Professional Documents
Culture Documents
Metode Penjadwalan Produktivitas Analisis Jadwal Linier Dengan Fungsi Singularitas
Metode Penjadwalan Produktivitas Analisis Jadwal Linier Dengan Fungsi Singularitas
Abstract: This paper describes a new integrated method of linear schedule analysis using singularity functions. These functions have
previously been used for structural analysis and are newly applied to scheduling. Linear schedules combine information on time and
amount of work for each activity. A general model is presented with which activities and their buffers can be mathematically described in
detail. The algorithm of the new method forms the body of the paper, including the steps of setting up initial equations, calculating
Downloaded from ascelibrary.org by Remote User on 07/24/21. Copyright ASCE. For personal use only; all rights reserved.
pairwise differences between them, differentiating these to obtain the location of any minima, and deriving the final equations. The
algorithm consolidates the linear schedule under consideration of all constraints and, thus, automatically generates the minimum overall
project duration. The model distinguishes time and amount buffers, which bears implications for the definition and derivation of the
critical path. Future research work will address float and resource analysis using the new model.
DOI: 10.1061/共ASCE兲0733-9364共2009兲135:4共246兲
CE Database subject headings: Scheduling; Critical path method; Network analysis; Geometry; Time dependence; Productivity.
具x − a典n = 再 0 for x ⬍ a
共x − a兲n for x 艌 a
冎 共1兲 General Model for Singularity Functions
where x = variable under consideration; a = upper boundary of The following section introduces the terminology and general
the current segment; and the exponent n = order of the phenom- model for applying singularity function to linear schedules of
enon that changes at the end of the segment. The exponential rule construction projects. This writer proposes to name the new
a0 = 1 applies to the brackets. Eqs. 共2兲 and 共3兲 describe how the method the productivity scheduling method 共PSM兲 to reflect its
brackets can be differentiated and integrated like regular math- intuitive way of combining time and amount information via the
ematical functions measure that links them, productivity. Eq. 共4兲 provides the general
model for modeling linear and repetitive activities and their buff-
d
具x − a典n = n · 具x − a典n−1 共2兲 ers with the Föppl-Macaulay notation, which is shown in Fig. 1
dx
冕 具x − a典ndx =
1
n+1
· 具x − a典n+1 + C 共3兲 y共x兲 = y 0 · 具x − 0典0 +
y1 − y0
x1 − x0
· 具x − 0典1
冋冉 冊 册
m−1
where C = integration constant. Beyond Eqs. 共1兲–共3兲, the follow- y k+1 − y k y k − y k−1
ing rules apply: parts of two singularity functions can be added or
+ 兺
k=1
−
xk+1 − xk xk − xk−1
· 具x − xk典1 共4兲
subtracted if the cutoff a and exponent n are identical. A singu-
larity function can be multiplied by any factor s to scale it. An where y = time variable of an activity with m segments;
exponent of n = 0 indicates a constant phenomenon where x = amount variable x; y 0 = intercept; and y k and xk = pairs of coor-
s⫽intercept and n = 1 indicates a linear phenomenon where dinates with the numbering index k. The summation term contains
s⫽slope. change terms where the present slope y k / xk is replaced with a new
slope y k+1 / xk+1. The buffer of an activity shall be defined as a
distance across time or amount that has to remain free of any
Advantages of Singularity Functions successor activities. It can theoretically take on any shape, but
often is a constant and, thus, has the same shape as the activity to
Singularity functions are a family of functions with several math- which it is attributed. A pair of activities, thus, is typically related
ematical properties that are desirable for the description and via a buffer between them. Such continuous buffers are a gener-
analysis of linear schedules: alization of the previous use where distances were only checked
• They describe the phenomenon of interest based on geometry; at isolated points between neighboring activities 共Harris and Io-
• They separate the components of the phenomenon of interest; annou 1998兲. Those buffers were called horizontal and vertical
• They capture any changes in progress across time and amount; logic constraints 共Thabet and Beliveau 1994兲, least distances
• They can include infinitely many segments of different be- 共Harmelink and Rowings 1998兲, or time and space dependencies
havior; 共Arditi et al. 2002兲. Amount buffers were also called stage buffers
• They are continuous and their value is defined for all 共Reda 1990兲 or location buffers 共Mubarak 2005兲.
arguments;
• They can be scaled with any factor and are independent of
units; Algorithm of Productivity Scheduling Method
• They can be added or subtracted for identical orders and
cutoffs; The following sections detail each step of the algorithm for the
• They can be differentiated and integrated like regular func- new method as shown in the flowchart of Fig. 2. The paper then
tions; and describes how the equivalent of a critical path can be derived
• They can be evaluated manually or via computer software. from the mathematical analysis.
to find locations of minimum values, indicating 1998兲. Activities C, E, and F each consist of two segments with
critical points (activities will touch via buffers) changes of productivities between them, from lower to higher for
C1 to C2 and from higher to lower for E1 to E2 and for F1 to F2.
Step 5: Final Activity and Buffer Equations The exact parameters are listed in Table 1, with the time distance
Follow precedence; subtract minimum values of DT in days, the amount distance DA in length units, and the time
differences from activity and buffer equations and amount buffers BT and BA. Compare this schedule with the
(rewrite intercepts) in consolidated configuration
simple network diagram in Fig. 4. The richness of the two-
dimensional information in linear schedules will allow for a
Fig. 2. Flowchart of analysis steps
deeper analysis than what is possible under CPM.
Step 1: Capture Schedule Data Step 2: Initial Activity and Buffer Equations
Fig. 3 shows a linear schedule that was analyzed under research Activities A through F are described in Föppl-Macaulay notation
for the Iowa Department of Transportation 共Harmelink and Row- in Eqs. 共5兲–共10兲. For clarity, the brackets are sorted from left to
ings 1998兲. Since time can only progress forward, the algorithm right by ascending segment boundaries and within that by ascend-
requires that it is considered the dependent variable y. To reflect ing exponents. The initial equations are generated by following
this mathematical convention, the amount is plotted on the x-axis the order of precedence and by inserting the buffers of Table 1
and time on the y-axis. This causes the slope to represent the between neighboring activities. The maximum value of y共x兲 of all
inverse of the productivity P, which is defined as amount over predecessors is used as the intercept of the successor. This initial
time. However, the mathematical model is separate from the configuration is on the safe side, as it keeps the successor high
graphical representation. If desired, the time axis can be drawn above its predecessors without any potential interference. It is
along the horizontal axis. Numeric values for time and amount analogous to only permitting finish-to-start relationships in CPM.
and their buffers were not provided in the original source and Fig. 5 shows this stacking of activities. Buffers are shown as gray
have been derived from the diagram in the original source shaded areas in the figure.
共Harmelink and Rowings 1998兲. Activities A – F are performed in Eqs. 共5兲 – 共10兲 = singularity functions for the activities under
a sequential order of precedence 兵A , B , C , D , E , F其. All activities consideration of their time buffers, which act in the vertical di-
rection. The start and finish dates of the equations are listed in
Table 2. The time buffer equations are omitted for brevity, but can
y [time] be recreated by adding their BT to the intercept of their activity
22 F
equation. The equivalent activity equations under consideration of
F2 the amount buffers are also omitted, but their respective start and
20 finish dates are listed in Table 3. Activity F is the last activity and
18 F1 E does not carry any buffers. The simple case of a constant buffer as
E2 per Table 1, which is added to the intercept of its activity equa-
16
D tion, can be extended to the general case of a buffer equation
14 E1
whose shape is entirely independent from its activity. Step 2
12 C would then alternate between activity and buffer equations
C2
10
7
8 B y共x兲A = 0 · 具x − 0典0 + · 具x − 0典1 共5兲
C1 A 50
6
4 4
y共x兲B = 8 · 具x − 0典0 + · 具x − 0典1 共6兲
2 50
0
10 20 30 40 50
x [amount] A B C D E F
Fig. 3. Linear schedule 共Harmelink and Rowings 1998兲 Fig. 4. Network diagram
22 Amount buffers
Critical Time
20 C Name Segment ST FT points distance
C2 A A1 0 7 75/ 2 = 37.50 3.0
18
B B1 7 11 0.00 3.0
16
C1 C C1 11 17 150/ 7 = 21.43 6.3
14
C2 17 18 150/ 7 = 21.43 6.3
12 B D D1 18 25 30.00 2.0
10 E E1 25 26 80/ 7 = 26.67 2.0
8 E2 26 30 80/ 7 = 26.67 2.0
A F F1 30 33 — —
6
F2 33 36 — —
4
Downloaded from ascelibrary.org by Remote User on 07/24/21. Copyright ASCE. For personal use only; all rights reserved.
2
value of x to describe where the plateau of the amount buffer
0
10 20 30 40 50 begins. Similar equations for the remaining amount buffers are
omitted for brevity
冓 冉 冊冔 冓 冉 冊冔
x [amount]
0 1
BA 50 7 50
Fig. 5. Activity stacking with time buffers y共x兲bufA =0· x− − + · x− −
4 50 4
y共x兲C = 13 · 具x − 0典 +0
6
· 具x − 0典1 −
6
−
1
冉· 具x − 35典1冊
−
7
冓 冉 冊冔
50
· x − 50 −
50
4
1
共11兲
共9兲 4 4
BA
y共x兲bufB = 8 · 具x − 0典0 + · 具x − 0典1 − · 具x − 37.5典1 共13兲
冉 冊
50 50
冓 冔
3 3 3
y共x兲F = 38.7 · 具x − 0典0 + · 具x − 0典1 − − · 具x − 40典1 1
40 40 10 653 6 11 150
BA
y共x兲bufC = · 具x − 0典0 + · 具x − 0典1 − · x−
共10兲 49 35 105 7
冓 冔
Amount buffers act in the horizontal direction, which is equiva-
lent to shifting the singularity functions for their activities side- 7 7 260 1
ways by the value BA. Eq. 共11兲, for example, includes the shift of
BA
y共x兲bufD = 19.8 · 具x − 0典0 + · 具x − 0典1 − · x−
50 50 7
⌬x = 12.5 in the xk values and by deducting it from the maximum
共15兲
232 1
Table 2. Initial Configuration for Time Buffers BA
y共x兲bufE = · 具x − 0典0 + · 具x − 0典1
9 30
冓 冔 冓 冔
Time buffers
Critical Time 1 1
1 20 1 80
Name Segment ST FT points distance + · x− − · x− 共16兲
6 3 5 3
A A1 0.0 7.0 50 4.0
B B1 8.0 12.0 0 4.0
C C1 13.0 19.0 35 5.9 Step 3: Differences of Activities and Buffers
C2 19.0 20.0 35 5.9
D D1 21.9 28.9 30 3.8 Pairwise differences are calculated between the time and/or
E E1 30.7 31.7 0 5.0
amount buffers of all predecessors and their successor activity
while following the order of precedence. In other words, these
E2 31.7 35.7 40 5.0
difference equations describe the white spaces in Fig. 5. If an
F F1 38.7 41.7 — —
activity has several predecessors, the minimum of these differ-
F2 41.7 44.7 — —
ences is used. The rule applies that the terms of two singularity
冉 6 4
冊 11 38 1
Downloaded from ascelibrary.org by Remote User on 07/24/21. Copyright ASCE. For personal use only; all rights reserved.
BT
y共x兲C−bufB = 5 · 具x − 0典0 + − · 具x − 0典1 − · 具x − 35典1
BA
y共x兲F−bufE = · 具x − 0典0 + · 具x − 0典1
9 24
冓 冔 冓 冔
35 50 105
1 1
共18兲 1 20 1 80 9
− · x− + · x− + · 具x − 40典1
6 3 5 3 40
BT
y共x兲D−bufC = 8.9 · 具x − 0典0 + 冉 7
−
6
50 35
冊
· 具x − 0典1 +
11
105
· 具x − 35典1
共26兲
冉 冊
ference changes its sign from negative to positive. Depending
3 1 1 on whether the neighboring activities overall are diverging or
BT
y共x兲F−bufE = 8 · 具x − 0典0 + − · 具x − 0典1 − · 具x − 30典1
40 30 6 converging 共Harris and Ioannou 1998兲, the boundaries x = 0 or
9 x = xmax are also locations of potential minima. Such critical points
+ · 具x − 40典1 共21兲 where activities touch via their buffers were also called vertices
40 共Harmelink and Rowings 1998兲. The x-coordinates of the critical
points are listed in Tables 2 and 3 along with the distances ⌬y
Amount Buffers between the buffers of the respective activity segments and the
successor activity. Note that the time buffer of activity E is closest
to activity F in two locations, for x = 0 and for x = 40
3 7
BA
y共x兲B−bufA = 5.25 · 具x − 0典0 − · 具x − 0典1 + · 具x − 37.5典1
50 50 3
⬘BT = 0 −
y共x兲B−bufA · 具x − 0典0 共27兲
共22兲 50
3 7
16 11 ⬘BA = 0 −
y共x兲B−bufA · 具x − 0典0 + · 具x − 37.5典0 共28兲
BA
y共x兲C−bufB = 3 · 具x − 0典0 + · 具x − 0典1 − · 具x − 35典1 50 50
175 105
4
+ · 具x − 37.5典1 共23兲 Step 5: Final Activity and Buffer Equations
50
All activity and buffer equations are consolidated to their earliest
冓 冔
possible configuration by deducting the differences of Tables 2
1 and 3 from their intercepts cumulatively while following the order
229 11 11 150
BA
y共x兲D−bufC = · 具x − 0典0 − · 具x − 0典1 + · x− of precedence. Activity A is the first activity and does not need to
49 350 105 7
冓 冔
be rewritten. Shifting the singularity functions of Eqs. 共6兲–共10兲
1
1 255 and their respective time or amount buffers downward by the sum
+ · x− 共24兲
15 7 of ⌬y to yield the final Eqs. 共29兲–共33兲 accomplishes the objective
of minimizing the overall project duration 共makespan兲. The final
time and amount buffer equations are omitted for brevity, but can
8 be recreated as described above. Evaluating these equations yields
BA
y共x兲E−bufD = 5.2 · 具x − 0典0 − · 具x − 0典1
75 the start and finish points listed in Table 4
+
1
6
· 具x − 30典1 +
7
50
· x− 冓 冔
260
7
1
共25兲 y共x兲B = 4 · 具x − 0典0 +
4
50
· 具x − 0典1 共29兲
22 F 22 F
F2 F2
20 20
18 F1 E 18 F1 E
E2 E2
16 16
D D
14 E1 14 E1
12 C 12 C
C2 C2
10 10
8 B 8 B
C1 A C1 A
6 6
4 4
Downloaded from ascelibrary.org by Remote User on 07/24/21. Copyright ASCE. For personal use only; all rights reserved.
2 2
0 0
10 20 30 40 50 10 20 30 40 50
x [amount] x [amount]
6 11 The time and amount critical paths thus obtained are shown in
y共x兲C = 5 · 具x − 0典0 + · 具x − 0典1 − · 具x − 35典1 共30兲 Figs. 6 and 7 as thick lines in the linear schedule. Critical points
35 105
are marked with small circles. Several interesting observations
can be made. Examining the figures shows that critical points may
7 occur at starts, changes, or finishes of activities. For example, in
y共x兲D = 8 · 具x − 0典0 + · 具x − 0典1 共31兲
50 Fig. 6, the change in activity C at x = 35 induces a time critical
point with its neighboring activity D. There is no equivalent re-
lationship for such “middle-to-middle” link in CPM. Moreover,
1 1
y共x兲E = 13 · 具x − 0典0 + · 具x − 0典1 + · 具x − 30典1 共32兲 the previously observed two closest locations at x = 0 and x = 40
30 6 causes the time critical path to split and merge again between the
two activities E and F already. Activities A and B are connected
3 9 by a finish-to-finish link whereas activities B and C are connected
y共x兲F = 16 · 具x − 0典0 + · 具x − 0典1 + · 具x − 40典1 共33兲 by a start-to-start link. Activity B is, therefore, fully time critical,
40 40
same as activities A and F. Activities C, D, and E are only par-
The mathematical analysis as formulated under this algorithm tially time critical across segments whose boundaries are induced
is independent of the graphical representation of the linear sched- by time critical points. The amount critical path shown in Fig. 7 is
ule, whether the x-axis is drawn horizontally or vertically. Essen- significantly different. It does not split at all. Only activity A is
tially, the previous application of the algorithm has analyzed two fully amount critical, all others are partially amount critical. The
different schedules, which happened to yield the same final con- boundaries at which they change from being noncritical to critical
figuration of activities. A mixture of time and amount buffers and vice versa in Fig. 7 are completely different from Fig. 6. It is,
inside a single project schedule can be processed by the algorithm therefore, prudent to distinguish time and amount criticality for
in the same manner, as both types of buffers are modeled with activities in linear schedules.
singularity functions. If an activity has both types of buffers, a
case distinction is needed to ensure that all constraints are ful-
filled. In this case, Steps 2 through 5 shall use whichever buffer Comparison with Previous Solution
yields the maximum intercept for a successor activity. The amount criticality of activities A and B in Fig. 7 also differs
from the results from the graphical analysis by Harmelink and
Rowings 共1998兲, which found activity A to be amount critical
Time and Amount Critical Paths from its start point to x = 200/ 7 ⬇ 28.57 and activity B to be
amount critical from its start point to x = 12.5. This inconsistency
With the activities and their buffer in their final configuration, it is was noted independently by Kallantzis and Lambropoulos 共2004兲.
now possible to construct the equivalent of a critical path under It was caused by considering the start of an activity to be critical
consideration of the scenario with time buffers as per Table 2 and if no link with any predecessor had otherwise been established in
the scenario with amount buffers as per Table 3. The following the downward pass 共Harmelink and Rowings 1998兲 and by check-
rules are used for connecting the critical locations to obtain the ing distances only at discrete critical points, not across a continu-
critical path for the scenarios with time or amount buffers: ous equation. These least time and distance intervals were discrete
• It is continuous from earliest start point to latest finish point; themselves, but no equivalent of a time critical path was derived.
• It follows the order of precedence and may split or merge; In a subsequent paper, time distances were checked during the
• It may include complete activities or segments thereof; graphical analysis of an amount critical path 共Harmelink 2001兲. In
• It jumps parallel to the time axis across time buffers; and comparison, time and amount buffers are clearly distinguished in
• It jumps parallel to the amount axis across amount buffers. PSM and their equations are continuous. The algorithm of PSM
There is no separate forward and backward pass as under CPM The support of the National Science Foundation 共Grant No.
共or upward and downward pass, respectively兲, to determine the CMMI-0654318兲 for portions of the work presented here is grate-
criticality of activities, but an initial configuration from stacking fully acknowledged. Any opinions, findings, and conclusions or
activities and their buffers and consolidating them to the final recommendations expressed in this material are those of the
configuration by using minimum differences. CPM requires spe- writer and do not necessarily reflect the views of the National
cific relationships to exist as input for its algorithm, at the least Science Foundation.
finish-to-start links. If predecessor and successor activities shall
overlap across time to minimize the overall project duration, other
Downloaded from ascelibrary.org by Remote User on 07/24/21. Copyright ASCE. For personal use only; all rights reserved.
Lucko, G. 共2007兲. “Flexible modeling of linear schedules for integrated Stradal, O. and Cacha, J. 共1982兲. “Time space scheduling method.” J.
mathematical analysis.” Proc., 2007 Winter Simulation Conf., IEEE, Constr. Div. 108共3兲, 445–457.
Piscataway, N.J., 2159–2167. Thabet, W. Y., and Beliveau, Y. J. 共1994兲. “HVLS: Horizontal and verti-
Macaulay, W. H. 共1919兲. “Note on the deflection of beams.” Messenger cal logic scheduling for multistory projects.” J. Constr. Eng. Manage.,
of Math., 48共9兲, 129–130. 120共4兲, 875–892.
Mattila, K. G., and Abraham, D. M. 共1998兲. “Linear scheduling: Past Wittrick, W. H. 共1965兲. “A generalization of Macaulay’s method with
research efforts and future directions.” Eng., Constr. Archit. Manage., applications in structural mechanics.” AIAA J., 3共2兲, 326–330.