Académique Documents
Professionnel Documents
Culture Documents
I N E M A N N
International Journal of Project Management Vol. 13, No. 3, pp. 187-194, 1995
Copyright ©1995 Elsevier Science Ltd
Printed in Great Britain. All rights reserved
0263-7863/95 $10.00 + 0.00
One of the criticisms of the PERT/CPM technique is that it ignores the uncertainty influence of
noncritical activities on project-duration estimation. As a result of this problem, the PERT/CPM
technique gives an optimistic estimation of project duration and float use. The paper introduces
a new concept of time-disturbance analysis for a project network. Time-disturbance analysis
is a process of simulating the change of the expected time of a given merge event or total project
duration with the changes in float use of noncritical activities. The computations of the late start
and finish times of the noncritical activities on the basis of the rational use of the float are
introduced by using the back-forward uncertainty estimation (BFUE) procedure, an analytical
risk-analysis model. Time-disturbance analysis can assist in project scheduling to reduce project
risks rationally.
Keywords: risk analysis, network scheduling, float
In project-duration estimation, there is a common pheno- the uncertainty of noncritical activities into the estimation
menon which can be called time disturbance. This refers to of project time. The actual calculated project duration can
the disturbance of the scheduled activities caused by the be longer than the one estimated with the PERT/CPM tech-
variance of activity durations in a network. A time dis- nique when the uncertainties of noncritical activities are
turbance can be caused by an activity finishing either later considered. Several techniques have been introduced to
than or in advance of the project-schedule date. In project correct the bias. These techniques include the analytical
planning, the disturbances caused by delays in some model developed by Clark 2 in 1961, the Monte Carlo
activities are of greatest concern. An activity delay can simulation model developed by Van Slyke in 19633, and
easily lead to the interruption of the scheduled following the analytical model developed by Ang et al. in 19754.
operations and a delay in the total project. Then, an adjust- However, these early pieces of research only concentrate
ment to the original project plan, acceleration of the fol- on the forward-pass calculation of a network. The use of
lowing activity operations, and an increase in the project float is not considered as a factor in influencing project
cost may be necessary. Avoiding an activity delay, or duration in risk-analysis-oriented scheduling techniques.
reducing the influence of an activity delay, is one of the The effect of float use on project duration can be ex-
main tasks in project time planning and project control. plained by the early research about near-critical activities.
A time disturbance can be caused by uncertainty of Consider that a noncritical activity can become a near-
critical activities in a network. We can estimate this type of critical activity with the increase of float use, or become a
time disturbance by using the PERT/CPM(Program Evaluation critical activity when the total float is used. Then, the float
Review Technique/Critical Path Method) technique. A use can influence the project duration if the results of the
time disturbance can also be caused by the uncertainty of early research about the influence of near-critical paths on
noncritical activities. In particular, when noncritical activities the total project duration are applied.
consume some or all of their floats, the possibility of a time The phenomenon of the influence of float use on project
disturbance caused by noncritical activities can increase. time is called the combined influence problem in this paper.
The PERT/CPMtechnique cannot estimate this type of time The combined-influence problem indicates the joint influence
disturbance. In early studies ~, this type of problem was of float use and the uncertainty of noncritical activities on
expressed in terms of the uncertainty of near-critical paths the project time. An unsolved problem in risk-analysis-
influencing the estimate of project duration. The study oriented network scheduling is that of how to measure the
results show that there is a bias relating to the estimate time risks caused by the combined influence.
using the PERT/CPMtechnique and the techniques combining The results of early research on the merge-event bias
187
Calculation of safe float use in risk-analysis-oriented network scheduling: D Gong and J E Rowings Jr
188
Calculation of safe float use in risk-analysis-oriented network schedufing: D Gong and J E Rowings Jr
use is as follows:
45-
i
44 t ]:U
*To simplify the calculation, the probability distributions of the activity
times are assumed to be normal for both activity A and activity B.
Equations 5 can be calculated using the author's prototype computer Figure 2 Influence of float use at merging activity B on expected
program or other software, such as Excel. time of network
189
Calculation of safe float use in risk-analysis-oriented network scheduling." D Gong and J E Rowings Jr
Event
: Critical Activity
L , Non-CriticalActivity
3,17(0.3t)
~ StandardDeviation
ActivityMeanTime
3.17(0.313E ,(~ 3(o3F
4,5(0.94) 3.67[0,833
23,17(3,44)4,17(1,56} 2(0.63)
J
activities should be adjusted to the safe float range so that Critical merge event
the float use will not lead to a critical disturbance or Merge event 10 is a critical merge event which connects
changes in the latest expected time of the merge event. three merging activities. To avoid a critical disturbance,
Then the latest start and finish times for each noncritical the expected time of merge event 10 should not be dis-
activity are calculated on the basis of safe float use. turbed. This implies that we should define a safe range of
There are two merge events in the network in Figure 3. float use for noncritical merging activities J and F so that
Merge event 8 is a noncritical merge event, and merge event the float use will not influence the expected time of merge
10 is a critical merge event. The time-disturbance analyses of event 10.
these two types of merge event are discussed separately. To define the safe range of float use for the merging
activities related to a critical merge event, we can go
through the following steps*. Let the safe range of float use
Table 1 Forward-pass calculation for network in Figure 3 with BFUE along a merging activity i be SFUi. In the case of n
procedure noncritical merging activities preceding a merge event,
Activity Event
• Step 1: set the float use F ~ equal to zero (j :~ i ; j = 1,
Activity EFT o Number E [t ] o 2 . . . . . i - 1, i + 1 . . . . . n) for each merging activity
A 4.50 0,94 2 4.50 0.94 except the merging activity i;
B 8,17 1,13 3 8.17 1.13 • Step 2: define SFUi, the safe range of float use along
C 12.17 1.30 4 12.17 1.30 the merging activity i, with the BFUE procedure.
G 13.08 2,32 5 13.08 2.32
H 17.34 3,10 6 17.34 3.10 Since merge event l0 is a critical event, its expected time
1 18.25 2.51 7 18.25 2.51
should be the same in both the backward-pass and forward-
E 11.34 1.17 pass calculations. The expected time is 40.41. Table 2
8 15.17 1.44
D 15.17 1.44
shows the calculation of the total float FT of merging
K 40.41 4.26 activities F, J and K. The total float is 16.74 weeks for
J 23.67 3.16 9-10 40.41 4.26
F 18.17 1.44
merging activity J, and 22.24 weeks for merging activity F.
Let the float use for the merging activities F and J be FUF
L 63.59 5,47 11 63.59 5.47
N 67.76 5.69 12 67,76 5.69
and FUj, respectively. We can analyze the influences of
O 69.76 5.72 13 69.76 5.72 FUe and FU~ on the expected time of merge event 10. The
following procedure analyzes the influence of FUj.
[Time unit: week. EFT: earliest finish time.]
• Step 1: Set FU F to zero.
Table 2 Float calculation for merging activities preceding merge • Step 2: Identify the nearest crosseventt of merging
event 10
Merging activity EFT Expected event time Total float
F 18.17 40,41 22.24 *A mathematical proof for applying these steps at a critical merge event
J 23.67 40,41 16.74 is presented in Reference 7.
K 40.41 40.41 0.00 tThe identification of the nearest crossevent copes with the assumption
of the basic model in Equations 2 and 3, The concept of the nearest
[Time unit: week. EFT: earliest expected finish time. ] crossevent is explained in Reference 6.
190
Calculation of safe float use in risk-analysis-oriented network scheduling: D Gong and J E Rowings Jr
activities F and K required in the BFUE procedure. The Table 3 Time-disturbance estimation at merge event 10 with float use
nearest crossevent here is event 3. at merging activity J
• Step 3: Substitute the earliest expected finish time, the FU Event 10 End event
time variances of merging activities F and K, and the
time variance of the nearest crossevent 3 (see Table 1) 0.00 40.41 69.75
10.00 40.62 69.96
in Equation 5 as follows: 12.00 40.88 70.22
15.00 41.67 71.01
t - 40.41 16.74 42.42 71.76
T~=
( 4 . 2 6 2 - 1.132) '~ [Time unit: week. ]
E[t]
45
Time unit: week
44
43
A
42
40.41
4.
40
39
88 FU
0 5 10 15 20
191
Calculation of safe float use in risk-analysis-oriented network scheduling: D Gong and J E Rowings Jr
indicates that SFUF along merging activity F is 18 weeks. D (see Figure 3), and the time variance of event 3 is
On the basis of the above time-disturbance analysis at 1.132. By substituting the earliest expected finish time, the
merge event 10, to keep the event time of merge event 10 time variances of merging activities E and D, and the time
undisturbed, we should restrict the use of the float to about variance of the nearest crossevent 3 into Equations 5, the
ten weeks for merging activity J, and about 18 weeks for expected time of merge event 8 can be set as a function of
merging activity F. Then, the latest expected finish time is the conditional SFUE as follows:
for activity F, where 18.17 is the earliest expected finish where 11.34 and 1.172 are the earliest expected finish time
time of activity J (see Table 4). and time variance of the merging activity E, and 15.17 and
1.442 are the earliest expected finish time and time
Noncritical merge event variance of the merging activity D.
A noncritical merge event has an earliest event time and By substituting for TE and To in Equation 2, and apply-
a latest event time. When the uncertainty of merging ing Equation 7, we can estimate the conditional SFUe of
activities influences the latest event time, the total project the merging activity E. The results in Table 5 show that,
time is disturbed. The time-disturbance analysis at a when the float use of merging activity E is less three weeks,
noncritical merge event identifies the safe float ranges of there is little change in the value of the earliest expected
the merging activities which do not disturb the latest event time of merge event 8. The conditional safe range of float
time. To analyze the time disturbance at a noncritical merge use for merging activity E is two weeks.
event, the following procedure is recommended: Since the difference between the latest and earliest
expected times of merge event 8 is 33.17 - 15.17 = 18
• Step 1: Identify the merging activity i which has, for the weeks (see Table 4), the safe range of float use for merging
noncritical merging activities, the largest value of the activity D is then 18 weeks. The safe range of float use for
earliest expected finish time. merging activity E is the sum of the conditional SFUF and
• Step 2: Define the conditional safe range of the float use 18 weeks. The calculation indicates that, when the float use
along a merging activity j which corresponds to the
merging activity i identified in step 1.
Table 4 Comparison of latest finish times for PERT/CPM and BFUE
• Step 3: Define the safe float use of a merging activity by
procedures
summing the conditional safe range of the float use and
the difference between the latest and earliest expected Activity Event
times of the noncritical merge event. Activity EFT LFT~P LFT~B Number Expected time*
192
Calculation of safe float use in risk-analysis-oriented network scheduling: D Gong and J E Rowings Jr
is limited to FUo <<,18 weeks for merging activity D, and E[t] Time unit week
FUF <~ 20 weeks for merging activity E, there is no 74
critical disturbance in the network. The latest expected
finish time of merging activity D can be calculated by
summing the earliest expected finish time EFT of activity D 72 .6
Monte Carlo Simulation Method
and the safe range of float use SFU for merging activity D.
We can compute the latest finish time of merging activity
E in similar fashion. The results are listed in Table 4. The 70 8
latest finish time is 31.34 weeks for merging activity E, and 69 ~ BFUE Procedure
33.17 weeks for merging activity D.
The remaining events in Figure 3 are each connected with 68
only one activity. On the basis of the derived latest finish 67
times of activities J, F, D and E with the safe use of float,
66J . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . FU
we can calculate the latest start time and finish time for 0 5 10 15
these remaining activities. The calculation rule is similar to
the backward-pass calculation in the CPM technique. The
Figure 5 Total project time influenced by float use at noncritical
results are also presented in Table 4.
activity J
Discussion and conclusions the expected time of a merge event can be viewed as the
One of the recognized conclusions from the merge-event change in the risk of time disturbance. In many cases, this
bias study is that the PERT procedure gives an optimistic change may not be significant. For instance, when the float
estimation of the total project time. This conclusion is use of merging activity J is 12 weeks, the expected time of
reached from the forward-pass calculation of network merge event 10 is 40.88 weeks (see Table 3). The expected
scheduling. By introducing the float-use concept into the time is only increased by 0.47 weeks in comparison with the
merge-event bias study, we can draw additional conclusions case in which the float use is zero in merging activity J. The
from the backward-pass calculation. The results from the magnitude of this difference can be ignored. Defining the
preceding sections show that the PERT/CPM procedure gives safe range of float use, therefore, may be related to the
an optimistic estimation for float use, because it ignores the project manager's attitude to the magnitudes of the changes
combined influence in a network. in the expected time of a merge event. In the above example,
Table 4 lists the latest expected finish times of the acti- if the 0.47 week change is an acceptable risk to a project
vities with the PERT/CaM procedure and 8FUE procedure. manager, the float use of 12 weeks can still be within the
The latest expected finish times of the noncritical activities safe range of float use. When the float use is about 16
are less in the BFUE procedure than in the PERT/CPM pro- weeks, the increase in the expected time of merge event 10
cedures. This occurs because the BFUE procedure restricts is two weeks, compared with no float use along merging
the float use from the total float FTto the safe float use SFU path J. The two-week delay could be an unacceptable risk.
to avoid a critical disturbance in a network. In the BFUE In this case, the float use of 16 weeks is not within the safe
procedure, the latest expected finish time is the sum of the range of float use.
earliest expected finish time of a noncritical activity and the In a risk-analysis problem, indicating the trend in the
safe range of float use SFU. In the PERTprocedure, however, time disturbance may be more important than giving a
it is the sum of the earliest expected finish time for a precise value of the time disturbance. To see whether the
noncritical activity and the total float FT. Since the safe BFUE procedure can indicate the trend in the time disturbance
range of float use SFU is less than the total FT, the latest in a network properly, the result of the BFUE procedure for
expected finish times of the noncritical activities are, time-disturbance analysis is compared with the result of
therefore, less in the BFUE procedure than in the PERT/CPM Monte Carlo simulation. Since the results of Monte Carlo
technique. simulation can vary when different assumptions are made
The earliest and the latest expected finish times of the about the activity duration's probability distributions, this
critical activities are the same in the BFUE procedure (see comparison is carried out under one condition: the mean
Table 4). This is also a conclusion from the PERT/CaM activity durations and their variances are approximately the
technique. However, the PERT/CPM technique achieves this same for the aFUE procedure and the Monte Carlo simulation.
by ignoring the combined influence on the total project This condition can be achieved by adjusting the probability-
time. Unlike the PERT/CPM technique, the BFUE procedure density distributions of the activity duration in the Monte
produces this result by restricting the float use of the Carlo simulation according to the same data of optimistic,
noncritical activities to eliminate the critical disturbances pessimistic and most likely estimation used in both the BFUE
caused by the combined influence. The introduction of safe procedure and the Monte Carlo simulation.
float use in a network can lead to the elimination or The two curves in Figure 5 show the changes in the total
reduction of the risk of project time overruns caused by project time with respect to the change of float use at
noncritical activities. activity J. The broken-line curve is the result from the BFUE
Figure 4 shows the changes in the expected time of merge procedure. The solid curve shows the result from the
event 10 with respect to the changes in the float use of the Monte Carlo method*. The two curves develop in almost
two noncritical merging activities. When the float use is
zero, the expected time of merge event 10 is 40.41 weeks.
With an increase in float use, the expected time of merge *The Monte Carlo simulation software used was supported by Primavera
event 10 increases gradually. The changes of magnitude of Systems.
193
Calculation o f safe float use in risk-analysis-oriented network scheduling: D Gong and J E Rowings J r
the same manner. This shows that the results of simulating References
the trend in the total project time with respect to the float
1 MacCrimmon, K R and Ryavec, C A 'An analytical study of the
use are similar for both the BFUE procedure and the Monte
PERT assumptions' Operations Research 1964 12 (1)
Carlo method. 2 Clark, C E 'The greatest of a finite set of random variables'
On the basis of the above discussions, w e can draw Operations Research 1961 9 (2)
several conclusions, as follows: 3 Van Slyke, R M 'Monte Carlo methods and the PERT problem'
Operations Research 1963 11 (5)
• The PERT/CPM technique gives an optimistic estimation 4 Ang, A H-S, Abdelnour, J and Chaker, A A 'Analysis of activity
networks under uncertainty' J Engineering Mechanics Division, Proc
of not only the total project time, but also the float
American Society of Civil Engineers 1975 101 (EM4)
available for noncritical activities. 5 Moder, J J, Phillips C R and Davis, E W Project Management with
• When the float use for noncritical activities is limited to CPM, PERT and Precedence Diagramming (3rd ed) Van Nostrand
the safe float use, the risk of time overrun caused by Reinhold (1983)
noncritical activities can be eliminated or reduced. 6 Gong, D and Hugsted, R 'Time uncertainty analysis in a project
network with a new merge-event time-estimation technique' Int J
• The definition of the safe range of float use can be Project Management 1993 11 (3)
related to the project manager's attitude to the time- 7 Gong, D 'Risk analysis oriented network scheduling - - backward pass
disturbance level. calculation and time disturbance analysis' Dissertation (1993)
194