLP 1986): 476-491.
model has been examined using an illus- REFERENCES 4. Clough, R.H., and G.A. Sears.
trative sample. Construction Project Management.
The size of the MOLP model, in terms 1. Arditi, D., and Z.M. Albulak. Line-of- New York: John Wiley and Sons, 1979.
of the number of constraints needed to rep- Balance Scheduling in Pavement 5. Handa, V.K. Linear Scheduling Using
resent a project, depends on the number of Construction. Journal of the Con- Optimal Control Theory. Journal of
activities and stages. This can be computa- struction Division/ASCE 112 (no. 3 Construction Engineering and Man-
tionally inefficient for very large, repetitive 1986): 411-424. agement 112 (no. 3 1986): 387-393.

44 Cost Engineering Vol. 38/No. 12 DECEMBER 1996


6. Johnston, D.W. Linear Scheduling Dr. Kamal Al-Subhi are in mathematical programming,
Method for Highway Construction. Al-Harbi is an assistant sequencing and scheduling, the control of
Journal of the Construction Division/ professor of construc- transient queues, activity networks, the
ASCE 107 (no. 2 1981): 247-261. tion engineering and analysis of large data sets, and cluster
7. O’Brien, J.J. CPM Scheduling for management at King analysis. He is an associate editor of
High-Rise Buildings. Journal of the Fahd University of Pattern Recognition, published by
Construction Division/ASCE 101 Petroleum and Miner- Pergamon Press and serves on the editorial
(no. 4 1975): 895-905. als in Dhahran, Saudi board of the Arabian Journal for Science
8. O’Brien, J.J., F.C. Kreitzberg, and Arabia. He received his M.S. from the and Engineering, which is published by
W.F. Mikes. Network Scheduling University of Michigan and his Ph.D. from KFUPM.
Variations for Repetitive Work. Journal NCSU-Raleigh. He has an interdiscipli-
of the Construction Division/ASCE nary minor in industrial engineering, com- Mazen Al-Sinan is a
111 (no. 2 1985): 105-116. puter science, and business administration. graduate student in the
9. Reda, R.M. RPM: Repetitive Project He was part of the team who developed a department of construc-
Modeling. Journal of Construction statewide computerized bridge mainte- tion engineering and
Engineering and Management 116 nance management system for North management at King
(no. 2 1990): 316-330. Carolina. He has done consultations for Fahd University of
10. Russell, A.D., and W.F. Caselton. multimillion-dollar projects and has con- Petroleum and Miner-
Extensions to Linear Scheduling ducted numerous training seminars als in Dhahran, Saudi
Optimization. Journal of the Con- throughout the Arabian Gulf States. He is Arabia. He currently works as a contracting
struction Division/ASCE 114 (no. 1 a member of PMI and ASCE, and of engineer for the Saudi Arabian Oil
1988): 36-52. AACE International. Company (Saudi Aramco) in Dhahran.◆
11. Selinger, S. Construction Planning for Back to Table of Contents
Linear Projects. Journal of the Dr. Shokri Z. Selim is Index to Advertisers
Construction Division/ASCE 106 an associate professor at Reader
Response
(no. C02 1980): 195-205. the department of sys- Number Advertiser
12. Steuer, R.E. Operating Manual for tems engineering at 200 Primavera Project Planner,
the ADBASF Multiple Objective King Fahd University of Primavera Systems, Inc.,
Linear Programming Computer Petroleum and Minerals inside front cover.
Package. Athens, GA: University of (KFUPM) in Dhahran, 221 ICARUS Project Manager,
ICARUS Corp., page 39.
Georgia, 1991. Saudi Arabia. He ob- 225 Success, U.S. COST, back cover.
13. Williams, H.P. Model Building in tained his Ph.D. from the Georgia Institute 916 R.S. Means, page 17.
Mathematical Programming. New of Technology in industrial and systems 919 QuickGantt, Ballantine and
York: John Wiley and Sons, 1990. engineering. His current research interests Company, Inc., page 8.
924 CostTrack, OnTrack Engineering, Ltd.,
page 34.

1996 TRAofNSAC
TION S Now Available!
A ACE InNforTETotRN
The Associatio
AT ement
al Cost Manag
IONAL

ENGINEERS
1996 Transactions of
AACE International
A N D C OS T
C OS T M A N AG E R S

ULER S
AND SCHED
PL ANNERS

A N AG E R S
PROJECT M
If you were unable to attend the 1996 Annual Meeting in
VA L U E E N G
INEERS Vancouver, you still have a chance to take advantage of the
many technical papers that made their debut there!
S
E S T I M AT O R

Don’t miss out on this important, up-to-date reference material.


Order your copy of the 1996 Transactions today!

Proceedings of
the 40th Annual
Meeting of AAC
CANADA
E INTERNATIONAL Members Nonmembers
H COLUMBIA,
VANCOUVER, BRITIS
$7995 $9495
Call 800.858.COST or 304.296.8444
to order yours today!

Cost Engineering Vol. 38/No. 12 DECEMBER 1996 45

You might also like