You are on page 1of 6

TOC or LP?

Mabel Qiu, Lawrence Fredendall and Zhiwei Zhu have addressed the
production scheduling problem from two fundamentally different
approaches, the theory of constraints (TOC) and linear programming (Lf).
LP is a complicated mathematical approach while TOC is much less
quantitative in nature. Both approaches were applied to the real operations
data from a fibre manufacturer. It was found that the schedule produced by
TOC is very similar to the one generated by L f

heuristic was developed problem, we use the data provided mined after the spinning schedule is
using part of Goldratts in Qiu and Burch, who proposed developed.
theory of constraint (TOC) and tested a tnatheniatical hierarchical The fibre plant under study has
to perform hierarchical production planning model for a fibre four spinning machines. Each spinning
production planning. Goldratt and plant. This data was disguised to protect machine is divided into two splits or
Coxs heuristic consists of the five steps the firms confidential information, but groups of positions. Each split can
described below: it still provides a realistic test of the produce only one product at a time. The
perforinance of the T O C heuristic. productivity of a spinning machine is
(I) identify the systems constraint(s), measured as pounds per position per
(2) decide how to exploit the systems Plant background hour. A position is the smallest produc-
constraint(s), In a typical yarn manufacturing firm, tion unit, and the capacity of a split is
(3) subordinate everything else to the there are three basic processes that have determined by the number of positions
above decision, to be schedulcd: polynier preparation, at the split. The number of positions
(4) elevate the systems constraint(s),and spinning, and drawing. The primary varies for each split for this plant is
(5) if in the previous steps a constraint process of a fibre yarn nianufacturer is shown in Table 1.
has been broken, go back to step 1. spinning. Spinning controls the entire Setup costs are significant. The
production process and is usually the temperature is reduced to setup the
Many successful applications of this plant bottleneck due to its relative iliachine and any fibres produced before
heuristic have been reported by prac- capacities. Clearly if a feasible schedule the inacliitie reaches the correct temp-
titioners. A detailed discussion and is developed for spinning the rest of the erature are a lower quality fibre. In this
review of TOC can be found in several production planning becomes trivial. In plant, there are four different levels of
books-3 and articles4.The five steps do fact, scheduling the polymer production machine setups. Table 2 shows the times
not guarantee an optimal solution, but and the drawing operation is deter- and costs corresponding to each level of
they do provide optimal solutions in machine setups. In addition to these
most situations tested to dates,6.How- Table I Spinning machine splits four levels of machine setups, some
ever, most of the cases reported in prior capacity products are only run on certain splits.
research have been small and academic Those products that have a preferred
in nature. This prior research has not split are called standard products, while
examined the effectiveness of TOC Numberof 1 2 3 4 5 6 7 8 the products that do not have a prefcrred
Dositions 10 4 16 12 20 8 16 12
using real-world scheduling problems. split are referred to as non-standard
This paper examines the use of the products.
five-step heuristic in a hierarchical
Table 2 Times and costs of machine
production planning system. To the setups by level Linear programming model
best of our knowledge, there has been Master p d w t i o n schedule
no prior research applying the TOC Qiu and Burch7 developed an aggre-
heuristic to a hierarchical production 1 4 hours $1,868 gate cost miniinisation model to make
2 2 hours $700
planning problem. To ensure that the 3 1.5 hours $868 production schedules for the fibre plant.
T O C heuristic is tested with a realistic 4 0.5 hours 0 The aggregate model is shown below,

I90 MANUFACTURING ENGINEER AUGUST 2002

Authorized licensed use limited to: SUNY Buffalo. Downloaded on February 1, 2009 at 03:00 from IEEE Xplore. Restrictions apply.
Table 3 Legend
and the notation is explained in Table 3.

Subject to:

inachinc setup occurs when product is found in Table 4.


change-over takes place on a machine.
Constraint 6 is used to set lower and
upper bounds of monthly inventory
level for individual products. In the S.T.
aggregate model, all the variables are
continuous, except machine setup, Y,,,,,,
is a zero-one variable.
Given the information on forecasted
demand, inachiiie capacities, and total
M is a number big enough to force inventory target (warehouse capacity),
YILp= 1 when XILp# 0. the aggregate model determines the
N M monthly lot size, machine (or split)
lb 5 E XlL1 LI,5 ub for all i,p
,=1 L=l
(6) assignment and ending inventory level
for each individual product. At the
YiLp E (0,l) for all i,L,p disaggregate level, a network model
schedules production on each split to
The objective fmction of the model niininlise setup costs on each split
is made of three cost components; individually. Thus the daily operational
production, setup, and inventory. The schedule is determined at the aggregate
constraints are of three functional level.
types. Constraint 1 is used for the first
month at the beginning of the planning Daily operational sclzedule
horizon because the initial inventory A network model is used to dis- The objective functioii of this network
was recorded in pounds. Constraint 2 aggregate the aggregate schedule into a model niniiiises total setup cost. The
is the balance constraint for the rest of detailed schedule. The notation for the model uses the output of the aggregate
the planning horizon between the model given in Qiu and Burch's model model (i.e. desired production quanti-
demand and the variables of production
and inventory (in days of production). Table 4 Legend
Constraint 3 is the capacity constraint
Indices
that limits production quantities and
d = 1,...,Q day
inventoiy levels during each period. i , j = l , ...,N product
Constraint 4 is the aggregate inventory L = I ,...,M split
P Time period
target that specifies level of inventories
to be used to smooth production. The Parameters
CSll set up cost to switch from product i to product j
inventory target I, is set based on CLP capacity (in days) of machine L during period p
the physical warehouse capacity. The DIP demand (in pounds) of product i during period p
inventory shortage variable S,, is utilised Decision variables
to avoid unfeasibility at the lower level YiLd = 1 if production of product i takes place on machine L on day d,
= 0 otherwise
of the model and maintain a stability of ZijLd = 1 if a changeover from product i to j occurs on machine L during day d,
the schedule. Constraint 5 ensures that = 0 otherwise

MANUFACTURING ENGINEER AUGUST 2002 191

Authorized licensed use limited to: SUNY Buffalo. Downloaded on February 1, 2009 at 03:00 from IEEE Xplore. Restrictions apply.
Table 5 Split preferences for each
ties and split assignments) and the initial between the selling price and the cost of
product number
split status (i.e. which products were last raw materials. The first 3 steps of the
run in the previous period) as input. Split TOC heuristic create a master schedule.
The variable Z y L d is defined as the 1 2 3 4 5 6 7 8 To adapt this heuristic to create both an
changeover from product i to product j 1 aggregate plan and a master schedule,
2 2
on split L at the beginning of day d: 4 4 we apply it in two separate terms. To
Zi,Lrl = I, if the changeover occurs and 5 5 create an aggregate production plan
0 otherwise. With this product a 6 6 6 we first use T O C to plan for each
7 7 7
production changeover can occur any 8 8 month. Then we use TOC to schedule
day, but there will not be more than 9 9 9 each machine, thus creating the master
one changeover on any machine in a 10 10 10 schedule. This two-step procedure is
11 11
given day. Thus, a 0-1 variable defines 12 12 similar to the traditional hierarchical
the possible changeovers. Constraint 7 14 14 planning procedure, which creates an
satisfies demand D,, which is deter- 15 15 aggregate plan and then creates a
16 16 16
mined at the aggregate level. Constraint 18 18 detailed schedule.
8 recognises capacity constraints. Con- 19 19 19
straints 9, 10 and I1 ensure that only 20 20 Master poductiou schedule
21 21
one setup is scheduled on a given split 22 An explanation of the use of the three
per day. Constraint 9 restricts splits to 23 23 steps for aggregate planning is below.
producing only one product a day. 24 24
25 25
Constraint 10 recognises that split L is Step 1: Identify the system constraint
changed from producing j on day d to This step was done by calculating
producing product i. And, constraint 11 schedule the plant. Step 4 of the five the capacity required at each split to
restricts changeovers between product step heuristic actually increases the produce the demand for the standard
j and i on day d to those where i is capacity of the bottleneck constraint. product. The first months demand for
produced on day d-I (is. the previous The purpose of this study, however, is standard products was then compared
day). It also ensures that a setup cost is to compare the effectiveness of T O C to the capacity available at each split.
assigned to the setup variable (ix. Z g L d ) scheduling to the Llmodel, so we The split with the tightest capacity was
when a production changeover occurs. do not consider the case of capacity considered to be the systems constraint.
Constraint 12 restricts each split to one increases. Likewise, Step 5, which is Only the standard products were used in
changeover per day to identify the new bottleneck after this calculation because these products
capacity is increased is not considered. are restricted to certain splits. The
Five-step heuristic As a first step in making TOC non-standard products can be run on
Only the first 3 steps of the T O C operational, we calculate the contri- multiple splits and can be scheduled
five step heuristic introduced at the bution margin for each product. The on a non-bottleneck split.
beginning of this paper are used to contribution margin is the difference The demand for the non-standard
products was computed and compared
Table 6 Preliminary loading of standard products to the available capacity remaining after
the standard products were scheduled.
Split ID Load of Products Time required on Total time required
the split respectively When the total deniand was greater
than the total capacity some products
could not be produced in that month.

Step 2: Exploit the system constraint


If there was any demand which could
not be scheduled, all the products were
ranked by their contribution margins
per pound divided by the processing
Table 7 Capacity analysis of production system
time per pound. The products were
then scheduled on a split in descending
Split ID total time required time available % required order of their contribution margin per
(in davs\ (in dam\
< , < > I ,
unit ofprocessing time. During the next
1 7 31 23% month if a standard product was to be
2 25 31 81 %
3 31 31 100% scheduled and there was no room on
4 0 31 0% its preferred split then the inventory of
5 13 31 42 % the standard product was used to satisfy
6 10 0 (testing new products) -
7 8 31 26% demand. If there was not enough
a 10 31 32% inventory that could be used and if there

I92 MANUFACTURING ENGINEER AUGUST 2002

Authorized licensed use limited to: SUNY Buffalo. Downloaded on February 1, 2009 at 03:00 from IEEE Xplore. Restrictions apply.
Split 4 (12 positions)
Month 1 2 3 4

processing 19 12 8 14 6 2 29 30
time +---A --__ + _---__ +A--+ __--_ A---+-+ ____-_____ +____________;
Product 19 T 19 21 14 14 T T

I = idle T = testing new products

w-ere noii-staiidard products scheduled about the use of other iliachines and excessivc setup times, we search for a
on its preferred split these non-standard personnel support the schedule devel- better split to place it onto.
products were moved to a different split. oped in step 2. This inay require that we An exaniple demonstrating the appli-
improve the daily schedule to reduce cation of these stcps is shown below.
Step 3: Subordinate everything else to setup times at the bottleneck splits. For Step 1 : the plant scheduler used the
the above decision exaniple, if loading of a non-standard split preference table (see Table 5) to
This step ensures that all decisions product to a bottleneck split creates assign products to their preferred split.

MANUFACTURING ENGINEER AUGUST 2002 I93

Authorized licensed use limited to: SUNY Buffalo. Downloaded on February 1, 2009 at 03:00 from IEEE Xplore. Restrictions apply.
Split 4 (12positions)
Month 1 2 3 4
processing 19 12 3 18 8 2 29 30
time +---A _ _ _ _ +_-__-_ +--+ _______ A_--+-+ ---__-____ +__-----___--
Product 21 T 14 19 1 19 T T

Split 5 (20positions)
Month 1 2 3 4
processing 6 15 10 1 1 0 15 2 1 7 14 8 5 25
time +--+ ___- AAA+-AA-+-~---+ _ _ - - +_- + - A - - + --_- +-AA+A-~, _ - - _ _ _ _ - I

Product 2 12 1 1 2 12 6 6 4 6 I I 6

Split 6 (8 positions)
Month 1 2 3 4
processing 31 3 25 1 1 4 7 9 30
time +--_-_-________-+-A ---_--____- + A __--- + --+ - A A + _ _ _ _ _ - A---
Product T I 10 10 20 21 I i

I = idle T = testing new products

As sta d earlier this iiiiproves both 1, 2, 4, 5, 8, 11, 12, 15, 18, 22, 23, 24 In table 7, split 6 was used for test-
quality aiid productivity. In Table 5, the aiid 25. The production times required ing new products, so no capacity was
bold faced product iiuinbers indicate to produce Jaii~arysdemand ofstandard available on split 6 in iiioiith 1 . Product
their preferred splits of each product. products on each split was calculated 12 had 18 days of inventory available,
For example, products 1 , 2 and 4 prefer (see Table 6). This was then compared which exceeded its requirements on
split 5. The standard products (Le. with the production time available on split 6 as shown in Table 6, so demand
products that prefer a split) are products each split (see Table 7). was met fi-ominventory Therefore, split

I94 MANUFACTURING ENGINEER AUGUST 2002

Authorized licensed use limited to: SUNY Buffalo. Downloaded on February 1, 2009 at 03:00 from IEEE Xplore. Restrictions apply.
Table 8 Non-standard products
3 was the most constrained resource. I n
January, 100% of the capacity on split 3
was needed for standard products. 9 $1.oo 1 day
Split 3 was the bottleneck with just 6 .95 5 days
the standard products loaded. The next 14 .90 1 day
7 .85 1 day
stage in step I was to load the non- 19 .80 12 days
standard products. Table 8 shows the 16 .75 13 days
noii-standard products, the splits that 21 .70 0
20 .60 7 days
the non-standard product can be loaded 10 .55 4 days
onto, and the demand for the product.
This product was loaded onto the Daily oyciatronal schedule and uses and generates good production
eligible split which had the most Once ~ lfour
l months of the projected schedules as cotiipared to the one by
capacity in descending order of their demand IE loaded onto a split, tlie L1. Further research needs to focus on
contribiition margin. Tlic capacity detailed cchedule is created ucing the testing TOC: on different industrial
analysis once the non-standard products firct three cteps of the TOC heuristic. settings and a variety of scheduling
were loaded is shown in Table 9. A detailed schedule wab created one problems.
This loading process was repeated iiioiith at J time using the T O C
for cach of the next three months. If heurictic. Step 1 in January is shown in References
a product could not be loaded in Table 9, and the constraint ie split 3. Step 1 Goldratt, E. M. and Cox,J.: 1992, The
the iiionth it was demanded, cither the Goal, second rcvised edition (Croton-ori-
Table 9 Loading of products for
previous months slack capacity was used January I-Iudson, N Y North Rivcr Press)
to produce enough of a product to 2 Goldl-att,E. M. and Fox, 11.:1986. Tile
Race (Cloton-on-Hudson, N Y North
eliminate the constraint or existing
River Press)
inventory was used to nicct demand. 3 Unible, M. M.and Srikantli, M. L.: 1990,
Step 2 of thc T O C heuristic is to 5 7 days Sytirhvmorrs A 4 a r ~ r ~ ~ c ~ i(Cincinnati,
thg
exploit the bottleneck. To do this nieaiis 15,23 25 days OH: South-Western Publishing)
18,22 3f days
that the iiiost profitable products are 14,, 20, 19 20 days 4 Fawcctt, S. E. and lcarsoii, J. N.: 1991,
loaded onto the bottleneck. As stated 1, 2,4, 9, 6, 7, 10 13 days Understanding and applying conscraint
earlier, the profitability measure used Test new product 31 days management in todays nianufacturiiig
25 8 days eiiviroiiiiients, Prodridoil arid Orver?fory
is the products contribution margin 24.16 13 davs
divided by its time on the bottleneck. l a / , (3),pp.46-55
M a / / a ~ C w ~ e n f J i J M u /32,

If a product has a contribution margin 2, examining the contribution margin, S L~iebbe,R. and Finch, 3.: 1992, Theory
of constraints and lincar progran~ming:
of $1 per pound weight and it takes shows that product I 8 is higher than
a comparison, I r z f e r r d o r z a l Jouri~al qf
4 minutes to produce a pound, then product 22, so product 18 is scheduled
Pvodirrfion Research, 30, (h), pp.1471-
its ration is $0,25/niin. In the example first. Since product 18 needs 30 days of 1478
given in Table 9 none of the loadings production, product 22 is scheduled for 6 Frcdendall, L. 13 and Lea, U. I<.:
determined in step 1 are changed. the last days of the month. Iniprovingthe product mix heuristic in
Step 3 requires inanagenient to sub- the theory of constraints, Oiferrzntiortal
ordinate all other decisions to decisions Conclusion J[JUV/?C?l (If
~ l ~ d l / c h Jl<CSCf?YC/l,
tl 35, ( h ) ,
made in step 2. An example of making This study attempted to look at the pp.1535-IS44
a decision to support step 2 would be production scheduling problem from 7 Qiu, M. M. and Uurch, E. E.: 1997,
to i~iove soiiie production forward two fuiidaiiientallydifferent approaches. Hierarchicalproduction and planning
when it is being produced on a niachine The theory of constraints (TOC) is sclicduliiig, Irrtrv~iatiowal Joiirrrcrl (I[
Production Rcscnvc-11,35, (1I ) , pp.3023-
that is a bottleneck in a given month. considered to be iiiucli less quantitative
3042
If a product requiring one day of in nature while tlie linear prograiiiiiiing
production on the bottleneck is moved (LP) ha^ a complicated iiiatheiiiatical Dr Mabel Qiu is with f k e Schod of Busi-
forward, the constraint may be eliini- approach. TOC was inipletnented in npss and Economics, Lorz~qwood Co//e$,
nated at the cost of carrying one days the five-step heuristic and was coni- Farnzuilk, VA 23909. Di Lawreme
invciitory. Fig. 1 shows the monthly pared with L1 which is kiiown to F:vedei/dall is iuiik t k e Dcpartrmwt qf
schedule for each of the four months. produce thc optimal solution. Both Marqemew t, Clemsor/ U/7ivevsity, C I P- ~
The plan shown, developed with the approaches were applied to the real S O M , SC 29634. Dr Zkiwei Zku ic
T O C heuristic, suggests that splits 1 , 7 operations data from a fibre manufac- witk the Dyartnient (rfUSAI; Uniuevsify
and 8 are not coiistraint. Consequently, turer (Figs. 1 and 2). It was surprising a ai Lafiycttc, L&)vtte,
inanagenient niay reduce the inventory that the schedules produced by the two LA 70504. T h y rtiay be corztacted by
levels for the products (5,25and 24) that approaches are so similar on this plants c-tmi/ir<y zxk u @lo CI isiarza. cdir
are loaded onto these splits, siiice thcre application. The value of this study is
is excess capacity on te splits, and less that TOC iiiipleiiieiited by the five-step
inventory is needed For protection. heuristic is much siiiipler to understand 0IEE 2002

MANUFACTURING ENGINEER AUGUST 2002 I95

Authorized licensed use limited to: SUNY Buffalo. Downloaded on February 1, 2009 at 03:00 from IEEE Xplore. Restrictions apply.

You might also like