Академический Документы
Профессиональный Документы
Культура Документы
Professor and Head, Department of Statistics, Institute of Social Science, B. R .Ambedkar University
Agra, Uttar Pradesh, India
3
ABSTRACT
This research paper focuses on using confounding an efficient tool of factorial experiment of design of
experiments (DoE) to optimize the facility layout in supply chain management. The critical factors that impact the
performance metrics of cycle time, work-in-process and throughput can be identified by using confounding in factorial
experiment. These experiments can be conducted in real world as well as, simulation models, which are used to evaluate
the various what-if scenarios.
editor@tjprc.org
26
observations obtained from such a controlled experiment could then be used to verify the hypothesis of direct
proportionality.Any disagreement between the observations and the hypothesis would suggest either that the hypothesis
should be discarded or some other assignable causes have to be looked for. A modified hypothesis can be set up, these
modification may be the inclusion of few more factors to be controlled or, a change in the functional relationship assumed
and the experiment is re-planned. The cycle of hypothesis, experiment, observation, analysis is repeated till a satisfactory
conclusion is reached, satisfactory does not mean any absolute sense but only relative to the standards set for the purposeof
the experiment.
In supply chain system experiment, the difference in performance of the companies depends upon location and
distance of inventory of raw material, finished product, repair centers, warehouse, transport destination, mode of transport,
route of transport, location of factories etc., these are considered as factors. Other than these factors there are cost of
inventory management, shutdown time of machinery, arrival and cost of outsourcing part, etc. Time required to shift the
lots from one set machineries to other, packaging, distances between operations, batch sizes, frequency of transactions,
resource utilizations etc., were studied closely for the feasibility of the re-layout scenarios. The performance measures and
results from the simulation models were then used to develop the layout for the server assembly process.
So there are n numbers of controlled and uncontrolled factors effecting the supply chain management of a
particular system. Now if anyone wants to apply any of the efficient and effective technique like six sigma, TQC, Lean,
etc., to improve the performance of the system. It is required to performed lots of experiments with different sets of
controlled and uncontrolled factors within and between the companies in the virtual models that is stimulation models.
Now question arise on what basis these experiments should be designed. Designs of an experiment means:
The specification of the experiment units to which the factors are to be applied.
The rules according to which the factors are allocated to the experimental units.
The specification of the measurements, or other records, to be made on each experiment unit
27
customers they are serving. To analyze such situations, confounding is best tool of factorial experiments which can be
used. The effects of less importance may be confounded so that the important effects can be estimated more efficiently.
Before using this concept first let us see what is factorial experiment? What does main effects, interaction effects and
confounding means?
interested. Then we have to estimate (i) the effects of each of the factors and, (ii) how the effect of one factors varies over
the different levels of other factors. To study these effects we investigate all possible replicate of the experiment.
treatment combinations (or composite treatments) to be assigned to the different experiment units.
A factorial experiment involves simultaneously more than one factor each at two or more levels, if the number of
levels of each factor in the experiment is same, the experiment is called symmetrical factorial experiment, and otherwise it
is called asymmetrical factorial or sometimes mixed factorial. The symmetrical experiments provide an opportunity to
study not only the individual effects of each factor but also their interactions. In general a factorial experiment can be
represented as
which represents each of the m at f levels. The simplest class of factorial designs is that involving
, these form the following six combinations taking one level from each factors in
The comparison required in this type of experiments are not the pair comparison, but a special type of comparison
+
called main effect and interaction effects. The comparison between the totals
+
are called main effects of Mode of Transport. Similarly, two independent comparisons among the totals,
respectively.Comparison among these three contrasts indicate the equality or otherwise the effect of Mode
Similarly
two
more
effects
are
there
for
and
which
are
of Transport by three different route of transport. Contrast representing such comparisons, that is(
(
), (
)(
) and (
)(
and
)
effects.
So this is an example of factorial experiment with two mode of transport and three route of transport of the raw
material or finished products. The method is illustrated with the help of following example.
Example
Let us consider a 24 factorial experiment where there are four factors effecting supply chains, each of these factors
are at two levels 0 and 1 (levels may be the types of the factors), the factors are Location of the factory denoted
www.tjprc.org
editor@tjprc.org
28
by
,
. Consider the experiment is conducted in RBD (Randomized Block Designs) with r replicates. Now
represents that location of the factory is at level one and rest of the
represents location of the factory and transport destination are at level zero but
mode of transport and route of transport are at level 1, and so on. Therefore the16 combinations are as follows:
= (1),
= ,
= ,
= ,
= ,
, ,
,
,
, ,
.
,
CD: 0011
BC: 0110
ACD: 1011
BCD: 0111
ABCD: 1111
Since in this example we are considering factorial experiment is conducted in RBDwith r replicates, therefore let
!
.!
01
23
29
.!
01
23
SSA represent sum of squares of main effect A, SSAB represents sum of Squares of interaction effect AB,,
SSABCD represents sum of square of interaction effect ABCD, these sum of squares for the main and interaction effects
are calculated by Yates method.
Sum of square of main effect A (SSA) =
[891
3. :
[8;91
3. :
[8;<91
3. :
Similarly, sum of squares of other main and interaction effects can be calculated. Now sum of square of error
//= = .// (// 45 67 + // + // + //
//
+ //
+ //
+ //
+ //
+ // + //
+ //
+ //
+ // + //
Analysis of variance (ANOVA) table for a 2 factorial experiment in RBD with r replicates is:
Table 1: Analysis of Variance (ANOVA) Table for A >? Factorial Experiment in RBD With R Replicates
Source of Variation
Degree of Freedom
Blocks
Main effect A
Main effect B
Main effect C
Main effect D
Interaction effect AB
Interaction effect AC
Interaction effect AD
Interaction effect BC
Interaction effect BD
Interaction effect CD
Interaction effect ABC
Interaction effect ABD
Interaction effect ACD
Interaction effect BCD
Interaction effect ABCD
Error
Total
r1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
15(r 1)
16 r 1
Sum of
Squares (SS)
SS Blocks
SSA
SSB
SSC
SSD
SSAB
SSAC
SSAD
SSBC
SSBD
SSCD
SSABC
SSABD
SSACD
SSBCD
SSABCD
SSE
TSS
ABB
Mean Sum of @ =
ABBC
Squares (MSS)
MSS Blocks MSS Blocks/ MSSE
MSSA MSSA/ MSSE
MSSB MSSB/ MSSE
MSSC MSSC/ MSSE
MSSD MSSD/ MSSE
MSSAB MSSAB/ MSSE
MSSAC MSSAC/ MSSE
MSSAD MSSAD/ MSSE
MSSBC MSSBC/ MSSE
MSSBD MSSBD/ MSSE
MSSCD MSSCD/ MSSE
MSSABC MSSABC/ MSSE
MSSABD MSSABD/ MSSE
MSSACD MSSACD/ MSSE
MSSBCD MSSBCD/ MSSE
MSSABCD MSSABCD/MSSE
MSSE -
www.tjprc.org
editor@tjprc.org
30
reduced block size. This device of reducing the block size by making one or more interaction contrasts identical with block
contrasts, is known as confounding.
When there are two or more replications, if same set of interaction is confounded from all the replications than
confounding is said to be complete and if different sets of interaction effects are confounded form the different replications
than it is said to be partial confounding. Both the methods are in practice. In complete confounding all the confounded
interactions are completely lost, though in partial confounding the confounded interaction can be recovered from those
replications in which they are not confounded.
Confounding in a 2 factorial without confounding we need blocks have size 2 . If any interaction or main effect is
D
of which r blocks get one set of treatments and the remaining r blocks get the other set. The
minimum number of blocks required is 2 when each treatment combination occurs once only. In both cases with or without
replication the arrangement is called a confounding arrangement in two blocks. This technique of Confounding could be
used to reduce the block size in the supply chain management system as well.
DF
effects and interactions are obtained as if there is no confounding. Of these we retain only the un-confounded main effects
and interaction effects making a total of (2 1) (2F 1) = 2 2F degree of freedoms. The sum of square of due to
the blocks is obtained -. 2F 1 degrees of freedom. The total of these two subtracted from the total sum of square
with-. 2 1 degrees of freedom gives the error sum of square with (- 1)(2 2F ) degrees of freedom.
In the above example, if we confound the interaction effect ABCD, we will get the two groups of eight treatments
each for2 factorial experiment from the solution of the equation
+
0
=H J
1
Table 2: Key Blocks of >? Factorial Experiment Confoundedwith Interaction Effect ABCD
Block 1
0000
0011
0101
0110
1001
1010
1100
1111
Block2
0001
0010
0100
0111
1000
1011
1101
1110
Block 1 is called principal or key blocks. This block is used further to create the whole design. Here we have
confounded one interaction, the experiment being carried out in 2- blocks of size 8 each. The sum of squares for the main
effects and interactions are obtained as if there is no confounding. Of these we retain only the un-confounded main effects
and interaction effects making a total of 14 degree of freedoms. The sum of square of due to the blocks is obtained with
2- 1 degrees of freedom.The total sum of square with 16- 1 degrees of freedom is obtained in the same manner.
Impact Factor (JCC): 5.7836
31
The error sum of square with 14(- 1) degrees of freedom is obtained by subtraction.
Analysis of variance (ANOVA) table for confounded arrangement of 2 factorial experiment in RBD withr
replicates in which ABCD interaction is confounded is as follows:
Table 3: Analysis of Variance (ANOVA) Table for A >? Factorial Experiment in
RBD with R Replicates Confounded with Interaction Effect ABCD
Source of Variation
Degree of Freedom
Blocks
Main effect A
Main effect B
Main effect C
Main effect D
Interaction effect AB
Interaction effect AC
Interaction effect AD
Interaction effect BC
Interaction effect BD
Interaction effect CD
Interaction effect ABC
Interaction effect ABD
Interaction effect ACD
Interaction effect BCD
Error
Total
2r 1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
14(r 1)
16r 1
Sum of
Squares (SS)
SS Blocks
SSA
SSB
SSC
SSD
SSAB
SSAC
SSAD
SSBC
SSBD
SSCD
SSABC
SSABD
SSACD
SSBCD
SSE
TSS
ABB
Mean Sum of @ =
ABBC
Squares (MSS)
MSS Blocks MSS Blocks/ MSSE
MSSA MSSA/ MSSE
MSSB MSSB/ MSSE
MSSC MSSC/ MSSE
MSSD MSSD/ MSSE
MSSAB MSSAB/ MSSE
MSSAC MSSAC/ MSSE
MSSAD MSSAD/ MSSE
MSSBC MSSBC/ MSSE
MSSBD MSSBD/ MSSE
MSSCD MSSCD/ MSSE
MSSABC MSSABC/ MSSE
MSSABD MSSABD/ MSSE
MSSACD MSSACD/ MSSE
MSSBCD MSSBCD/ MSSE
MSSE -
CONCLUSIONS
Factorial designs are widely used in experiments involving several factors where it is necessary to study the effect
of the factor on the response. When the factors are independent, then the factorial experiments give a broad picture of the
effect of one factor over varying levels of the other factors. They allow effects of a factor to be estimated at several levels
of the others, yielding conclusions that are valid over a range of experimental conditions. With m factors we shall need m
simple experiment but the same can be done with only one complex experiment that is factorial experiment. Thus it is
economic from time, labor and material point of view, they are more efficient than one factor at a time experiments. In
these designs individual factorial effect is estimated with precision, a whole of the experiment is devoted to it. To avoid
misleading conclusions, a factorial design is necessary when interactions are present. In fact, with independent factors
factorial experiments are the only means to provide estimates for interactions. These designs form the basis of the other
designs of considerable practical value. An additional advantage of these designs is that we can get a test for the additively
assumption from the experimental data of the factorial experiment. Confounding could be used to reduce the block size.
Factorial experiments arewidely used in research work. Also, in supply chain management on stimulation models as well
as real world situations, factorial experiment with confounding could be used for the study of different combinations of
variables and attributes at nlevels.
REFERENCES
1.
Rangaraj. N, Raghuram. G, Srinivasan. M. M, (2009); Supply Chain Management for Competitive Advantage;
Tata McGraw- Hill Publishing Company Limited; ISBN No. 978-0-07-022163-5.
www.tjprc.org
editor@tjprc.org
32
2.
Jain. K, Singh. V, (2011) The Six Sigma Management- Customers and Organizations Delight; Humanities and
Social Sciences: Interdisciplinary Approach; 2011, Volume: 03, Issue: 02, Page No.125-129, ISSN No.09757090.
3.
Chaudhuri. D, (2012) Comparative study for Lean and Six Sigma for organizational improvement; International
Global Research Analysis; 2012, Volume: 1, Issue: 4, Page no. -28-29, ISSN No. 2277-8160.
4.
The RIGHT source for Six Sigma learning and application. [on-line] Available on URL: >http://benchmark.groupsite.com/discussion< [cit.: 2011-01-19].
5.
Efektvnyprstup k zlepovaniuprocesov. FBE Bratislava. [on-line] Available on URL: > http://www.sixsigma.sk/< [cit.: 2011-01-23].
6.
The
Wall
Street
Journal;
[On-line]
Available
on
URL:>http://blogs.wsj.com/corporate-
intelligence/2013/12/18/factories-look-for-supply-chain-risks-in-all-the-wrong-places/<[cit.: 2013-12-18].
7.
Dass. M.N, Giri. N.C, (2002); Design and Analysis of Experiments; New age International (P) Limited,
Publishers; Second Edition, Page No. 98-159, ISBN No. 0-852-266-914-5.
8.
Joshi. D.D, (2003);Linear Estimation and Design of Experiments; New Age International (P) Limited, Publishers;
page no. 183-251; ISBN No. 0-852-26517-4.