Vous êtes sur la page 1sur 12

Cut order planning for apparel manufacturing

CHARLOTTE JACOBS-BLECHA
1
, JANE C. AMMONS
2
, AVRIL SCHUTTE
3
, and TERRI SMITH
4
1
School of Civil and Environmental Engineering, Georgia Institute of Technology, Atlanta, GA 30332, USA
2
School of Industrial and Systems Engineering, Georgia Institute of Technology, Atlanta, GA 30332, USA
3
Cummins Engine Company, Inc., Columbus, IN 47201, USA
4
MCI, Roswell, GA, USA
Received February 1993 and accepted November 1996
Cut order planning is the problem of planning the fabric cut for a set of apparel orders. A specied set of garments is ordered by
the customer to be produced by a given date. Aliated with this order is a set of garment sizes, and patterns for cutting each size.
The fabric for cutting the order must be arranged on the cutting table in such a way as to minimize the cost of cutting the order.
This plan must incorporate a partition of the pattern pieces to be arranged on top of the fabric layout. A mathematical model of
the problem is developed and analyzed. On the basis of this analysis, solution approaches are developed that have been imple-
mented on a desktop PC-based computer. Validated on representative industry problems, the approach is shown to be eective and
versatile.
1. Introduction
Like many domestic manufacturers competing in today's
international marketplace, the apparel industry has been
forced to upgrade its responsiveness to customer needs.
As a result, smaller orders are placed in a more dynamic
fashion, requiring the ecient production of smaller lot
sizes. Eective and economical production thus depends
upon the interaction of many system components, one
of the most critical being an ecient workow control
system.
Cut order planning (COP) is an important linkage in
the workow control system. As illustrated in Fig. 1,
COP is one of the initial stages in the introduction or
release of work-in-process (WIP) into the assembly sys-
tem. COP occurs for each order to be produced, and is
the starting point in the manufacture of the order. COP is
the activity of planning the cutting of the order, as input
into the marker making stage (where the layout of pattern
pieces on the fabric is designed) so that the cutting room
receives complete spreading and cutting instructions.
Once the cutting operation has produced bundles of cut
pieces, they are moved through the assembly system in
operation precedence order.
The cut order planning process is a dynamic one. The
function must respond to the ever-changing status of
many critical factors such as sales, inventory levels, raw
materials, and availability of labor and equipment. The
variety of sizes, styles, fabrics, and colors induces signi-
cant complexity into the problem. Adding to the com-
plexity, and thus potentially increasing total production
costs, are setup or changeover costs, the question of
appropriate lot sizes, and the necessity to meet customer
demand competitively.
Current industry approaches for performing COP
range from manual ad hoc procedures to customized
proprietary software. Many apparel manufacturing fa-
cilities are still using very unsophisticated methods, de-
pending on the expertise of one individual who has the
necessary data and decision making tools only in his or
her memory. Prot margins in this industry severely limit
capital investment, resulting in few resources for com-
puting equipment. Commercial software has been devel-
oped to solve COP, but eective application requires
extensive customization and the necessary hardware for
implementation. Therefore, because of industry limita-
tions, simple methodologies that produce `good' solutions
rather than complicated `optimal' ones will have the
highest impact on improving the operation of the cut
order planning process.
We found no previous papers for solving COP in the
literature. In this research we have addressed three major
thrusts, with the nal goal being a reasonable, and easily
implementable, solution method for COP. Our work be-
gan with developing an in-depth understanding of COP
and the examination of existing commercial software
packages for solving it, including a comparative analysis
of their performances. This analysis was performed using
testbed data representative of specic industrial problems.
The second thrust included formulating a mathematical
model of the cut order planning process, and analyzing the
complexity of the problem. On the basis of our under-
0740-817X 1998 ``IIE''
IIE Transactions (1998) 30, 7990
standing of this model and the results from the rst thrust,
several heuristic algorithms for solving COP were devel-
oped in the third phase of the project. These solution
approaches were tested on representative industry prob-
lems, and the results indicate that the algorithms are im-
portant for future COP systems, toward the improvement
of productivity and the competitiveness of apparel man-
ufacturing. Specic details of the various topics presented
throughout this paper can be found in [1].
The paper is organized as follows. In Section 2 the
COP problem is dened and a corresponding mathe-
matical model is developed. Our analysis of the model
provides insight for algorithm development, with solution
approaches described in Section 3. In Section 4 we report
the performance of the algorithms on a representative
problem provided by a manufacturing partner. The paper
concludes with a summary and overall insights detailed in
Section 5.
2. Problem denition
2.1. Problem description
In the conventional scope of cut order planning, the
problem begins with a given set of orders to be cut. The
order is composed of a set of garments to be manufac-
tured, in varying sizes. Each garment will be referred to as
a unit. The various units in the various sizes must be
partitioned in such a way that ecient cutting of the
order can take place. The cut order plan is a specication
for spreading the fabric and assigning the units to various
sections of the spread.
The cut is performed by spreading fabric onto a table.
The spread is divided into sections. Sections contain dif-
ferent combination of sizes to be cut and have various
number of layers, or plies, of fabric for cutting eciency.
Such a spread is illustrated in Fig. 2. Note that using
various size combinations within a section implies one
could cut a single size, a mixture of dierent sizes, or a
mixture of sizes that could include more than one of the
same size. For example, a section could be planned for
cutting two `larges'. This means that the pattern pieces for
two large garments would be needed for cutting that
section. This idea is illustrated in Fig. 3. To determine the
conguration for a fabric spread, the sizes in the order
must be partitioned in such a way that the order is com-
plete and that fabric use is minimized. For a further ex-
ample, an order might consist of 110 pairs of jeans, with
30 smalls (S), 50 mediums (M), and 30 larges (L). One way
of planning the cut would be to spread two sections. In
the rst section we would cut only size M jeans, with ply
height 20. This would yield 20 of the 50 pairs of jeans
ordered. Then we would cut all three sizes, S, M, and L, in
Section 2, with ply height 30. This would yield the 30 S
and 30 L ordered, and the remaining 30 M needed to
complete the order size of 50. Clearly there are many
Fig. 1. Traditional cut order planning.
Fig. 2. Fabric spread on the cutting table in varying ply heights.
80 Jacobs-Blecha et al.
other congurations that could be used. One must choose
the solution yielding the least costly cutting process.
Note that we are not concerned here with designing the
exact layout of the pattern pieces for the cutting of the
jeans. If we plan to cut only `mediums' in Section 1, we
need to know approximately how much fabric is required
to do that, but traditionally that quantity is often esti-
mated. If such a layout has been done before, historical
data can be substituted for estimates. The actual layout of
the pattern pieces that is used to guide the cutter is called
a marker, and the marker-making process is a variation of
the well-known stock cutting problem (e.g., [27]). In
current practice, the marker-making process is performed
independently of and subsequent to the COP process.
The spread length of a cut is the total length of all
sections as they are laid out end-to-end on the cutting
table. Note that in the example in Fig. 2 the spread length
is equivalent to the length of the cutting table, which
serves as a physical upper bound on the many possible
values for spread length. The perimeter length of a pattern
piece is the total number of inches in the perimeter of that
piece. The input to the COP problem consists of:
v the sizes required for the order,
v the quantity of each size to be cut,
v the total perimeter inches of each pattern piece re-
quired for the cut,
v the total area of the pattern pieces required for the
cut, and
v the standards for spreading (marker xed costs,
marker variable costs, cost to copy, minimum and
maximum plies, number of sizes per marker, cutting
costs, cutting speed, and cutting setup).
The output from the COP process then consists of the
following:
v the sizes to be combined in each section of the
marker,
v the estimated eciency of the marker (percent of
fabric utilization),
v the cutting cost per unit (garment),
v the total perimeter to be cut, and
v the total area to be cut.
The objective of the cut order planning problem is to
minimize costs, which introduces a need for a tradeo
between cutting costs and fabric costs. The key decisions
to be made are (1) the number of sections required to ll
the order, (2) ply height in each section, and (3) the sizes
to be cut in each section. The determination of (3) pro-
vides the input to the marker making function for actual
design of the marker itself.
2.2. Mathematical model
2.2.1. Parameters
Consider an order to be cut consisting of sizes
s = 1Y 2Y F F F Y S. The notation d
S
will represent the number
of units of size s required to ll the order. The marker for
the order will contain sections = 1Y 2Y F F F Y J. The com-
bination and multiples of sizes used in a section of the
marker are indexed by i. An example (for sizes small,
medium and large) is partially illustrated in Table 1.
The possible size combinations consist of the single
individual sizes, as well as multiples of the individual
sizes, and combinations of the single and multiple sizes.
The total number of these size combinations is repre-
sented by I, a very large number. For example, for an
order containing six sizes, the number of combinations is
O(10
7
). The magnitude of I can be signicantly reduced
by limiting the number of sizes allowed to be combined in
a section. Other problem parameters are:

i
= estimated fabric length required to cut size
combination i,
or the exact length required if a marker for
combination i already exists;
e
i
= number of cutting inches to cut size combina-
tion i;
M
i
= increased cost of marker making due to size
combination i;
d
si
= number of units of size s in size combination i;
c = fabric cost per length unit (normally yards or
meters);
P = maximum allowable ply height;
L = maximum allowable spread length;
1 = labor cost for time required to spread one length
of the table;
U = cost per perimeter length unit for cutting;
d
S
= number of units production permitted over or
under the total units in the order.
Table 1. Size combination example
i Size combination
1 1 small, 1 medium
2 1 large
3 1 medium, 1 large
4 3 mediums
5 2 smalls, 1 medium
F
F
F
F
F
F
Fig. 3. A marker for a section containing two units of the same
size.
Cut order planning for apparel manufacturing 81
2.2.2. Decision variables
There are two sets of integer decision variables repre-
senting the ply height of a section and the assignment of
sizes to a section:
y

= ply height of section Y y

= 0Y 1Y 2Y F F F Y PX
x
i
=
1 if size combination i is assigned to section ,
0 otherwise.
&
2.2.3. Objective
The objective of the cut order planning problem is to
minimize the total cost of cutting the order, including the
cost of fabric and labor. Specically, these costs are ac-
tual fabric costs, spreading costs, cutting costs, and the
impact on marker making costs.
(a) Fabric cost: The term c
i
is the fabric cost of one layer
of size combination i in any section. Thus, the total fabric
cost over all sections and all size combinations i is

J
=1

I
i =1
c
i
y

x
i
X
(b) Spreading cost: The variable
i
is the length of fabric
required to cut size combination i in any section. Thus,

i
aL is the fraction of the table length needed in spreading
the section for size combination i in any section, and
1(
i
aL) is the labor cost for spreading a one-ply section
containing size combination i. Hence, 1(
i
aL)y

is the
total cost of spreading size combination i in section , and
the objective function term for spreading cost is

J
=1

I
i =1
1

i
L
y

x
i
X
(c) Cutting cost: The term Ue
i
is the cost of cutting size
combination i, and the objective function term is

J
=1

I
i =1
Ue
i
x
i
X
(d) Increased marker making cost: The total increased cost
of marker making is expressed in the objective function
by the term

J
=1

I
i =1
M
i
x
i
X
If the marker already exists, the cost of determining the
marker length is based simply on the minor costs of
storage and retrieval. However, if such a marker does not
exist, the cost increases due to the expanded time needed
for a skilled operator to design the marker and then
transfer the data to the COP function.
The complete objective function can then be expressed as
follows:
Minimize Z =

J
=1

I
i =1
c
i
y

x
i
1

i
L
y

x
i
M
i
x
i
Ue
i
x
i
!
or
Minimize Z =

J
=1

I
i =1
c
i
y

1

i
L
y

M
i
Ue
i
!
x
i
X
2.2.4. Constraints
A demand constraint is required for the order to be lled;
i.e., the total number of units planned for should be equal
to the total number of units ordered. This is expressed as

J
=1

I
i =1
d
si
y

x
i
= d
s
\ sX (A)
Note that if production over or under the specied order
(overages or underages) is allowed, constraint (A) can be
modied accordingly.
Table length constraint: The following constraint re-
stricts the total length of the marker to be less than or
equal to the physical length of the cutting table.

J
=1

I
i =1

i
x
i
_ LX (B)
The marker is made up of all the planned sections in their
various ply heights.
Enforce the upper bound on ply height. The constraint is
represented as
y

_ P \ X (C)
This constraint is needed because both manual and au-
tomatic cutters have limited capacity in the thickness of
fabric that can be cut.
Variable restriction constraints. Decision variable y

is
restricted to be a positive integer less than or equal to the
maximum ply height, and x
i
is restricted to be a binary
variable:
y

0Y 1Y 2Y F F F Y P \ Y (D)
and
x
i
0Y 1 \ iY X
The model as presented above is dicult to solve by
standard mathematical programming methods. Both the
objective function and the constraints in (A) contain
nonlinear terms. The model can be linearized by using a
variable substitution as follows:
Let :
i
= y

x
i
= the number of replicates (layers) of size
combination i that will be cut in section X
It should be clear from the denition of y

and x
i
that :
i
is either 0 or the ply height of section . Implementing this
82 Jacobs-Blecha et al.
substitution results in the following linear integer model
for the Cut Order Planning problem.
(COP)
Minimize Z =

J
=1

I
i =1
c
i
1

i
L
!
:
i
M
i
Ue
i
[ [x
i
(1)
subject to

J
=1

I
i =1
d
si
:
i
d
s
= d
s
\ sY (demand constraint) (2)

J
=1

I
i =1

i
x
i
_ LY (tab!e !enqth constraint) (3)
:
i
_ x
i
P \ iY Y (p!y heiqht restriction constraints) (4)
:
i
0Y 1Y 2Y F F F Y P \ iY Y (tariab!e restriction
constraints) (5)
x
i
0Y 1 \ iY X
In summary, our model states that we will minimize the
cost of cutting a set of orders (1), and in so doing will
meet customer demand (2), will not exceed the table
length constraint (3), and will respect ply height and size
combination restrictions (4). In the next section, this
model is useful in discerning the mathematical complexity
of (COP), and will be useful for algorithmic development
described in Section 3.
2.3. Characteristics and complexity
The model (COP) is very dicult to solve to optimality
when the parameters are of realistic size. Intuitively this
diculty can be explained by pointing out that the
number of solutions grows exponentially as the size of the
problem increases, and therefore the time to search all
feasible solutions for the best one also grows exponen-
tially in the size of the problem. Having modeled COP
with an integer program bearing no exploitable structure
for ready solution provides further evidence of the di-
culty of the problem. In fact, (COP) is NP-complete. The
proof is provided in Appendix A.
3. Heuristics
Because (COP) is NP-complete, ecient algorithms for
realistically sized problems will necessarily be heuristic in
nature. This insight leads to the need for analyzing (COP)
for characteristics that can be exploited for development
of heuristic methods. This section describes the heuristics
developed for (COP), the reasoning behind these types of
algorithm, and justication for the evaluation techniques.
Heuristic development is based on the examination of
typical industry cases that COP cost (equation 1) is
dominated by total fabric length. Section 4.1 explains the
experimental design that we used to establish this char-
acteristic of the cost function. It should be noted that in
some cases the cost factors that we consider in the model
developed in Section 2.2 may have a signicant role in the
cost of cut order planning. For example, spreading costs
may be very high due to negotiated labor rates; cutting
costs may be driven up by manual or equipment pa-
rameters; or a large data base of historical markers may
not exist, greatly increasing the cost of that process.
However, we assume that the statistical results, which
conrm practitioners' intuitions, are valid for the types of
problem addressed by our work, and therefore the model
in Section 2.2 can be modied to reect this assumption.
Note that under this assumption the only change in the
model occurs in the objective function, where all terms go
to zero except those involving the fabric length parame-
ters. An alternative method for problem solution is to
solve the linear relaxation and check the resulting solu-
tion for satisfaction of the integer constraints. However,
this approach is not practicable: for realistically sized
problems the number of variables, :
i
, prohibits explicit
computation. Furthermore, most apparel manufacturers
who would use these solution methods do not have suf-
cient computing capability on site to utilize sophisti-
cated integer programming solvers.
Therefore the development of heuristic algorithms to
solve (COP) focuses on nding computationally ecient
procedures for nding good (i.e., relatively low cost) so-
lutions to (COP) for a robust set of problem instances.
We have selected two types of algorithm for the devel-
opment of such heuristics, constructive and improvement.
A constructive algorithm takes the input data and builds
a feasible solution using intuition, clues from the spatial
aspects of the problem, and guidelines found in the
mathematical model. An improvement algorithm begins
with an existing feasible solution and attempts to change
the solution in some manner so that the cost of the so-
lution is reduced while feasibility is maintained. The value
of the cost function associated with the feasible solution
produced by one of these heuristic methods can then be
compared with some numerical bound, or other bench-
mark solutions.
There are three greedy heuristics presented in this sec-
tion. Two of these algorithms, Savings and Cherry Pick-
ing, are constructive in nature. The Savings heuristic
assigns size combinations to a section on the basis of the
fabric savings achieved by combining them into one sec-
tion as opposed to having them assigned to separate
sections. The Cherry Picking algorithm builds sections by
combining certain sizes based on the best utilization of
fabric. The algorithm picks the rst and second most
numerous sizes in the order and places those in sections
rst, then repeats until all sizes are assigned to a section.
The third heuristic is an improvement heuristic rather
than a constructive one. The Improvement algorithm
takes a current solution and tries to improve it by
exchanging sizes in dierent sections or by combining
Cut order planning for apparel manufacturing 83
existing sections into one section. These three methods
are detailed in Table 2 and the remainder of this section.
All algorithms were coded in ANSI Standard C and
implemented on a DOS-based IBM PC platform. The
results described in Section 4 were obtained with a 386/20
MHz processor, with each solution completed in less than
10 minutes.
3.1. Savings computation
Step 2 of the Savings algorithm and Step 3 of the
Improvement algorithm require a computation of sav-
ings achieved for combining two sections into one.
Described below are the details of this computation,
based on whether or not the two sections to be com-
bined contain the same set of sizes, denoted Case A
and Case B.
Case A: The two sections contain exactly the same size(s).
The merger can be accomplished in one of two ways.
(i) Increase ply height by spreading one section on top
of the other and making no change to the size combina-
tion in the section, as illustrated in Fig. 4. Note that the
length of fabric required for the section is the same before
and after the merger, and hence has no eect on the cost
savings for the merger. Therefore if only fabric cost is
being considered, the savings for this case is equal to zero.
However, if cutting costs must be considered, let e
represent the number of cutting inches in the pattern for
the size combination in the two sections being considered.
Then e is also the number of cutting inches required for
the merged section as well. Recall that U is the cutting
cost per inch. Thus, Ue Ue is the cost of cutting the two
unmerged sections, and Ue is the cost of cutting the
merged sections. Hence, under this consideration, Ue is
the savings in cost obtained by merging the two sections.
(ii) Change the size combination, leaving the ply height
the same, as illustrated in Fig. 5. Here the savings will be
the decreased cost of fabric required for spreading the
merged sections.
Table 2. A summary of COP heuristic algorithms
Step no. Savings heuristic Cherry Picking heuristic Improvement heuristic
0 Begin with no units
assigned to a section.
Begin with no units assigned to a
section.
Begin with all units assigned to a
section.
1 Assign each unit to a
separate section, using the
initial ply height of 1.
Let q1 be the largest quantity of any
size remaining in the order, and q2
be the second largest, where q2 < q1.
If there is no such q2, then set q2 = q1.
Form set S by selecting all sizes
remaining which have a quantity
greater than or equal to q2.
Consider the next portion of one
section.
2 For each pair of sections,
compute a list of savings
(see Section 3.1) achieved by
combining them into a
single section.
The new section will have ply
height = min{q2, max ply height}.
Combine the sizes from set S to
minimize fabric use. All combinations
of the sizes in set S should be
considered which do not exceed
the maximum number of sizes
allowed per section.
Attempt to reassign the portion from
its original section to one or more of
the remaining sections, satisfying
the feasibility checks (see Section 3.2)
listed on the next page. If feasible to
reassign, compute the savings (see
Section 3.1) that would be achieved by
making the reassignment.
3 Feasibly merge sections
based on the maximum
savings achievable.
Reduce the order demand quantities
for the sizes in set S by q2.
Perform the reassignment based on
the best savings computed.
Start again at Step 1.
4 Continue merging other
sections with this one until
it contains the maximum
number of units. This section
is saved and becomes ineligible
for further merges.
If the order contains a size with
positive quantity larger than the
number of units allowed under the
specied demand, go to Step 1.
The iterations must be tracked. In
the current iteration, if there are no
improvements after examining all
possible exchanges, then the
algorithm terminates.
5 After k mergers in Step 3,
update the savings list (k
being an arbitrary input factor).
6 Continue until no more savings
can be achieved or no mergers
are possible.
84 Jacobs-Blecha et al.
Assume the following notation:

i1
= length of fabric required to cut one layer of the
rst unmerged section,

i2
= length of fabric required to cut one layer of the
second unmerged section,

i3
= length of fabric required to cut one layer of the
third merged section, and
p = ply height of the unmerged and merged sections.
Recall that c is the unit cost of fabric.
Then the savings can be computed as cp(
i1

i2

i3
).
Case B: The two sections do not contain the same size(s).
Fig. 6 illustrates this case.
(i) If the ply heights are the same, this case is precisely
the same as case A(ii). We simply merge the size combi-
nations, leaving the ply height unchanged. Then the
savings can be computed as cp(
i1

i2

i3
).
(ii) If the two sections do not contain the same size and
have dierent ply heights, the only way to merge two such
sections is to merge the size combinations, choosing the
ply height so that overages or underages are minimized.
This again is the same as case A(ii). Hence the savings
computation is cp(
i1

i2

i3
). If overages/underages
are not allowed, this option is simply denoted infeasible,
and the saving computation is equal to zero.
3.2. Feasibility checks for the improvement algorithm
The feasibility of such mergers is based on two condi-
tions:
(1) Will the maximum number of sizes allowed per
section be violated? If so, do not merge.
(2) Will the maximum number of units over and under
the demand be violated? If so, do not merge.
4. Computational results
4.1. Testbed data
We developed a set of problems representative of an in-
dustrial application to evaluate the performance of the
heuristic algorithms. These problems were based on a
problem provided by an apparel manufacturer, and were
structured with test cases to examine the robustness of the
algorithms relative to various problem instances. The
testbed data contains 20 problems that range from one to
six sizes per order. The various parameters of the problem
were varied according to the experimental design de-
scribed below. These factors are listed in Table 3. Esti-
mated marker lengths for the various size combinations
were obtained by using a COP commercial software
package.
Fig. 4. Illustration of case A(i).
Fig. 5. Illustration of case A(ii).
Fig. 6. Illustration of case B.
Cut order planning for apparel manufacturing 85
To facilitate the experimental trials, two order types,
normal and pathological, were dened. We used the nor-
mal orders to represent typical industrial problems, and,
in this case, implying that both ply height and order
quantities are multiples of 12. We chose pathological
orders to test algorithmic performance on odd numerical
combinations, in this case, ply height and/or order
quantities that are not multiples of 12. Three normal
orders and two pathological orders were created. For
these ve orders, three alternative ply heights were spec-
ied for dierent runs. In addition, the results from the
algorithms were recorded before and after the improve-
ment algorithm was applied. These various combinations
of the problem parameters resulted in 20 problem in-
stances, presented in Table 4. In addition to the results
shown in this table, one more trial was run with all fac-
tors set at their `high' settings to perform validation of the
model results.
We used these testbed data to obtain results for the
various problem instances by using two commercial
software packages that are widely applied to solve the
COP in industrial settings. The seven performance mea-
sures of interest were as follows:
(1) total cost ($);
(2) number of patterns;
(3) number of sections;
(4) total ply count;
(5) number units over demand required;
(6) number units under demand required;
(7) fabric utilization.
Using the output from the trial runs, we performed a
statistical analysis based on a 12-run PlackettBurman [8]
design. The analysis indicated that the performance
measures were not sensitive to most of the factors shown
in Table 3. The most conclusive nding of the analysis is
the signicantly large impact of the cost of fabric on the
total cost of the cut order planning process. Not only
does this conclusion make logical sense from the problem
environment, but is a fact that can be capitalized on for
solution of the cut order planning problem. Because
fabric cost is directly related to fabric length, we used the
criterion of total fabric length alone to compare the per-
formance of the algorithms. The remainder of this section
Table 3. Factors for the experimental design
Factor label Factor description Values used
A Fabric cost $0.50 or $10.00
B Number of pieces in the order 48 or 1200
C Distribution of sizes in order Uniform or
non-uniform
D Number sizes within order 1 `or' 6
E Cutting labor cost $10 or $30
F Spreading labor cost $8 or $25
G Maximum ply height 47 m or 108 m
H Order lling requirements Exact or
approximate
I,J,K Estimating sample error
Table 4. Experimental problem instances
Problem
instance
Normal order
(d
1
/d
2
/d
3
/d
4
/d
5
/d
6
)
Pathological order Ply
height
Improvement
algorithm applied?
1 72/144/360/360/144/72 48 No
2 72/144/360/360/144/72 48 Yes
3 72/144/360/360/144/72 108 No
4 72/144/360/360/144/72 108 Yes
5 0/0/0/0/960/240 48 No
6 0/0/0/0/960/240 48 Yes
7 0/0/0/0/960/240 108 No
8 0/0/0/0/960/240 108 Yes
9 0/0/0/0/1200/0 48 No
10 0/0/0/0/1200/0 48 Yes
11 0/0/0/0/1200/0 108 No
12 0/0/0/0/1200/0 108 Yes
13 163/239/599/45/124/30 47 No
14 163/239/599/45/124/30 47 Yes
15 163/239/599/45/124/30 108 No
16 163/239/599/45/124/30 108 Yes
17 200/200/200/200/200/200 47 No
18 200/200/200/200/200/200 47 Yes
19 200/200/200/200/200/200 108 No
20 200/200/200/200/200/200 108 Yes
86 Jacobs-Blecha et al.
will describe the results of our experimental study to ex-
amine the performance of the COP algorithms.
4.2. Empirical testing of the COP heuristics
The performances of the algorithms described in Section 3
were compared with each other and with the results ob-
tained from two commercial COP packages. Additional-
ly, we applied the improvement algorithm to all of these
solutions, as well as to randomly generated solutions.
To illustrate the results, the normal orders with ply
height 48 are given in Figs. 7 and 8. In Fig. 7, the solu-
tions denoted `Improvement' are the result of applying
the improvement algorithm to an initial solution com-
posed of each size in the marker in a separate section of
the marker. Note that in each case this solution is rela-
tively close to the best of all the others. Solutions for the
problems with ply height 108 and normal orders are
similar and consistent with these results.
Figs. 9 and 10 show the objective solution value for the
pathological orders with ply height 47 before and after
improvement, respectively. Similar results were obtained
for the pathological orders with ply height 108.
4.3. Interpretation of experimental results
Three primary conclusions can be drawn from the nu-
merical results:
1. The Savings algorithm performs signicantly better
than the Cherry Picking algorithm.
2. The Savings algorithm provides solutions that are as
good as or better than the commercial packages.
3. The Improvement algorithm is able to make im-
provements in all solutions, even the commercial ones.
After the improvement algorithm is applied, all solutions
are similar in the total required fabric length.
5. Conclusions and extensions
This research documents the rst known experimental
verication of fabric costs being the dominant determi-
nant of the cut order planning solution. This drives home
the point that heuristic solutions, including those used in
commercial packages, are critically dependent on the es-
timation of fabric length required to cut a particular
Fig. 7. Total fabric inches for normal orders with ply height 48.
Fig. 8. Total fabric inches for improved normal orders with ply
height 48.
Fig. 9. Total fabric inches for pathological orders with ply
height 47.
Fig. 10. Total fabric inches for improved pathological orders
with ply height 47.
Cut order planning for apparel manufacturing 87
combination of sizes together in a marker. Therefore it is
important that this information be obtained from his-
torical data or correctly estimated by experienced oper-
ators. Ideally these data would be available from the
actual markers used in the cutting process.
Major analytical contributions of this paper include the
development of a mathematical model and its complexity
analysis, and the improvement algorithm. The mathe-
matical model for the COP problem is new to the apparel
manufacturing literature. The complexity analysis de-
scribed in Section 2.2 establishes NP-completeness. For
large-scale industry problems, this means that no solution
method (including commercial packages) based on cur-
rent computational capabilities is likely to guarantee
mathematically optimal COP solutions. Simple heuristic
methods were shown to be at least as good as the com-
mercial packages, and an improvement algorithm was
developed that increased fabric utilization when applied
to solutions obtained from commercial software. Exhib-
iting versatility, this method was able to produce solu-
tions from a random starting solution that were
essentially equivalent to improved solutions from com-
mercial packages and other simple heuristics. This means
that COP can be solved easily by anyone with a desktop
computer, our improvement algorithm, and reasonably
accurate information for marker lengths.
The COP problem as addressed here is performed in-
dependently of downstream production considerations.
This is done even though the output of COP is the direct
input for marker making and the cutting room. Given the
ever changing status of current orders, current WIP in the
system, and production conguration, there is a signi-
cant opportunity to make the production system more
ecient and responsive by better coordination of dy-
namic, integrated planning and scheduling of WIP re-
lease, WIP movement, and conguration of the
associated exible production capacity. For these reasons
it is recommended that future COP research extend this
work to capitalize on the adaptive capabilities of the
improvement algorithm. Such extensions should explic-
itly include material ow control considerations. This will
allow consideration of the current status of the assembly
operations and the reection of competing system ob-
jectives. This extension will bring the COP process closer
to true integration within the total production planning
process.
The results reported here expand readily into the more
comprehensive planning environment. An illustration of
this concept is presented in Fig. 11.
Acknowledgements
We thank Mr Bill Warden, who worked tirelessly to per-
form the computational work for the PlackettBurman
study; Dr Donna Llewellyn, in the School of Industrial
and Systems Engineering at Georgia Tech, for her devel-
opment of the complexity proof; the two apparel vendors
(who wish to remain anonymous) for providing us with
their software for use in this work; Ms Cynthia Holeridge
from a very helpful apparel manufacturer, who spent long
periods on the telephone with us to provide and validate
our testbed data; Randolph Case of the Georgia Tech
Research Institute for moral support as well as for giving
us his best graduate student for this project; the Apparel
Demonstration Center on the Southern Tech Campus for
many instances of help and assistance; and Dale Stewart
and Carol Ring, at the Demonstration Center, for their
patience and assistance. The presentation has been sig-
nicantly improved by the helpful comments of the as-
sociate editor and three anonymous referees. Finally, we
must acknowledge the interest and encouragement in this
work by our sponsor's representatives, Donald O'Brien,
Dan Gearing, Helen Kerlin, and Julie Tsau. This work
was supported by the Defense Logistics Agency under
contract number DLA900-87-D-0018-0012 and by the
Georgia Institute of Technology.
References
[1] Jacobs-Blecha, C. et al., (1990) Cut order planning Final technical
report DLA 900-87-D0018-0012
[2] Johnson, D.S. et al., (1974) Worst-case performance bounds for
simple one-dimensional packing algorithms. SIAM Journal of
Computing, 3, 299325.
[3] Eilon, S. and Christodes, N. (1971) The loading problem.
Management Science, 17, 259267.
[4] Farley, A.A. (1988) Mathematical programming models for cut-
ting-stock problems in the clothing industry. Journal of the Op-
erational Research Society, 39 (1), 4153.
[5] Gilmore, P.C. and Gomory, R.E. (1961) A linear programming
approach to the cutting stock problem. Operations Research, 9,
349359.
Fig. 11. A comprehensive cutting room scheduling system.
88 Jacobs-Blecha et al.
[6] Hinxman, A.I. (1980) The trim-loss and assortment problems: a
survey. European Journal of Operational Research, 5, 818.
[7] Jacobs-Blecha, C. and Riall, W. (1991) The feasibility of im-
proving the marker making process. International Journal of
Clothing Science and Technology, 3(4), 1324.
[8] Plackett, R.L. and Burman, J.P. (1946) Design of optimal multi-
factorial experiments. Biometrica, 3, 305325.
[9] Garey, M.R. and Johnson, D.S. (1979) Computers and Intracta-
bility, A Guide to the Theory of NP-Completeness, W. H. Freeman,
New York.
Appendix A. The COP problem:
proof of NP-completeness
We will show that our (COP) model is equivalent to the
integral knapsack problem, which is known to be NP-
complete (e.g. [9] ). Consider the linearized formulation
for (COP) developed in Section 2.2.
Minimize

J
=1

I
i =1
c
i
1

i
L
!
:
i
M
i
Ue
i
[ [x
i
(COP)
subject to

J
=1

I
i =1
d
si
:
i
d
s
= d
s
\ sY

J
=1

I
i =1

i
x
i
_ LY
:
i
_ x
i
P \ iY Y
:
i
0Y 1Y 2Y F F F Y P \ iY Y
x
i
0Y 1 \ iY X
Now consider a special case where:
(a) c = 0Y 1 = 0, and
(b) all d
si
= 0 and d
s
= d
s
.
Let M
i
Ue
i
= W
i
Y \i. The model for this special case
then becomes:
Minimize

J
=1

I
i =1
W
i
x
i
(SCOP)
subject to

J
=1

I
i =1

i
x
i
_ LY
x
i
0Y 1 \ iY X
Now let u
i
Z

, where u
i
=

J
=1
x
i
. The (SCOP) for-
mulation becomes the following:
Minimize

I
i =1
W
i
u
i
(K)
subject to

I
i=1
!
i
u
i
_ LY
u
i
_ J \ iY u
i
Z

X
(K) is the formulation for an integral knapsack problem,
and so is NP-complete (e.g. [8]).
Proposition A1. x
i
solves (SCOP).
If we let u
i
=

J
=1
x
i
, then u
i
is feasible in K.
Proof:
u
i
=

I
i =1
x
i
_

I
i =1
1 x
i
0Y 1

(A1)
= J
= u
i
_ JY \ iX

I
i =1

i
u
i
=

I
i =1

J
=1
x
i
=

I
i =1

J
=1

i
x
i
_ LX (A2)
u
i
=

J
=1
x
i
Z

X (A3)j
Proposition A2. u
i
solves (SCOP).
If u
i
= 0 !et x
i
= 0 \ X
If u
i
= d ,= 0Y !et x
i1
= x
i2
= F F F = x
id
= 1;
x
i
= 0 )or b dX
Observe that

J
=1
x
i
= u
i
.
Proof:
x
i
0Y 1 \ iY X (A4)

I
i =1

J
=1

i
x
i
=

I
i =1

J
=1
x
i
=

I
i =1

i
u
i
_ LX (A5)j
With Propositions A1 and A2 we have shown that fea-
sible solutions to (K) are also feasible in (SCOP). Since
(K) is NP-complete, then so is (SCOP). Since (SCOP) is a
special case of (COP), then (COP) is also NP-complete.j
Biographies
Charlotte Jacobs-Blecha is a Senior Research engineer in the School of
Civil and Environmental Engineering (CEE) at the Georgia Institute of
Technology. Dr Jacobs-Blecha is working with the faculty in CEE to
develop, promote, and manage continuing education programs related
to developing and emerging technologies in civil engineering. Her own
research interests have focused on optimization and heuristic ap-
proaches to problems in the area of transportation, logistics, and
manufacturing. In 1989 she was placed second in the national Doctoral
Dissertation competition of the Institute of Industrial Engineers.
Dr Jacobs-Blecha is a member of IIE, INFORMS, and serves on the
Education Subcommittee for ITS America's Institutional Issues
Cut order planning for apparel manufacturing 89
Committee. She works with business and industry in the Atlanta
community, supporting eorts to improve transportation and envi-
ronmental problems. In 1995 she served on the executive committee for
the Metro Atlanta Clean Commute Day.
Jane Ammons is an Associate Professor of Industrial and Systems
Engineering at the Georgia Institute of Technology, where she teaches
and performs research in the areas of production system design and
analysis with a focus on manufacturing systems and quality manage-
ment. She has co-authored over 75 technical papers and presentations
on topics in the areas of electronic assembly systems, apparel manu-
facturing, material handling, electric generation capacity planning,
exible manufacturing and exible assembly systems, environmentally
conscious systems, quality, and total quality management. Dr Ammons
serves in a leadership role on a number of national boards and panels
related to her professional activities. She is a member of IIE, SME,
INFORMS, and ASEE.
Avril Schutte is the Maintenance Team Manager with Cummins En-
gine Company in Columbus, Indiana. She holds both Bachelor's and
Master's degrees in Industrial Engineering from the Georgia Institute
of Technology. Before her current position she worked as a graduate
research assistant at the Georgia Institute of Technology. In addition
to her work with Cut Order Planning, she assisted with a material-
handling project dealing with multiload carriers and the associated
system design. Her current responsibilities at Cummins Engine include
management of four plant maintenance teams, project management for
the relocation of two plants, and facility management of a 1000-person
oce. Ms Schutte is an associate member of IIE.
Terri A. Smith is currently working with MCI as a Project Manager for
the Small Business Online Lead Database System. She organizes and
plans the developments of the project while working closely with the
users of the system to establish requirements for changes and upgrades
to the current system. Ms Smith also does development work for the
Sybase database, Online system and Intranet projects. She was for-
merly an employee of the Georgia Tech Research Institute at the
Georgia Institute of Technology, where, in addition to her contribu-
tions to the Cut Order Planning research, she designed, developed,
tested, and integrated a ight-planning system and a penetration
analysis system for the US Air Force. She received a Bachelor's degree
in computer science from the University of Georgia, and later worked
at Milliken and Company.
90 Jacobs-Blecha et al.

Vous aimerez peut-être aussi