Maihl. Covnput. Modelling Vol. 16, No. 1, pp. 1-8, 1992 0895-7177192 $5.99 + 0.

06
Printed in Great Britain. All rights reserved Copyright@ 1992 Pergamon Press plc

ONE-DIMENSIONAL CUTTING STOCK PROBLEMS

AND SOLUTION PROCEDURES
ROBERT W. HAESSLER
School of Business, University of Michigan
Ann Arbor, MI 48109, U.S.A.

Abstract-This paper provides an introduction to one-dimensional cutting stock problems and

solution procedures. The first problem considered requires that both trim loss and pattern changes
be controlled. Both linear programming and sequential heuristic procedures are discussed along with
the ways they can be used jointly to generate the best possible solutions to this type of problem. Two
other important classes of one-dimensional problems me discussed along with ways in which they can
be solved.

INTRODUCTION

Common industrial problems, such as determining how to slit production rolls of paper, cut up
steel bars, assign products to rail cars, containers or trailers, balance assembly lines and cycle
products on a machine, appear on the surface to be quite different, but are, in fact, very closely
related. They are all examples of a class of problems known as cutting and packing problems. A
simple example of this type of problem is the trim loss minimization problem which occurs in the
paper industry. In this problem, known quantities of various width rolls of the same diameter are
to be slit from stock rolls of some standard width and diameter. The objective in this case is to
find slitting patterns and their associated usage levels which produce the requirements for ordered
rolls, at the least possible cost for scrap and other controllable factors. The basic restriction in
this problem is that the sum of the roll widths slit from each stock roll must not exceed the
usable width of the stock roll. The other problems listed above have exactly the same type of
restrictions. The only difference is that the focus in these problems may be on the utilization of
space or time, rather than material.
A cutting or packing problem is a one dimensional problem if this basic restriction can be
stated in the form:

c Wi Ai 5 VW, (1)
i
where, in this roll example,
- Wi is the size of requirement i (roll width i),
- Ai is the integer number of times requirement i appears in the pattern, and
- UW is the capacity (usable width of the stock roll).
Examples of higher dimensional cutting and packing problems for which restriction (1) above
is not sufficient to describe a feasible pattern are given below:
- cutting rectangular pieces of glass from larger stock rectangles,
- cutting irregular shapes from a steel plate or bolt of cloth,
- cutting boards from a log,
- packing a container based on the dimensions of the items to be packed rather than their
weight.
This paper will focus on one-dimensional cutting stock problems. The techniques described for
cutting stock problems will work equally well for the packing problems described above, except

Typeset by A,#-‘&X

1
2 R.W. HAESSLER

for the assembly line balancing problem. In that situation, the presence of complex precedence
relationships may make it necessary to use other approaches to solve the problem.
The next section contains a formulation of a common one-dimensional cutting stock problem
along with a discussion of the major solution alternatives for solving this problem. The following
section discusses some common variants of this problem along with solution approaches. The
paper concludes with an identification of an important problem on which additional research is
needed.

INITIAL PROBLEM DEFINITION

An initial statement of a one-dimensional cutting stock problem is given below in terms of
slitting ordered rolls of some material such as paper from a single stock size. It is assumed that
the order requirements are for Ri rolls of width Wi, i = 1,. . . , n, to be cut from stock rolls of
usable width UW. All orders are for the same diameter.

Min Cr c TjXj + CZ c 6(Xj),

j j

s.t. RLi 5 CAijXj 5 RUi, for all i, (3)

Xj 2 0, integer, where (4)
Aij is the number of rolls of width Wi to be slit from each stock roll that is processed using
pattern j (s.t.: subject to). In order for the elements Ai;, i = 1,. . . ,n to constitute a feasible
cutting pattern, the following restrictions must be satisfied:

c Aij Wi 5 VW, (5)

Aij >_ 0, integer. (6)
- Xj is the number of stock rolls to be processed according to pattern j;
- Tj is the number of units of trim loss incurred by pattern 3,

Tj = UW-CAijWi;

- Ci is the dollar value of trim loss per unit;

- Cz is the cost of changing patterns in dollars;
- S(Xj) is 1 for rij > 0 and 0 otherwise; and
- RLi, RVi are the lower and upper bounds on the order requirement, Ri, for customer
order i reflecting the general industry practice of allowing overruns within specific limits.
Depending on the situation Ri may be equal to either R L; or R Ui or both.
Note that the objective in this example is not simply to minimize trim loss. In virtually all
industrial applications, it is necessary to consider other factors in addition to trim loss. In this
example, a cost is associated with a pattern changes and, therefore, controlling the number of
patterns used to satisfy the order requirements is an important consideration. In other applica-
tions, some other factor might be important. For example, when assigning discrete products to
containers by weight, to minimize the number of containers used, it might be desirable to have
like products shipped in the same container to facilitate material handling at both the origin and
destination of the shipment.
Because optimal solutions to this problem can be found only for values of n smaller than those
typically found in practice, heuristic procedures represent the only feasible approach to solving
this type of problem. Two types of heuristic procedures have been widely used to solve one-
dimensional cutting stock problems. One approach uses the solution to a linear programming (LP)
problem as its starting point. The LP solution is then massaged in some way to provide a solution
to the problem. The second approach is to generate cutting patterns sequentially to satisfy some
portion of the remaining requirements. The sequential heuristic (SH) procedure terminates when
all ordered requirements (3) are satisfied.
 One-dimensional cutting stock problems 3

LP SOLUTIONS

Almost all LP based procedures for solving cutting stock problems can be traced back to
the seminal work of Gilmore and Gomory [1,2]. They described how the next pattern to enter
the basis could be found by solving an associated knapsack problem. Problems of minimizing
trim losses could then be solved by linear programming techniques without first generating every
feasible slitting pattern.
A large number of slitter patterns may exist when narrow widths are to be slit from a wide
stock roll. Pierce [3] showed that, in such situations, the number of slitting patterns can easily
run into the millions. However, only a small fraction of all possible slitting patterns need to be
considered in finding the minimum trim loss solution. The delayed pattern generation technique
developed by Gilmore and Gomory made it possible to solve trim loss minimization problems in
much less time than would be required to generate the slitting patterns to prepare the input for
a general-purpose linear programming algorithm.
Referring to (2)-(4), if the cost of changing patterns is dropped from the objective function,
and the integer requirement on pattern usage is relaxed, the linear programming problem to be
solved can be stated as
Min c Xi, (7)
j

cj
AijXj > RLi, for all i, (8)

Xj 20. (9)

Note that minimizing the number of production rolls rather than trim loss forces the upper bound
on each order width to be automatically satisfied because (8) can be restated as an equality.
If Vi is the dual price associated with constraint i in (8), the reduced cost for any nonbasic
pattern A = (Ai,. . . , A,,) is 1 - Ci UiAi. Therefore, the LP problem (7)-(g), can be solved
without first generating all possible cutting patterns. The next pattern to enter the basis, if one
exists, can be found by solving a knapsack problem:

2 = maxx UiAi, (10)

i

c WjAi < VW, (11)

Ai 2 0, integer. (12)

If Z 5 1, the current solution is optimal. If Z > 1, then A = (Al,. . . , An) can be used to improve
the solution.
This LP solution can then be massaged in a number of ways to obtain integer values for Xj
and completion of each order. The primary disadvantage of using LP to solve the problem in
(2)-(4) is that the number of nonzero cutting patterns in the LP solution will be close to n, the
number of sizes ordered. This may be acceptable only if controlling trim loss is very difficult and
LP is the only way to find a low trim loss solution.

SEQUENTIAL HEURISTIC SOLUTIONS

With this approach, a solution is constructed one pattern at a time, until all the order require-
ments (3) are satisfied. The first documented SH procedure capable of finding better solutions
than those found manually by human schedulers was described by Haessler [4]. The key to SUC-
cess with this type of procedure is to make intelligent choices as to which pattern will be selected
next. The pattern selected should have low trim loss, high usage and leave a set of requirements
for future patterns that does not guarantee trouble because the remaining sizes do not combine
well.
4 R.W. HAESSLER

The following procedure is capable of making effective pattern choices in a variety of situations:
1. Compute descriptors of the order requirements yet to be scheduled. Typical descriptors
would be the number of stock rolls still to be slit and the average number of ordered rolls
to be cut from each stock roll.
2. Set goals for the next pattern to be entered into the solution. Goals should be established
for trim loss, pattern usage and number of rolls in the pattern.
3. Search exhaustively for a pattern that meets those goals.
4. If a pattern is found, add this pattern to the solution at the maximum possible level
without exceeding Ri, for all i, reduce the order requirements and return to 1.
5. If no pattern is found, reduce the goals for the next pattern and return to 3.
The pattern usage goal is a lower bound on the level at which the pattern w.ill enter the solution
if it is chosen; for example, if some width has an unmet requirement of 10 rolls and the pattern
usage goal is 4, that width may not appear more than twice in a pattern that satisfies the goal.
If no pattern is satisfactory, then some goal-most commonly pattern usage-must be relaxed,
allowing more patterns to be considered. If the pattern usage goal is changed to 3 in the above
example, then the width for which 10 rolls are required can appear in the pattern three times.
Termination can be guaranteed by eventually reducing the goals to permit a pattern of lowest
trim loss at a usage level of one.
The primary advantage of the sequential approach is its ability to limit pattern changes by
searching for high usage patterns, and to eliminate rounding problems by working only with
integer values. It may, however, result in greatly increased trim loss because of what might
be called ending conditions. For example, if care is not taken as each pattern is accepted and
the requirements reduced, the widths remaining at some point in the process may not have an
acceptable trim loss solution. Such would be the case if only 36-inch rolls are left to be slit from
loo-inch stock rolls.

JOINT SOLUTION PROCEDURES

In order to find the best possible solution to a class of one-dimensional cutting problem such
as the one defined in (2)-(4), th e t wo basic solution approaches can be combined in a number of
ways as described below.
The SH procedure can be used with the LP solution to obtain an integer solution which
satisfies all order requirements in the following way. The LP solution could first be rounded
down, and then nonzero usage patterns could be increased in unit increments so long as
production of any size does not exceed RUi. Any orders for which production falls below
R Li can be completed by using the SH procedure to generate new patterns.
The SH procedure is used to generate a solution which is saved, and also used as an initial
basis in the LP procedure. Additional LP iterations are made only to the extent the trim
loss is reduced. The better of the SH and LP solutions is selected according to the criterion
specified in (2).
The problem is first solved as an LP problem in order to obtain optimal dual prices.
These dual prices are used as an additional test before accepting a pattern in a SH
procedure to ensure that the pattern does not contain a disproportionate share of the
low relative dual price sizes. For patterns in the optimal or alternate optimal trim loss
solutions,
Z = C AiUi = 1.
i
If the value of Z is less than .97 or .98, for a pattern accepted in a SH procedure, the total
trim loss of the SH solution may be increased significantly. Although this test makes it
possible to avoid making some mistakes when selecting a pattern using a SH approach, it
is not foolproof because the SH may use the pattern at too high a level.
As the SH nears the completion of the requirements and the pattern selection decision
becomes more difficult, patterns for residual requirements are generated using LP. If the
residual LP solution does not meet some target value which is based on the original LP
 One-dimensional cutting stock problems 5

solution to the entire problem, the sequentially generated patterns are dropped in reverse
order of generation and the expanded residual problem is resolved using LP. This dropping
of sequentially generated patterns continues until either a satisfactory solution is obtained
or all the patterns are dropped, at which point the LP solution with the best possible trim
loss is generated.
The advantage of this last approach is that it truly integrates the strengths of the SH procedure
to consider factors such as slitter changes and the LP procedure to minimize trim loss into a single
procedure. This procedure is capable of giving either a pure SH or LP solution if that is what is
best. Most importantly, however, is its ability to generate solutions which are part SH and part
LP and, therefore, likely to be better than either the pure LP or SH solutions. Haessler [5] has
used it effectively to solve difficult, from a trim loss standpoint, problems of the type defined in
C+(4).

OTHER ONE-DIMENSIONAL PROBLEMS

To this point, the only problem discussed has involved controlling the number of patterns used
along with the trim loss. Although this is an important class of one-dimensional problems, it is
by no means the only important class. Two other important classes will be considered here. The
first deals with multiple stock sizes and the second deals with pattern restrictions imposed by
the product or process.

Multiple Stock Sizes

A general statement of this problem, for the same order requirements used earlier, is as follows.
In this case, we permit the possibility that different stock sizes may be available at different
locations, and therefore freight cost also influences the choice of which stock size will be used.

s.t. RLi < CxAijkXjk 5 RUit for all 6

k j

c
xjk 5 Mk, for all k,

xjk 2 0, integer,

where
- Aijk is the number of rolls for order i to be cut from stock width k using pattern j. In
order for Aijk to be a feasible cutting pattern, for i = 1,. . . , n, the following condition
must be satisfied

c AijkWi 5 UWk

Ajk 1 0, integer.

- xjk is the number of stock of width k to be processed according to pattern j.

- Tjk is the trim loss incurred by using pattern j with stock width k.

Tjk = UWk - C AijkWi.

i

- Clk is the dollar value of trim loss per unit for stock width k.
- Caki is the cost of shipping one roll for order i which is produced from stock width k.
It is assumed that the stock width defines the production location. If all the production
options are at the same location, this value can be set to 0.
- Mk is the maximum number of rolls of stock width k which can be used.
6 R.W. HAESSLER

Real world problems of this sort usually require the power of LP to deal with the complex
tradeoffs which can arise.
The LP relaxation of this problem, developed by Beged Dov [6] is

min Cjkxjk 9
c
jk

s.t. C AijkXjk 2 R Li, for all i,

jk
dYj, 10.

For the most general case with varying costs for trim loss and freight, C’j, can be represented as
follows
cjk = ck + Cpk (C Ajk Wi)+ c C3kiAijk.
i i

ck is the cost excluding material of making one stock roll of width Iz.
Cpk is the material cost per inch of that portion of production roll Ic actually used.
Cski is the cost of shipping one roll for order i from the location where stock width R is
made.
In this situation, the shadow prices must be adjusted before generating the cutting pattern
that enters the solution next, if one exists. The following associated problem must be solved for
each stock width.
Zk =maxC(Ui - CpkWi - C3ki) Aijk - ck,

s.t.
c WiAijk <
i
uwk,

Aijk 2 0, integer.

The pattern for which zk has the largest value greater than zero enters the solution.

Pattern Restrictions

For this discussion refer back to the problem defined in (2)-(4). The definition of an acceptable
pattern will be extended to consider restrictions beyond

c AijWi 2 UW.

1. Limited slitter capacity

If C is the maximum number of rolls which can be slit from each stock width on the
primary slitter, then each pattern must satisfy the additional restriction

If the rolls are too narrow, then it may be necessary to generate master rolls to be slit
at a secondary operation. In this case, the objective would be to minimize the number
of such rolls. In a SH procedure, this can be accomplished by initially combining the
narrowest rolls into patterns that contain master rolls for secondary slitting, until the
remaining rolls are wide enough to be slit at the primary operation. In a LP procedure
this can be handled using two stock sizes where the second stock size has a higher cost to
reflect the secondary slitting, and a lower usable width to reflect the edge trim required
at a secondary slitter.
 One-dimensional cutting stock problems 7

2. Two-stage processing
In some applications, such as the production of film or coated grade of paper, the
stock rolls must be slit, processed through a finishing operation which is not capable
of processing the full stock width roll, and then slit to the width required by the final
customer. Haessler [7] d iscussed a procedure for solving this problem in the film industry.
In this case, each pattern (Al, A-_?, . . . , An) must be capable of being partitioned to meet
the requirements of the finishing operation. Each component Ai of the pattern can be
thought of as being the requirements R; of a trim problem with a limited number of stock
sizes of usable width, FW, where FW is the maximum width roll which the finishing
equipment can process,
3. Quality variations across the width of the stock size
Sweeney and Haessler [8] d eveloped a procedure for solving the one-dimensional trim
problem in the case where the order requirements may be satisfied by lower quality grades
which may occur during the production process. If there are no quality variations across
the width of the stock roll, this first quality material can be used to satisfy any order for
that grade. A two phase procedure is used. In the first phase all non perfect stock rolls
are considered. Each roll is assigned a value based on the nature of the quality variations.
Patterns are generated for each stock roll, beginning with those with the severest quality
variations. Each pattern is given a value based on the shadow prices obtained from solving
an LP problem for the order requirements and stock rolls without any quality defects. If
the value of the pattern exceeds the value of the stock roll, the pattern is accepted for that
one stock roll. Any order requirements not met from stock rolls with quality variations are
slit from first quality stock rolls based on an LP solution to this residual problem found
in phase two.
4. Order contiguity
In cutting stock problems where the time required to satisfy all the requirements is
larger, it may be necessary to try to produce all of each order within some limited time
period. This may be due to a need to ship some order by a certain time, or it may be due
to a general need to be shipping something all the time because of limited dock or work in-
process storage space. This problem is extremely important in mills which produce large
volumes of commodity grades of paper, such as kraft liner and medium for corrugated
boxes. The issue may be defined as limiting the number of orders that have been partially
produced at any point in time. The problem is further complicated if some orders require
multiple railcars or truck trailers to ship, and the issue is defined as limiting the number
of partially filled vehicles at any point in time.
There are a number of ways to approach this problem. Although using one might work
reasonably well in a given situation, none really provides a satisfactory general purpose
method of dealing with this problem. The most powerful of these approaches involve
pattern restrictions based on predetermined or dynamic sequencing of orders.
A predetermined sequencing would involve partitioning either orders, or full railcar or
trailer portions of orders into subgroups. In an LP approach, the only patterns generated
initially would be those containing ordered sizes assigned to the same subgroup. Once an
optimal LP solution for each subgroup is obtained, the procedure stops if a judgment is
made that the trim loss is low enough. If the trim loss is judged to be too high, additional
patterns can be generated using ordered sizes from adjacent subgroups. This expansion of
ordered widths which can be combined in the same pattern continues until a satisfactory
trim loss solution is obtained. Clearly, the effectiveness of this type of procedure depends
on how well the subgroups have been formed and the production sequence in which they
have been arranged.
A more dynamic approach would involve selecting subgroups during the solution process
based on trim and shipment considerations. An LP solution for all orders regardless of
contiguity would indicate overall attainable trim loss and optimal dual prices. Once a
starting order is established, an integrated LP and SH procedure could simultaneously
generate patterns and start new orders as needed to provide low trim loss patterns which
complete orders in a systematic fashion.
8 R.W. HAESSLER

CONCLUSION

This paper has considered a variety of one-dimensional cutting stock problems and the ways
in which they can be solved. Of the problems identified, the most difficult one is the order
continuity problem. If the number of orders is more than a few, the number of sequences in
which the orders can be arranged for completion becomes very large. Research is needed to find
ways to sequence orders for production or to partition groups of orders for joint production,
so that the requirements of the material handling and shipping departments can be considered
when generating solutions to large cutting stock problems. An even more complex variation of
this problem exists when the same grade is being run simultaneously on two or more machines.
In this case, product flow off each machine must be coordinated to ensure efficient handling and
timely shipment of each order.

REFERENCES

1. P.C. Gihnore and R.E. Gomory, A linear programming approach to the cutting stock problem, Oper. Rec.
9 (6), 848-859 (1961).
2. P.C. Gilmore and R.E. Gomory, A linear programnu ‘ng approach to the cutting stock problem, Oper. Res.
11 (6) 863-888 (1963).
3. J.F. Pierce, Some Large Scale Producfion Problems in the Paper Industry, Prentice-Hall, Inc., Englewood
Cliffs, NY, (1964).
4. R.W. Haessler, A heuristic progr amming solution to a nonlinear cutting stock problem, Management Science
17 (12), 793-802 (August, 1971).
5. R.W. Haessler, A new generation of paper machine trim programs, TAPPI 71 (8), 127-130 (August, 1988).
6. A.C. Beged-Dov, Some computational aspects of the A4 paper mills and P printers problem, Journal of
Business Research 1, 15-34 (1970).
7. R.W. Haessler, Solving the two-stage cutting stock problem, Omega 7 (2), 145-151 (1979).
8. P.E. Sweeney and R.W. Haessler, One dimensional cutting stock decisions for rolls with multiple quality
grades, European Journal of Operations Research 44 (2), 224-231 (January, 1990).