Professional Documents
Culture Documents
Abstract: Critical path analysis on a project network having non-finish-to-start 共FS兲 logical relationships with lags is generally referred
to as precedence diagram method 共PDM兲. A PDM-based scheduling analysis is facilitated by mainstream project scheduling software
共such as P3兲. However, PDM compounds total float determination and interpretation, potentially causing anomalous effects on critical
Downloaded from ascelibrary.org by New York University on 05/15/15. Copyright ASCE. For personal use only; all rights reserved.
path identification. In the present research, we generalize those particular circumstances that entail applying non-FS logical relationships
on construction projects. We then propose generic transform schemes such that non-FS relationships in a PDM network can be detected
and transformed—automatically—into equivalent FS. Moreover, we provide analytical proofs for the transform schemes being proposed
to justify the logical equivalency between the original PDM network and the transformed activity-on-node 共AON兲 network only having
FS logical relationships. A PDM network example demonstrates that confusions would arise in interpreting P3’s critical path analysis
results, but not in the case of the transformed AON counterpart. In conclusion, the transform schemes being proposed lead to better
understanding of the scheduling results when critical path analysis is performed on a PDM network. This also paves the way for
conducting further sophisticated scheduling analysis 共such as resource loading or Monte Carlo simulation兲 on a PDM network.
DOI: 10.1061/共ASCE兲CO.1943-7862.0000062
CE Database subject headings: Scheduling; Critical path method; Construction management.
Introduction and Background to a more complicated form called precedence diagram method
共PDM兲.
Since its formalization in the 1950’s, the critical path method After the circle-and-connecting-line technique 共later called
共CPM兲 has contributed significantly to planning, control, and per- AON兲 was introduced in a Navy Bureau of Yards and Docks
formance monitoring of construction projects. To enable CPM- study in the 1960’s, Fondahl 共1961兲 recognized that AON-based
based scheduling analysis, it is a prerequisite to apply network critical path analysis could not allow overlapping unless activities
diagramming techniques so to represent a project with activity were further divided. The PDM was introduced in a 1964 IBM
publication 共the user’s manual for the IBM 1440 program兲, using
nodes linked by precedence relationships that are imposed by
three relationships 关finish-to-start 共FS兲, start-to-start 共SS兲, and
construction technology or any other logical constraints on the
finish-to-finish 共FF兲兴 and positive lags to depict partially concur-
project. The resulting network diagram serves as a model of a
ring or overlapped working progress between activities 共Moder
complex project system and facilitates analysis and control of
and Philips 1970兲. The start-to-finish 共SF兲 relationship and nega-
time, resource, and cost. In comparison with the activity-on-arrow
tive lags were later added to the PDM by Ponce-Campos 共1970兲.
format, activity-on-node 共AON兲 network diagramming does not The PDM permits modeling mutually dependent activities, which
require the use of any artificial dummy activities and allows the are performed partially in parallel instead of in series. In short,
planner to specify more than one logical relationship between compared with conventional AON networks featuring FS logic
activities. only, PDM networks with “smart” relationships are more com-
Moreover, AON lends itself well to the object oriented pro- pact, flexible, and realistic to represent construction projects 共i.e.,
gramming scheme as implemented in major project management Harris 1978; Wiest 1981; Valls and Lino 2001; Lock 2003兲.
software systems 共i.e., each activity is a specific instance of the Nowadays, popular commercial CPM scheduling software
generic “activity” object definition兲. Herein, we present a brief systems—such as Primavera Project Planner 共P3, Oracle Corp-
history of AON to shed light on how AON originated and evolved oration, Redwood Shores, Calif.兲 and Microsoft Project 共MS
Project, Microsoft Corporation, Redmond, Wash.兲—provide the
1
Associate Professor, Construction Engineering and Management, functionality for defining four relationship types between activi-
Dept. of Civil and Structural Engineering, Hong Kong Polytechnic Univ., ties.
Hong Kong, China 共corresponding author兲. E-mail: cemlu@ Harris 共1978兲 and Ioannou and Harris 共2003兲 emphasized the
polyu.edu.hk
2
need for the use of two links 共SS and FF兲 between two overlap-
Research Associate, Dept. of Civil and Structural Engineering, Hong ping activities to model consecutive construction processes re-
Kong Polytechnic Univ., Hong Kong, China. E-mail: ceching@ peating on multiple work units. Notably, MS Project only permits
polyu.edu.hk
specifying one relationship between two activities, while P3 is
Note. This manuscript was submitted on June 24, 2008; approved on
March 4, 2009; published online on August 14, 2009. Discussion period deemed more construction friendly by providing the flexibility of
open until February 1, 2010; separate discussions must be submitted for connecting two activities with more than one relationship. In Fig.
individual papers. This paper is part of the Journal of Construction 1, we generalize two common cases for possible use of two smart
Engineering and Management, Vol. 135, No. 9, September 1, 2009. relationships to connect two construction activities. The case of
©ASCE, ISSN 0733-9364/2009/9-863–873/$25.00. “SS/FF staggered” uses a pair of SS/FF to relate two partially
tains SS/FF/SF relationships with positive lag time. Note a SS b) Equivalent AON Network (without lag)
relationship with 1-day lag and a FF relationship with 1-day lag
relate B to C while a SF relationship with 1-day lag links E to D. Fig. 4. PDM-to-AON conversion—staggered SS/FF
All the other relationships are straightforward FS with zero lag.
The activity times are given in Table 1. Next, we entered the
project data in P3 and executed P3 under the interruptible option
for scheduling activity duration. P3-produced CPM analysis re-
sults are shown in Fig. 3 and in Table 1. finish end. For example, the TF is zero on Activity B’s start,
Note, P3 provides three options for deriving TF for an activity, but 2 days on Activity B’s finish; Activity D has 4-day TF on
namely; 共1兲 the start float 共i.e., the difference between late start its start and 2-day TF on its finish.
time and early start time兲; 共2兲 the finish float 共i.e., the difference 3. By articulating activity start ends or activity finish ends with
between late finish time and early finish time兲; or 共3兲 the most zero TF, a continuous critical path is identified to span Ac-
critical float 共i.e., taking the smaller value of start float and finish tivity A—FS relationship linking A and B—B’s start end—SS
float.兲 The following observations can be made from P3’s sched- relationship linking B and C—Activity C—FS relationship
uling results: linking C and E—Activity E. Adding up activity times and
1. By the “most critical” option, all activities are found critical relationship lags along the critical path results in the total
except for Activity D 共TF= 2 days兲. project duration of 12 days.
2. By contrasting the start float and the finish float, given one The project network used in the case study is small and
particular activity, the presence of non-FS logical relation-
simple. Yet, it is not straightforward to comprehend and commu-
ships leads to various float values on the start end and the
nicate the scheduling results, making it difficult to utilize them to
control project execution. In particular, the following two issues
remain unclear and hence justify further clarification: 共1兲 the TF is
Table 1. Activity Times and P3-Produced TF Results in PDM Case zero on B’s start, but 2 days on B’s finish; what does that imply?
Study 共2兲 How come D has 4d TF on start while only 2 days TF on
TF finish? To clarify the CPM results, we develop transform schemes
Duration 共most critical, for analytically processing non-FS relationships with positive lags
Activity 共days兲 start float, end float兲 in PDM into equivalent FS with zero lag, as presented in the
A 3 0,0,0 following sections.
B 3 0,0,2
C 5 0,0,0
D 2 2,4,2 Transform Scheme for SS/FF Staggered
E 3 0,0,0
Fig. 4 shows the transformation of a SS/FF staggered case into
straightforward FS relationships. Note activity duration is shown
in the bracket inside the activity node and the relationship lag
time is marked next to the relationship arrow. A1 and A2 can be
taken as components of the original Activity A. The duration of
Activities A1 and A2 is “s” 共equal to the SS lag time兲 and “a − s”
共i.e., the duration of whole Activity A minus SS lag time兲, respec-
tively. Similarly, the duration of Activities B1 and B2 is “b − f”
共duration of whole Activity B minus FF lag time兲 and “f” 共FF lag
time兲, respectively. To simplify mathematical representations, we
assume the project start time is zero; and a relative origin on the
timeline aligns with the start event of the preceding Activity A.
The equivalency of the two networks in Fig. 4 共PDM versus
AON兲 in terms of CPM analysis can be analytically proven as
Fig. 3. P3 scheduling results output snapshot will be discussed next.
Forward Pass Calculation Start float Finish float Start float Finish float
For activities with no predecessor relationships, ES = 0 Activity 共LS-ES兲 共LF-EF兲 Subactivity 共LS-ES兲 共LF-EF兲
A 0 T−a− f Al 0 0
Activity A: ESA = 兵project start time = 0其 = 0 共1兲
A2 T−a− f T−a− f
EFA = 兵ESA + DA = 0 + a = a其 = a 共2兲 B T−b−s 0 Bl T−b−s T−b−s
再 冎
B2 0 0
project start time = 0
共3兲
再 冎
Activity B: ESB = max =s
ESA + SSAB = 0 + s
再 冎
EFA2 = a
T = max = max共a + f, b + s兲
ESB + DB = s + b EFB2 = max共a + f, b + s兲
Downloaded from ascelibrary.org by New York University on 05/15/15. Copyright ASCE. For personal use only; all rights reserved.
T = max 再 EFA = a
EFB = max共a + f, b + s兲
冎 = max共a + f, b + s兲 共5兲
Backward Pass Calculation
For activities with no successor relationships, LF = T
再 冎
LSB1 = 兵LFB1 − DB1 = 共T − f兲 − 共b − f兲其 = T − b 共22兲
再 冎
total project duration = T
Activity A: LFA = min =T−f
EFB − FFAB = T − f total project duration = T
Activity A2: LFA2 = min =T−f
共8兲 LSB2 = T − f
再 冎
共23兲
LFA − DA = T − a − f
LSA = min = min共T − a − f,T − b − s兲 = 0
LSB − SSAB = T − b − s LSA2 + 兵LFA2 − DA2 = 共T − f兲 − 共a − s兲其 = T − a − f + s 共24兲
再 冎
共9兲
LSA2 = T − a − f + s
Activity A1: LFA1 = min
LSB1 = T − b
Critical Path Analysis on Activity-on-Node Network
= min共T − a − f + s, T − b兲 共25兲
Forward Pass Calculation
For activities with no predecessor relationships, ES = 0 LSA1 = 兵LFA1 − DA1 = min共T − a − f + s,T − b兲 − s其
Activity A1: ESA1 = 兵project start time = 0其 = 0 共10兲 = min共T − a − f, T − b − s兲 共26兲
EFA1 = 兵ESA1 + DA1 = 0 + s = s其 = s 共11兲 The above deductions on the PDM network and the counterpart
AON network arrive at the identical total project time 共T兲, which
Activity A2: ESA2 = 兵EFA1 = s其 = s 共12兲 is max共a + f , b + s兲. In Table 2, we further contrast the TFs result-
ing from early or late activity times, as determined from the for-
EFA2 = 兵ESA2 + DA2 = s + 共a − s兲 = a其 = a 共13兲 ward and backward analysis on the two networks.
The following interpretations of PDM results are made by re-
Activity B1: ESB1 = 兵EFA1 = s其 = s 共14兲 lating to those results from the equivalent AON counterpart:
1. The start float on A is 0 and the finish float on A is T − a − f.
EFB1 = 兵ESB1 + DB1 = s + 共b − f兲 = b − f + s其 = b − f + s 共15兲 This actually denotes that the initial portion of A 共i.e., A1, the
再 冎
duration of which is equal to the SS lag time s兲 has zero TF
and is critical; while the remaining portion of A 共i.e., A2兲 has
EFA2 = a
Activity B2: ESB2 = max = max共a,b − f + s兲 a TF of T − a − f. Thus, A2 can be critical or noncritical, de-
EFB1 = b − f + s pending on the value of T − a − f being equal to zero or not.
共16兲 2. The start float on B is T − b − s and the finish float on B is 0.
This actually denotes that the final portion of B 共i.e., B2, the
EFB2 = 兵ESB2 + DB2 = max共a,b − f + s兲 + f其 = max共a + f, b + s兲 duration of which is equal to the FF lag time f兲 has zero TF
and is critical; while the remaining portion of B 共i.e., B1兲 has
共17兲
a TF of T − b − s. Thus, B1 can be critical or noncritical, de-
Total project duration pending on the value of T − b − s being equal to zero or not.
A1 A2
(s) (a - s)
Downloaded from ascelibrary.org by New York University on 05/15/15. Copyright ASCE. For personal use only; all rights reserved.
再 冎
A2 is s 共SF lag time兲 and a − s 共duration of whole Activity A
minus SF lag time s兲, respectively. Similarly, Activities B1 and EFA1 = s
Activity B2: ESB2 = max = max共b, s兲 共42兲
B2 in the AON are two components making up the original Ac- EFB1 = b
tivity B in the PDM network. It is noteworthy that to represent the
SF logic, Activity B2 in the AON is actually a dummy, which has EFB2 = 兵ESB2 + DB2 = max共b,s兲 + 0其 = max共b, s兲 共43兲
zero duration and denotes the finish milestone for Activity B.
Total project duration
再 冎
Thus, the duration of Activity B1 in the AON network is the same
as that of Activity B in the PDM network 共i.e., b兲. To simplify EFA2 = a
mathematical representations, we assume the project start time is T = max = max共a, b, s兲 共44兲
zero; a relative origin on the timeline is aligned with early start EFB2 = max共b, s兲
events on A and B. The equivalency of the two networks in Fig. 5
共PDM versus AON兲 in terms of CPM analysis can be analytically Backward Pass Calculation
proven as will be shown next. For jobs with no successor relationships, LF = T
Activity B2: LFB2 = 兵total project duration = T其 = T 共45兲
Critical Path Analysis on Precedence Diagram Method
Network
LSB2 = 兵LFB2 − DB2 = T − 0其 = T 共46兲
Forward Pass Calculation
Activity B1: LFB1 = 兵LSB2 = T其 = T 共47兲
For activities with no predecessor relationships, ES = 0
Activity A: ESA = 兵project start time = 0其 = 0 共27兲 LSB1 = 兵LFB1 − DB1 = T − b其 = T − b 共48兲
EFA = 兵ESA + DA = 0 + a = a其 = a 共28兲 Activity A2: LFA2 = 兵total project duration = T其 = T 共49兲
Activity B: ESB = 兵project start time = 0其 = 0 共29兲 LSA2 = 兵LFA2 − DA2 = T − 共a − s兲其 = T − a + s 共50兲
共51兲
Total project duration
T = max 再 EFA = a
EFB = max共b, s兲
冎 = max共a, b, s兲 共31兲
LSA1 = 兵LFA1 − DA1 = min共T − a + s,T兲 − s其 = min共T − a, T − s兲
共52兲
Through the above analysis, the transformed AON model Application on Case Study
Downloaded from ascelibrary.org by New York University on 05/15/15. Copyright ASCE. For personal use only; all rights reserved.
yields the same CPM results as the original PDM model, such as
identical total project duration 共T兲, activity start times and finish Application of the analytical transformation schemes as derived
times. In Table 3, we further contrast the TFs, which are deter- above leads to automatic production of the equivalent AON net-
mined from early or late activity times as derived from the for- work with straightforward FS logical relationships only, as given
ward and backward analysis on the two networks. in Fig. 6. Also shown in Fig. 6 are activity times and floats cal-
The following interpretations of PDM results are made by re- culated by classic CPM. In this case study, additional relation-
lating to the results from the equivalent AON counterpart: ships are inserted to the basic networks for case “SS/FF
1. The start float on A is min共T − a , T − s兲 and the finish float on staggered” and case “SF” 共i.e., as seen in Fig. 2, an additional FS
A is T − a. This actually denotes that the initial portion of A relationship is linked to B’s start in the case of SS/FF staggered;
共i.e., A1, the duration of which is equal to the SF lag time s兲 additional FS relationships enter the start ends of D and E, respec-
has a TF of min共T − a , T − s兲; while the remaining portion of tively in the “SF” case兲. Nonetheless, the logical equivalency of
A 共i.e., A2兲 has a TF of T − a. Because min共T − a , T − s兲 is no the basic networks still holds. In general, the basic network trans-
greater than T − a, thus, A1 is more critical than 共or at least as formation can be dealt with at a local level and then be plugged
critical as兲 A2. In other words, if A2 is critical 共i.e., T − a into the project at a global level.
= 0兲, then A1 is also critical and the whole activity A is criti- The critical path in the original PDM network were identified
cal. from previous P3 analysis as A-B’s start-SSBC-C-E 共12 days in
2. The start float on B is T − b and the finish float on B is total project duration兲, which now translates to A-B1-C1-C2-
E1-E2 in the equivalent AON network 共also 12 days in total
min共T − b , T − s兲. With reference to the counterpart AON
project duration shown Fig. 6兲. As to the SS/FF staggered sce-
network, B1 is equivalent to B, as B2 is a dummy denoting
nario between B and C, the implication of B’s start being critical
the finish milestone on B. Thus, Activity B has a TF of T
and B’s finish having 2-day TF is clarified: the initial portion of B
− b. Thanks to the SF logical constraint imposed between A
共i.e., B1, whose duration is equal to the SS lag of 1 day has zero
and B, B’s finish event needs to be intentionally delayed,
TF and stays on the critical path, while the remaining portion on
thus possibly resulting in a lesser TF on B’s finish given that B 共i.e., B2, whose duration is 2 days, or B’s duration of 3d minus
T − s is less than T − b. B1’s duration of 1d兲 has 2-day TF and is off the critical path. As
It is noteworthy in Fig. 5 when A’s duration 共a兲 is less than the to the SF logic binding E and D, the fact that D has 4d TF on its
lag time of SF 共s兲, the duration of A2 in the converted AON 共i.e., start and 2-day TF on its end can be explained as follows: Activity
a − s兲 turns negative. However, this does not overthrow the valid- D has 2-day less TF on its finish is due to an mandatory 2-day
ity of the transform scheme, nor brings about confusions on criti- delay on its early finish time imposed by the SF logic.
cal path results from both PDM and AON networks. For instance, In short, the transform schemes being proposed can be auto-
all the floats are actually independent of the difference of a − s 共as matically applied to aid the interpretation of critical path results
seen from Table 3兲. When a − s is zero or negative, A2 in the AON for the PDM network featuring SS/FF staggered or SF relation-
can be taken as a dummy inserted for maintaining particular logic. ships between activities. To make the proposed approach more
Next, we apply the above two transform schemes on our case generic so as to address various scenarios for applying non-FS
study so as to shed light on the critical path analysis results from logical relationships on construction projects, we generalize a
the PDM example network. total of six transform schemes that are likely to be encountered in
0 3 3 3 1 4 4 2 6 6 2 8 10 0 10
SS A B1 B2 D1 D2
0 0 3 3 0 4 6 2 8 10 4 12 12 2 12
FF
ES Dur EF 4 4 8 8 1 9 9 1 10 10 2 12
Task Name C1 C2 E1 E2
LS TF LF 4 0 8 8 0 9 9 0 10 10 0 12
planning construction projects 共Fig. 7.兲 Similar to the above Moreover, we provided analytical proofs for the transform
SS/FF staggered scheme and “SF” scheme, we analytically schemes being proposed to justify the logical equivalency be-
proved the equivalency of the other four transform schemes, tween the original PDM networks and the converted AON net-
namely, 共1兲 “SS” with lag, 共2兲 “FS” with lag, 共3兲 “FF” with lag, works only having FS. As illustrated with a P3 application on a
and 共4兲 SS/FF enveloped. The proofs along with the contrast of PDM network example featuring SS, FF, and SF, confusions
TF results between the original PDM network and the equivalent would arise in interpreting P3‘s results from the PDM network,
AON network are presented in Appendix 1. but not in the case of its AON counterpart.
In conclusion, the proposed transform schemes lead to better
understanding of the scheduling results when critical path analy-
Conclusions sis is performed on a PDM network. This also paves the way for
performing further sophisticated scheduling analysis on a PDM
In forming an AON network diagram for project scheduling, con- network, as the transformed AON network provides a nonambigu-
struction professionals tend to prefer the use of non-FS logical ous straightforward foundation to take on the addition of more
relationships with lags 共1兲 to represent the partially overlapping complex constraints to project scheduling analysis. Examples are
logic between activities or 共2兲 to model simple repetitive projects 共1兲 resource requirements can be specified for each activity in the
comprising a small number of construction units. Critical path resulting AON network for conducting resource leveling analysis;
analysis on the resulting project network is generally referred to 共2兲 by defining pessimistic, optimistic, and most likely times for
as PDM. The “smart” precedence relationships of PDM com- each activity on the resulting AON network, the PERT—coupled
pound TF determination and interpretation, potentially causing with Monte Carlo simulation—can be applied so as to assess
anomalous effects on critical path identification. One straightfor- variability in project duration.
ward solution is to transform non-FS relationships with lags into
FS with zero lag at the expense of adding extra activities prior to
Acknowledgments
conducting forward and backward pass calculations. However,
such relationship transformations are dealt with on case-specific
This research was funded by a Niche Area Research Grant of the
basis in the literature; generic transform schemes are yet to be
Hong Kong Polytechnic University 共A/C No. BB89兲.
formalized.
In the present research, we generalized those particular cir-
cumstances that entail applying non-FS logical relationships on Appendix I
construction projects. We then proposed generic transform
schemes such that non-FS relationships in a project network can The details of PDM-to-AON transformation of four cases 共i.e.,
be detected and transformed—automatically—into equivalent FS. SS, FS, FF, SS/FF enveloped兲 are shown as follows.
Forward Pass Calculation. For activities with no predecessor Activity B: ESB1 = 兵EFA1 = s其 = s 共66兲
relationships, ES = 0
Activity A: ESA = 兵project start time = 0其 = 0 共53兲 EFB = 兵ESB + DB = s + b其 = b + s 共67兲
Total project duration
EFA = 兵ESA + DA = 0 + a = a其 = a 共54兲
再 冎
relationships: LF = T
EFA = a
T = max = max共a, b + s兲 共57兲 Activity B: LFB = 兵total project duration = T其 = T 共69兲
EFB = b + s
Activity A: LFA = 兵total project duration = T其 = T 共60兲 Activity A1: LFA1 = min 再 LSA2 = T − a + s
冎
再 冎
LSB = T − b
LFA − DA = T − a = min共T − a + s, T − b兲 共73兲
LSA = min = min共T − a, T − b − s兲
LSB − SSAB = T − b − s
共61兲 LSA1 = 兵LFA1 − DA1 = min共T − a + s,T − b兲 − s其
= min共T − a, T − b − s兲 共74兲
Critical Path Analysis on Activity-On-Node Network
Table 5. Float Times of PDM Network with FS Relationship and Equivalent AON Network
Network Activity Early start Late start Early finish Late finish TF
PDM A 0 0 a a 0
B a+ f a+ f a+b+ f T=a+b+ f 0
AON A 0 0 a a 0
L a a a+ f a+ f 0
B a+ f a+ f a+b+ f T=a+b+ f 0
Activity B1: ESB1 = 兵project start time = 0其 = 0 共108兲 EFB = 兵ESB + DB = s + b其 = b + s 共122兲
Total project duration
Downloaded from ascelibrary.org by New York University on 05/15/15. Copyright ASCE. For personal use only; all rights reserved.
再 冎
EFB = 兵ESB + DB = max共a, b − f兲 + f其 = max共a + f, b兲
LFA − DA = T − a
共111兲 LSA = min
LSB − SSAB = T − b − f − s
Total project duration
= min共T − a, T − b − f − s兲 共125兲
T = 兵EFB2其 = max共a + f, b兲 共112兲
Activity B: LFB = 兵LFA − FFBA = T − f其 = T − f 共126兲
Backward Pass Calculation. For activities with no successor
LSB = 兵LFB − DB = 共T − f兲 − b其 = T − b − f 共127兲
relationships, LF = T
Activity B2: LFB2 = 兵total project duration = T其 = T 共113兲 Critical Path Analysis on Activity-on-Node Network
LSB2 = 兵LFB2 − DB2 = T − f其 = T − f 共114兲 Forward Pass Calculation. For activities with no predecessor
relationships, ES = 0
Activity B1: LFB1 = 兵LSB1 = T − f其 = T − f 共115兲 Activity A1: ESA1 = 兵project start time = 0其 = 0 共128兲
Forward Pass Calculation. For activities with no predecessor Activity B: ESB = 兵EFA1 = s其 = s 共134兲
relationships, ES = 0
EFB = 兵ESB + DB = s + b其 = b + s 共135兲
Activity A: ESA = 兵project start time = 0其 = 0 共119兲
Total project duration