Professional Documents
Culture Documents
Jurnal CPM Pert
Jurnal CPM Pert
ABSTRACT: Formal stochastic simulation study has been recognized as a remedy for the shortcomings inherent
to classic critical path method (CPM) project evaluation and review technique (PERT) analysis. An accurate and
efficient method of identifying critical activities is essential for conducting PERT simulation. This paper discusses
the derivation of a PERT simulation model, which incorporates the discrete event modeling approach and a
simplified critical activity identification method. This has been done in an attempt to overcome the limitations
and enhance the computing efficiency of classic CPM/PERT analysis. A case study was conducted to validate
the developed model and compare it to classic CPM/PERT analysis. The developed model showed marked
enhancement in analyzing the risk of project schedule overrun and determination of activity criticality. In ad-
dition, the beta distribution and its subjective fitting methods are discussed to complement the PERT simulation
Downloaded from ascelibrary.org by New York University on 05/16/15. Copyright ASCE. For personal use only; all rights reserved.
model. This new solution to CPM network analysis can provide project management with a convenient tool to
assess alternative scenarios based on computer simulation and risk analysis.
ities based on the ratio of average slack (TF ) to activity du- tion of the classic CPM network, which is built using the ac-
ration standard deviation and the criticality index.’’ Criticality tivity-on-arrow structure. The network is constructed with two
index CI for an activity in a percentage term is defined as the basic elements that are interconnected by arrows; these are the
number of simulation runs in which the activity is critical, merge node and the burst node. In the proposed model, a
divided by the total number of simulation runs. merge node is strictly defined as a node with two or more
Halpin and Riggs (1992) presented a PERT simulation ap- incoming branches (activity arrows) and only one outgoing
proach using CYCLONE (Halpin 1990). By performing non- branch, and a burst node is defined as a node with only one
cyclic simulation in CYCLONE, the forward pass calculation incoming branch and two or more outgoing branches. If a node
was solved and the statistics for total project completion time has only one incoming branch and only one outgoing branch,
were obtained; these were the average project completion it is treated as a burst node for the sake of simplicity in cal-
time recorded by a COUNTER node of CYCLONE. However, culation, because no attribute needs to be tracked and deter-
Halpin and Riggs pointed out that ‘‘a somewhat troublesome mined in the calculation for a burst node. (This is clarified
aspect in the interpretation of any computer simulation of a later in the section entitled Forward Pass Calculation.) If a
network regards the critical path.’’ They identified the statistics node has more than two incoming branches and more than
of the waiting times of entities converging at a merge node in two outgoing branches, as is often encountered in the CPM
the CPM network as a good indication of activity criticality; network, it must be changed into a merge node and a burst
the lower the average waiting time, the more critical the ac- node by adding one dummy activity between nodes as shown
tivity (Halpin and Riggs 1992). in Fig. 1.
The limitation of this CPM is that the ‘‘amount of time an Only one terminal node is allowed to exist in a network to
entity had to wait in a QUEUE node (converging activity) earmark the finish milestone of the whole project. A dummy
before being processed by the following COMBI activity activity (zero duration) is added to link the last node of the
(merge node)’’ is actually the local slack or free float in CPM CPM network to a terminal node, as shown in Fig. 2.
terminology. According the CPM analysis, the condition re- As long as a node has no preceding activities or incoming
quired for an activity to be deemed critical is for the TF value branches, it is defined as an origin node. The following rules
apply to the origin node:
{2N ⫹ 1}i (6) methods are overviewed, and the methods of fitting beta dis-
i=1 tribution based on subjective time estimates are discussed.
where m = total number of merge nodes (note that the terminal Beta Distribution
node here is also treated as a merge node with only one in-
coming branch); and N = number of converging branches for Beta distribution, which was originally used in the classic
merge node I, (N ⱖ 2, except N = 1 for the terminal node). CPM-based PERT to model activity duration, has been thor-
The simplified CPM/PERT simulation ignores the burst oughly studied and conditioned. Many methods have been
nodes and origin nodes in the forward pass and does not need found to fit a beta distribution and have been documented in
to calculate TF in the backward pass. Therefore, much fewer literature (AbouRizk et al. 1991, 1994). When a sufficient
event time variables are stored than with the classic CPM, in amount of sample data is available (this may be historically
which five time variables are required for every activity (ES, accumulated or simulation generated), an objective approach
EF, LS, LF, and TF ). Additionally, in the classic CPM, the can be taken to fit the parameters of beta distribution to the
duration of every activity needs to be stored in a variable for sample data. If sample data is unavailable, or if historical data
the purpose of the backward pass calculation. So each activity is missing or too expensive and time-consuming to be col-
needs a total of six time variables for classic CPM/PERT anal- lected, a subjective approach may be utilized to fit a beta dis-
ysis. In contrast, in the simplified CPM/PERT simulation, ac- tribution based on human experience. Furthermore, during the
tivity duration is used only once in the forward pass calcula- project planning and execution stages of a project, subjective
tion and does not need to be stored. Given a CPM network beta distribution fitting provides a more flexible and effective
with 17 activities, the simplified CPM/PERT takes up only means for the project manager to experiment with different
one-third of the time variables that are needed for the classic scenarios and analyze the risk of schedule overrun through
CPM analysis. This means savings of computer memory and PERT simulation. This is demonstrated later in the section en-
enhancement of computing efficiency, which will become even titled Sample Application. In short, an accurate and efficient
more considerable for large-scale networks. method to fit the beta distribution parameters based on sub-
222 / JOURNAL OF CONSTRUCTION ENGINEERING AND MANAGEMENT / MAY/JUNE 2000
jective time estimates will make PERT simulation more useful TABLE 2. Base Case Scenario: PERT Simulation Results ver-
to assist risk analysis and decision making. sus Classic PERT Analysis
Downloaded from ascelibrary.org by New York University on 05/16/15. Copyright ASCE. For personal use only; all rights reserved.
Simplified
Subjective Beta Distribution Fitting Statistics of project CPM/PERT Classic CPM/
completion time simulation PERT analysis
Three time estimates describing the activity duration are (days) results results
used in fitting the beta distribution (McLaughlin and Pickhardt (1) (2) (3)
1997):
Minimum duration 39.3 N/A
Maximum duration 48.9 N/A
• The ‘‘most optimistic’’ time, which is an estimate of the Mean 43.6 43.5
minimum time required to complete an activity and be Standard deviation 2.1 2.2
obtained only if everything goes right. 95% confidence interval 43.2–44.1 N/A
• The ‘‘most pessimistic’’ time is an estimate of the maxi- Probability of completing by Day 45 72% 75%
mum time required to complete an activity and be ob-
tained only if everything goes wrong. TABLE 3. Base Case Scenario: Comparison of Activity Criti-
• The ‘‘most likely’’ time is an estimate of the normal time cality and Path Criticality
required to complete an activity. It is an estimate of the
result that would occur most often if the activity were CI (simplified On/off critical
repeated a large number of times. CPM/PERT path (classic
simulation) CPM/PERT
Activity description (%) analysis)
The two shape parameters of the beta distribution can be de-
(1) (2) (3)
termined using five methods from subjective time estimates
(AbouRizk et al. 1991). The mean and variance specified 1-3 (Prefab metal building) 85 On
1-2 (Clear site) 15 Off
method is found to be accurate and efficient to fit the shape 2-3 (Foundations) 15 Off
parameters and is recommended for use in the simplified CPM/ 3-4 (Erect building) 100 On
PERT simulation. 4-5 (Finish building) 100 On
SAMPLE APPLICATION
standard deviation of which are obtained based on the central
Base Case Scenario limit theorem. In the sample application, the mean and stan-
dard deviation of a normal distribution from the classic PERT
The sample CPM network for building a convenience store
analysis are determined as 43.5 and 2.2, respectively [refer to
is taken from Ahuja et al. (1995, Chapter 15) as shown in Fig.
Ahuja et al. (1995) for details]. Next, the control date is con-
2, which was originally used to illustrate the PERT calculation.
verted to a standardized term for the standard normal distri-
The activity description, the factors causing uncertainty, and
bution [(45 ⫺ 43.5)/2.2 ⫺ 0.6818]. By looking up the standard
the time estimates of activity duration are listed in Table 1. In
normal distribution table, the cumulative probability of com-
the sample application, the simplified CPM/PERT simulation
pleting the project by Day 45 can be determined to be 75%.
and mean and variance specified method are utilized to do a
The comparison in Table 2 exposes the drawback of the
PERT simulation analysis. Note that in the CPM network (Fig.
classic CPM/PERT analysis due to the merge event bias: the
2); node 1 is an origin node with the control time set to zero;
mean calculated by PERT is an underestimate of the mean
node 6 is the terminal node; node 3 is a merge node; and nodes
resulting from simulation (43.5 < 43.6), and the schedule over-
2, 4, and 5 are the three burst nodes.
run risk is also underestimated (72% < 75%). If the network
A program is developed using Microsoft Excel 97 and Vi-
is larger and has more merge nodes, the bias will become more
sual Basic for application to conduct beta distribution fitting
obvious.
and random sampling and to implement the simplified CPM/
In Table 3, the activity criticality in terms of CI resulting
PERT simulation. For the base case scenario, the simulation
from simplified CPM/PERT simulation is compared to the path
results are generated after 100 runs as listed in Tables 2 and
criticality resulting from the classic CPM/PERT analysis. It is
3, and are compared with classic CPM/PERT analysis results
observed that simplified CPM/PERT simulation can better
if applicable. The histogram and cumulative probability curve
quantify the criticality of an activity than the classic CPM/
of the project completion time drawn from the simulation re-
PERT analysis. This is more obvious in Scenario 1.
sults are shown in Fig. 3.
Note that for the simplified CPM/PERT simulation, the
Scenario 1
probability of completing the project by Day 45 is read from
the generated cumulative probability curve, as shown in Fig. If the general contractor feels uncomfortable with the over-
3. For the classic CPM/PERT analysis calculation, the proba- run risk of project schedule, it is easy to do some simulation
bility of completing the project by Day 45 is calculated from experiments by changing duration estimates for an alternative
a theoretical normal cumulative distribution, the mean and scenario. The following scenario is an example: What if the
JOURNAL OF CONSTRUCTION ENGINEERING AND MANAGEMENT / MAY/JUNE 2000 / 223
general contractor cooperates with another fabrication subcon- TABLE 4. PERT Simulation Results: Scenario 1 versus Base
tractor who is able to deliver the job in a shorter time? Based Case
on factors such as reputations and past experiences on similar Simplified Simplified
work, the three time estimates for activity 1-3 are reevaluated Statistics of project CPM/PERT CPM/PERT
for an alternative supplier as L = 18, M = 20, and U = 22. completion time simulation simulation
Next, the general contractor runs the simplified CPM/PERT (days) scenario 1 base case
simulation again and obtains the results for a different scenario (1) (2) (3)
(scenario 1) as shown in Tables 4 and 5 and Fig. 4. Minimum duration 37.4 39.3
Table 4 compares the overrun risk of project schedule to Maximum duration 50.1 48.9
that of the base case scenario as well, showing that turning to Mean 42.4 43.6
the alternative supplier can reduce the overrun risk of project Standard deviation 2.7 2.1
95% confidence interval 41.9–42.9 43.2–44.1
schedule. For example, the mean project completion time is
Probability of completing by Day 45 85% 72%
reduced from 43.6 to 42.4 days, and the probability of com-
pleting the project within 45 days increases from 72% in the
base case scenario to 85% in scenario 1.
The change of CI shown in Table 5 reveals that the value
of CI for activity 1-3 decreases from 85 to 48%. With the TABLE 5. Scenario 1: Comparison of Activity Criticality and
classic CPM/PERT analysis, activity 1-3 is off the critical path Path Criticality
and therefore may not be brought to the project manager’s CI (simplified On/off critical
attention. However, the simplified CPM/PERT simulation re- CPM/PERT path (classic
sults give a CI value for activity 1-3 that is only 4% less than simulation CPM/PERT
that of activities 1-2 or 2-3, which are on the critical path from scenario 1) analysis
the classic CPM/PERT calculation. Therefore, to complete the Activity description (%) scenario 1)
convenience store project within schedule, the general con- (1) (2) (3)
tractor should almost equally prioritize prefab metal building 1-3 (Prefab metal building) 48 Off
(activity 1-3), clear site (activity 1-2), and foundations (activity 1-2 (Clear site) 52 On
2-3). This demonstrates the advantage of the activity criticality 2-3 (Foundations) 52 On
3-4 (Erect building) 100 On
concept in PERT simulation over the path criticality concept 4-5 (Finish building) 100 On
in the classic CPM/PERT analysis.
Part 1 of Rule 1 can be put into a simple mathematical For a merge node I given CIIJ = 0, then CIkI = 0
Downloaded from ascelibrary.org by New York University on 05/16/15. Copyright ASCE. For personal use only; all rights reserved.
representation as follows:
for all the converging branches at the merge node I
For a merge node I given CIIJ = 1, if WTkI = 0, then CIkI = 1; Because CIIJ = 0, the emanating activity I-J must be noncritical
otherwise CIkI = 0 so that we can have TFIJ > 0 or
where CIIJ = criticality index variable for activity I-J, which ESIJ < LSIJ (14)
emanates from node I and flows into node J, as shown in Fig. We also know
5; WTkI = waiting time of an entity flowing into merge node
I from converging branch k; CIkI = criticality index of con- EFkI ⱕ ESIJ (15)
verging branch k; k = 1, 2, 3, . . . , N, N = total number of
i.e., the earliest finish time of converging branch k must be
converging branches. (Seasoned schedulers will soon realize
less than or equal to the earliest start time of merge node I.
the equivalence of the term WT in the classical CPM: the local
Now from (14) and (15), we can have
slack or free float.)
Because CIIJ = 1, branch I-J must be on the critical path by EFkI < LSIJ (16)
the definition of CI. According to the classic CPM, TF for the
branch I-J (TFIJ) must be 0, therefore It also holds true that
Hence i.e., the latest start time of merge node I is equal to the latest
finish time of any converging branch. Now from (16) and (17),
ESIJ = LSIJ (8) we can have EFkI < LFkI , i.e.
where ESIJ = earliest start time of activity I-J in the classic TFkI = LFkI ⫺ EFkI > 0 (18)
CPM terms; and LSIJ = latest start time of activity I-J in the
classic CPM terms. If CIIJ = 0, any converging branch at the merge node is non-
By the definition of the arrival time of converging branch critical; hence, the CI values for all of them should be equal
k at merge node I, ATkI is equivalent to the earliest finish time to 0, i.e., CIkI = 0, which proves the second part of Rule 1.
of converging branch EFkI , therefore
Proof of Rule 2
EFkI = ATkI (9)
Rule 2 can be put into a simple mathematical representation
From (8) and (9) we can have as follows:
WTkI = DT(I ) ⫺ ATkI = ESIJ ⫺ EFkI (10) For a burst node I given 兺 CIIk = 0, then CIJI = 0;
where DT(I ) = entity departure time or earliest release time otherwise, i.e., 兺 CIIk > 0, then CIJI = 1
from node I and should be equal to the ESIJ .
If WTkI = 0, then from (10) ESIJ ⫺ EFkI = 0, and so we have where k = 1, 2, . . . , N, N = total number of outgoing branches
ESIJ = EFkI from (8), and at burst node I shown in Fig. 6.
If 兺 CIIk = 0 (i.e., none of the emanating branches at the
LSIJ = EFkI (11) burst node is critical, hence TFIk > 0 for every emanating
branch) or LSIk ⫺ ESIk > 0, we have
In the classic CPM backward pass calculation, we have
ESIk < LSIk (19)
LFkI = LSIJ (12)
At burst node
i.e., the latest finish time of converging branch k is equal to
the latest start time of merge node I. ESIk = ES(I ) (20)
From (11) and (12), we have
i.e., the earliest start times of all the emanating activities equal
LFkI = EFkI (13) to the earliest start time of burst node I
FIG. 5. Illustration for Proving Rule 1 FIG. 6. Illustration for Proving Rule 2
i.e., the latest start time of burst node I is equal to the latest Pidd, M. (1996). Tools for thinking, modelling in management science.
Wiley, London.
finish time of its incoming branch J-I. Pritsker, A., Sigal, C., and Hammesfahr, R. (1989). SLAM II network
From (22)–(24), we can conclude EFJI < LFJI or TFJI = LFJI models for decision support. Prentice-Hall, Englewood Cliffs, N.J.
⫺ EFJI > 0, thus the incoming branch J-I burst node I is not ProjectGear Inc. (Risk⫹ for MS Project). 具http://www.cs-solutions.com/
critical, and a 0 value is accordingly assigned to CIJI , which riskplus.htm典
proves the first part of Rule 2. ‘‘Monte Carlo娃 3.0 for Primavera.’’ (1995). Project Risk Analysis Soft-
If 兺 CIIk > 0 (i.e., at least one of the emanating branches, ware, Primavera Systems, Inc., Bala Cynwyd, Pa.
say branch k, at burst node I is critical), we have TFIk = 0 or
ESIk = LSIk (25) APPENDIX III. NOTATIONS
For burst node I, according to the classic CPM The following symbols are used in this paper:
ACKNOWLEDGMENTS
Subscripts
This work was funded by a number of construction companies in Al-
berta, Canada, and by the National Science and Engineering Research IJ = branch or activity starting from node I and ending at node
Council of Canada under grant number IRC-5558/96. J;
I, J = network node ID;
APPENDIX II. REFERENCES Ik = kth outgoing branch linking with burst node I;
AbouRizk, S. M., Halpin, D. W., and Wilson, J. R. (1991). ‘‘Visual in- k = positive integer index, k = 1, 2, 3, . . . , N; Ik or kI also
teractive fitting of beta distributions.’’ J. Constr. Engrg. and Mgmt., stands for kth branch linking with node I; and
ASCE, 117(4), 589–605. kI = kth incoming branch linking with merge node I.