Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

Int J Adv Manuf Technol (2009) 41:760–769

DOI 10.1007/s00170-008-1526-1

ORIGINAL ARTICLE

Taguchi-based Six Sigma approach to optimize plasma


cutting process: an industrial case study
Joseph C. Chen & Ye Li & Ronald A. Cox

Received: 19 September 2007 / Accepted: 10 April 2008 / Published online: 18 June 2008
# Springer-Verlag London Limited 2008

Abstract This case study outlines the use of Taguchi are smaller than those before Taguchi parameter design is
parameter design to optimize the roundness of holes made conducted.
by an aging plasma-cutting machine. An L9 array is used in
a Taguchi experiment design consisting of four controllable Keywords Taguchi method . Quality . Six Sigma .
factors, each with three levels. With two non-controllable Process optimization
factors included in the setting, we conduct 36 experiments,
compared to the 81 parameter combinations (four factors,
three levels or 34) required in a traditional DOE setting. 1 Introduction
Therefore, using the Taguchi method significantly reduces
the time and costs of a quality improvement process. Many small- and medium-sized industries have imple-
Conducted for two response variables—bevel magnitude mented the two most popular process improvement meth-
and the smallest diameter deviation of the hole—the odologies—lean manufacturing and Six Sigma—which
Taguchi experiments gave the optimal combination originated at Toyota and Motorola, respectively. Each
A1B2C1D3 (small for tip size, 93 in/min for feed rate, methodology has its unique structure and tools. The central
100 V for voltage, and 63A for amperage), which is focus of lean manufacturing is to provide value by
verified with a confirmation run of 30 work pieces. All 30 eliminating waste, which is defined as anything that is not
cuts meet the quality requirement for subsequent assembly. value-added from the customer’s perspective. The “seven
Furthermore, statistical analysis indicates that the mean deadly wastes,” as defined by this method, are over-
value and standard deviation of the confirmation run data production, inventory, waiting, movement, transportation,
defects, and over-processing. By continuously eliminating
these wastes, the customer receives a high value product.
J. C. Chen
Department of Agricultural and Biosystems Engineering, When the variation of a part or a service does not meet the
Iowa State University, specifications of the downstream internal and/or external
221 I. ED. II, customers, the methodology and tools of Six Sigma can be
Ames, IA 50011-3130, USA
implemented to improve the quality of the product or
Y. Li (*) service. The latest structure of Six Sigma is defined as the
Department of Industrial and Manufacturing Systems define-measure-analysis-implementation-control (DMAIC)
Engineering, Iowa State University, model. Each stage of the model provides tools for
2019 Black Engineering,
conducting Six Sigma quality improvements for any
Ames, IA 50011-2164, USA
e-mail: yeli@iastate.edu process or service.
Figure 1 shows a summary of tools used in each stage of
R. A. Cox Six Sigma management. In order to reduce the variation of
Center for Industrial Research and Service (CIRAS),
a process, many Six Sigma teams use design of experiments
Iowa State University,
Campus, 2272 Howe Hall, Suite 2620, (DOE) methodology to find solutions that will align
Ames, IA 50011-2272, USA products with customer expectations. DOE is a statistical
Int J Adv Manuf Technol (2009) 41:760–769 761

Phase 0: Define
• Scope and Boundary Phase I: Process Measurement
• Define Defects • Map Process and Identify Inputs and Outputs
• Team Charter • Cause and Effects Matrix
and Champion • Establish Measurement System Capability
• Estimated $ Impact • Establish Process Capability Baseline
• Leadership approval

Phase II: Process Analysis


• Complete FMEA
• Perform Multi-Vari Analysis
• Identify Potential Critical Inputs
• Develop Plan for Next Phase

Phase III: Process Improvement


• Verify Critical Inputs
• Optimize Critical Inputs via
Taguchi
Fig. 3 Plasma table

Phase IV: Process Control


• Implement Control Plan
• Verify Long Term Capability factor, three-level design, there would be 27 treatment
• Continuously Improve Process combinations, i.e., 33 or 3×3×3. In addition, each treatment
combination should be run twice to achieve a more reliable
Fig. 1 DMAIC process improvement methodology
statistical data analysis, bringing the total number of
experiments to 54 (27×2). If the cost of each experiment
tool that studies the relationship between independent is $100, the Six Sigma team will spend $5,400 to analyze
variables, the Xs (process variables), and their interactions a three-factor, three-level experiment design with two
on a dependent variable, the Y, which is considered the replicates.
critical-to-quality (CTQ) of the product. In many manufacturing processes, achieving a solution
The most commonly used DOE tool is the 2k factorial that meets customers’ specifications may require an
design, where k is the number of factors, each with two evaluation of five or six factors. Costs will increase
levels. Thus, in a three-factor design, there are eight accordingly. For example, assuming a base cost of $100
treatment combinations, i.e., 23 or 2×2×2. Unfortunately, per experiment, a six-factor, three-level experiment design
for most manufacturing processes, two levels of each factor with two replicates will cost $145,800 (3 6 or 729
may provide insufficient information about quality im- combinations×2=1,458). This is the cost of conducting
provement. Often, three levels for each factor are needed. DOE alone and does not include time spent on DOE, which
For example, when studying the feed rate of a turning may reduce the productivity of other jobs and delay
operation, three levels could be evaluated (e.g., 0.005, resolution of the quality problem. In addition, during this
0.010, and 0.015 in. per revolution). Thus, for a three- time, the process will produce more defects and wastes.
Based on the aforementioned analysis, resolving industrial
problems cost-effectively and in a timely manner requires a
Fig. 2 Electric switchboard

Fig. 4 Indicator light


762 Int J Adv Manuf Technol (2009) 41:760–769

Fig. 5 Illustration of bevel Bevel magnitude

more economical DOE approach. Taguchi parameter conventional approaches used in off-line quality control,
design, which is capable of providing the optimal solution Taguchi’s philosophy is based on the belief that once
with a reduced number of experiment runs, is one such quality is designed into both the product and process, only
approach. minimal inspection is necessary. Taguchi proposes a
One small-sized, Midwest manufacturer has an aging holistic view of quality related to cost, which extends the
plasma-cutting machine that produces defects and causes focus of quality beyond manufacturers at the time of
production delays. Though it is easier to replace older production by integrating the customer and society as a
equipment with new machines, most small manufacturers whole. Taguchi defines quality as the (minimum) loss
have limited capital. The manager of this company imparted by the product to society from the time the
approached the researchers for assistance in answering the product is shipped. This economic loss is associated with
following questions: losses due to rework, waste of resources during manufac-
turing, warranty costs, customer complaints and dissatis-
1. What is the optimal setting to produce the highest
faction, time and money spent by customers on failing
quality products?
products, and the eventual loss of market share. Taguchi
2. Can the optimal setting lower the defect rate but
methods provide an efficient and systematic way to
maintain the required productivity?
optimize designs for performance, quality, and cost. These
If the defect rate remains high after implementing methods have been used successfully in Japan and the
optimal settings, the manager will have the justification he United States in designing reliable, high quality products at
needs to invest in new equipment. low cost in such areas as automotives and consumer
To address this challenge, the researchers conducted a electronics.
Taguchi parameter design study, using the results of that Taguchi breaks down off-line quality control into three
study to make recommendations. The procedure and process stages, concept design, parameter design, and tolerance
is demonstrated in the four phases outlined in Fig. 1. design, which are summarized below:
– Concept design results in either a design concept or an
“up and limping” prototype. In the initial phase, more
2 Overview of Taguchi parameter design than one design concept, each with its own set of pros
and cons, can be presented. The ideal design concept
Taguchi parameter design is one of several methods will be the one that research shows best addresses
developed by Dr. Genichi Taguchi [1]. One of the customer needs and is inherently robust.

Fig. 6 Illustration of smallest


diameter deviation
D normal
Smallest diameter deviation: | Dsmallet - Dnormal |
Dsmallest : Smallest diameter
D smallest Dnormal : Nominal diameter
Continuous curve: Actual plasma-cut hole
Dashed curve: Nominal diameter hole, maximal roundness
Int J Adv Manuf Technol (2009) 41:760–769 763

Fig. 7 Fish bone diagram

– Parameter design is a critical production step. The optimal combinations of parameters for better performance.
nominal design features or selected process factor Ghani et al. [2] applied Taguchi parameter design to
levels are tested and the combination of product optimizing parameters for end milling process. Low
parameter levels or process operating levels least resultant cutting force and good surface finish were found
sensitive to changes in environmental conditions and with high cutting speed, low feed rate and low depth of cut.
other uncontrolled factors (noise) is determined. Zhang et al. [3] used a Taguchi parameter design
– Tolerance design is used to further reduce variation, if application to optimize surface quality in a CNC face
required, by tightening the tolerance of those factors milling operation, where the best surface roughness
shown to have a significant impact on variation. This (response) and signal-to-noise ratio were sought. Davim
stage utilizes loss function to determine whether and Reis [4] studied cutting parameters of composite
spending more money on materials and equipment will milling process using Taguchi-based experiments. Kirby
result in a better product, thus emphasizing the et al. [5] discussed the application of Taguchi parameter
Japanese philosophy of invest last not invest first. design to optimize turning operations for best surface
roughness. Palanikumar and Karthikeyan [6] used Taguchi
Taguchi parameter design is an experiment-based pro-
methods to conduct experiments in turning composite
cess that uses the following steps to identify settings of
material to achieve maximum material removal rate and
design parameters that maximize performance character-
minimum surface roughness. Taguchi parameter design has
istics (e.g., yield or productivity, etc.):
also been used to analyze optimal parameters for drilling
process [7, 8, 9, 10].
1. Identify initial and competing settings of design
Although many of the aforementioned projects were
parameters, as well as important noise factors and their
conducted primarily in laboratories, research can also be
ranges.
done in an industrial setting. The Iowa State University
2. Construct the design and noise matrices, and plan the
Center for Industrial Research and Service (CIRAS), an
parameter design experiments.
extension service unit of this Midwest land grant university,
3. Conduct the parameter design experiments and evaluate
the performance statistic for each test run of the design
matrix.
4. Use the values of the performance statistic to predict
new settings of the design matrix. Factor Units Level 1 Level 2 Level 3
5. Confirm that the new settings truly improve the A Tip Size - Small Medium High
performance statistic. B Feed Rate in/min 83 93 103
C Voltage volts (V) 100 105 110
Considering multiple factors simultaneously, the Taguchi D Amperage amperes (A) 43 53 63
parameter design method uses orthogonal experimental
combinations to shorten the product development cycle, Noise Low High
1 Air Pressure lbs/in2 (PSI) 45 60
which, in turn, saves time and money, Taguchi parameter 2 Pierce Time seconds (s) 0.70 1.40
design has been utilized in traditional manufacturing
processes such as milling, turning, and drilling to determine Fig. 8 Test parameters
764 Int J Adv Manuf Technol (2009) 41:760–769

Fig. 9 Taguchi experiment Noise Level Setting


table Noise 1 2 3 4
1-Air Pressure 45 45 60 60
2-Pierce Time 0.7 1.4 0.7 1.4
Control Factors and Levels
Run A B C D N1 N2 N3 N4 Avg Circularity σ σ2 η
1 1 1 1 1 y11 y12 y13 y14
2 1 2 2 2 y21 y22 y23 y24
3 1 3 3 3 y31 y32 y33 y34
4 2 1 2 3 y41 y42 y43 y44
5 2 2 3 1 y51 y52 y53 y54
6 2 3 1 2 y61 y62 y63 y64
7 3 1 3 2 y71 y72 y73 y74
8 3 2 1 3 y81 y82 y83 y84
9 3 3 2 1 y91 y92 y93 y94

has proposed this methodology to help industry to obtain the holes have poor circularity (roundness). Roundness is
optimum manufacturing processes to meet customer measured by finding the smallest diameter deviation, which
demands within a reasonable timeframe and budget. The is the difference of the smallest diameter of the actual hole
following sections will present the application of Taguchi from the nominal diameter (Fig. 6). Beveling and roundness
parameter design in helping a small business achieve the problems that prevent hardware from fitting properly into
Six Sigma paradigm. the switchboard is a defect that requires rework. To
minimize production costs due to this rework, a Taguchi
analysis was undertaken to determine the optimal setting to
3 Current problem with an older plasma-cutting produce holes with minimum bevel and minimum out-of-
machine round diameter. Ideally, all of the holes cut by the plasma
machine would have best round shape and enable the
The case study in this paper is based on a problem hardware to be inserted smoothly.
presented by an electrical manufacturing company (Com- To accomplish this project for Company A, a Six Sigma
pany A) located in Des Moines, Iowa. The major products team consisting of operators, engineers, researchers and a
of Company A are large electrical switchboards (Fig. 2). manager was formed, and the Taguchi parameter design
One step in the process of making the boards requires the procedure was applied.
use of a plasma table (Fig. 3) to cut holes for hardware such
as an indicator light (Fig. 4). The current problem of the
plasma cutting process is that some of the holes do not 4 Six Sigma improvement process
allow the fitting for the hardware. A close examination of
the holes reveals that two reasons may keep the hardware 4.1 Define
from passing through the holes. First, the plasma cutting
process is beveling the edge of the holes it creates (Fig. 5), 4.1.1 Factors and levels
which, in turn, obstructs the hardware. Meiji EMZ-5TR
zoom stereo microscope was used to measure the magni- Before the Six Sigma team could conduct the Taguchi
tude (unit: 0.001 inch) of the bevels for analysis. Second, parameter design experiment, they needed to thoroughly

Fig. 10 Data of bevel (unit: Noise Level Setting


0.001”) Noise 1 2 3 4
1-Air Pressure 45 45 60 60
2-Pierce Time 0.7 1.4 0.7 1.4 S/N
Control Factors and Levels Ratio
Run A B C D N1 N2 N3 N4 Avg Bevel σ σ2 η
1 1 1 1 1 39 31 22 42 33.5 8.96 80.33 -30.73
2 1 2 2 2 35 25 35 35 32.5 5.00 25.00 -30.31
3 1 3 3 3 55 19 32 37 35.75 14.91 222.25 -31.60
4 2 1 2 3 40 47 44 82 53.25 19.38 375.58 -34.94
5 2 2 3 1 40 52 62 87 60.25 19.97 398.92 -35.94
6 2 3 1 2 45 54 60 62 55.25 7.63 58.25 -34.91
7 3 1 3 2 52 48 49 49 49.5 1.73 3.00 -33.90
8 3 2 1 3 39 47 46 52 46 5.35 28.67 -33.30
9 3 3 2 1 56 51 40 45 48 6.98 48.67 -33.69
Int J Adv Manuf Technol (2009) 41:760–769 765

Avg Bevel
A B C D Optima combinations(Raw Data)
done by determining normal settings for each factor and
level 1 33.92 45.42 47.25 47.25 smaller the better then setting one level lower (level 1) and one level higher
level 2 56.25 46.25 44.58 45.75 A1, B1, C2, D3 (level 3), using the existing settings as level 2. For
level 3 47.83 46.33 48.50 45.00
uncontrolled factors (noise factors), two levels were chosen,
S/N Ratio low and high. The levels and units of each factor are shown
A B C D Optima combinations(S/N Ratio) in Fig. 8. These levels were set to determine if changes to
level 1 -30.9 -33.2 -33 -33.5 larger the better
level 2 -35.3 -33.2 -33 -33 A1, B2, C1, D2 the factors impacted hole quality, and, if so, how much. In
level 3 -33.6 -33.4 -33.8 -33.3 this way, the most significant factor was identified, and the
optimal level for each factor determined. For the uncon-
Fig. 11 Response of controllable factors to bevel and S/N ratio
trolled factors, the two levels helped determine if back-
ground noises produce a significant effect. This effect will
be further analyzed with T-tests.

understand the plasma-cutting process. A brainstorming 4.1.2 Experimental design


exercise that examined all of the factors of the process
provided the necessary information. Since these factors Using these four controllable factors and two uncontrolled
involve knowledge in different domains and could be either noise factors, we constructed an L9 Taguchi experiment
controllable or non-controllable, it was necessary to have table with the appropriate settings of each factor (Fig. 9).
all team members participate in the brainstorming. A For each controllable parameter setting combination, four
fishbone diagram based on this exercise revealed all runs were conducted, each under a different noise factor
possible causes of defective plasma-cut holes from the setting combination. Each trial run is represented as yij,
perspectives of method, material, machine, operator and where i ranges from 1 to 9, denoting controllable parameter
environment (Fig. 7). Through group discussion and setting for experimental run, and j from 1 to 4, denoting
ranking, four factors were identified as controllable factors noise factor setting. Thus, a total of 36 experimental runs
(voltage, feed rate, amperage, and tip size) and two as are conducted as a setting shown in Fig. 9. The average
uncontrollable noise factors (air pressure and pierce time). value, standard deviation, variance and signal-to-noise (S/
The Six Sigma team then determined ranges of the levels N) ratio of the four runs under the same controllable
to explore for each factor. For controllable factors, this was parameter setting were calculated. The same experimental

S/N Ratio S/N Ratio


Factor A: Tip Size Factor B: Feed Rate (ipm)
Ave Bevel Avg Bevel

-28 -31 50
sm md lg 60
-29 83 93 103
-32 49
-30
Avg Bevel

Avg Bevel
52
S/N Ratio

-31 -33 48
S/N Ratio

-32 44 -34 47
-33
-35 46
-34 36
-35 -36 45

-36 28 -37 44

S/N Ratio S/N Ratio


Factor C: Voltage (Volts) Avg Bevel
Factor D: Amperage (Amps) Avg Bevel

-32 100 105 110 54 -32 43 53 63 52

52 -33 50
-33
Avg Bevel
Avg Bevel

S/N Ratio

50
S/N Ratio

-34 -34 48
48
-35 -35 46
46

-36 -36 44
44

-37 42 -37 42

Fig. 12 Response graphs for bevel magnitude


766 Int J Adv Manuf Technol (2009) 41:760–769

Fig. 13 Data of smallest Noise Level Setting


diameter deviation (unit: inch) Noise 1 2 3 4
1-Air Pressure 45 45 60 60
2-Pierce Time 0.7 1.4 0.7 1.4 S/N
Control Factors and Levels Ratio
Avg
Run A B C D N1 N2 N3 N4 Bevel σ σ2 η
1 1 1 1 1 0.002 0.007 0.018 0.020 0.012 0.0087 0.000075 37.116
2 1 2 2 2 0.001 0.001 0.006 0.016 0.006 0.0071 0.000050 41.337
3 1 3 3 3 0.006 0.001 0.003 0.019 0.007 0.0081 0.000066 39.925
4 2 1 2 3 0.012 0.018 0.021 0.052 0.026 0.0179 0.003200 30.442
5 2 2 3 1 0.037 0.044 0.060 0.055 0.049 0.0104 0.000109 26.051
6 2 3 1 2 0.023 0.036 0.040 0.052 0.038 0.0120 0.000143 28.147
7 3 1 3 2 0.037 0.027 0.042 0.035 0.035 0.0062 0.000039 28.956
8 3 2 1 3 0.009 0.005 0.027 0.014 0.014 0.0096 0.000092 35.888
9 3 3 2 1 0.041 0.032 0.021 0.041 0.034 0.0095 0.000090 29.184

design table was used for two response variables, the bevel determine the optimal combination of levels of the
magnitude and the deviation of the smallest diameter of controllable factors that creates the minimum bevel.
plasma-cut holes. Figure 12 is the graphical depiction of the response and
S/N ratio for bevel magnitude. Since the quality character-
4.2 Measure and analysis istic for beveling is the-smaller-the-better, the optimal
setting combination for minimum bevel is A1-B1-C2-D3,
This section provides details of the Taguchi-based experi- which is interpreted as small tip size, a feed rate of 83 in/
ments conducted for bevel magnitude and smallest diameter min, a voltage of 105 V, and amperage of 63A. For the S/N
deviation. Data were collected and analyzed using the ratio, we are looking for the largest value. Therefore, the
experimental design described in the previous section. optimal setting combination for S/N ratio response is A1-
B2-C1-D2, which means small tip size, a feed rate of 93 in/
4.2.1 Taguchi experiment for bevel min, a voltage of 100 V, and amperage of 53A. The arrows
in Fig. 12 indicate the chosen level of each factor, and the
Figure 10 shows the experiment data for the bevel size. The directions of the arrows show the quality characteristic with
response of each controllable factor on bevel magnitude upward denoting the-larger-the-better and downward the-
and the signal-to-noise ratio are shown in Fig. 11. The smaller-the-better.
quality characteristic is the-smaller-the-better; the formula
to calculate S/N ratio is given below and its derivation can 4.2.2 Taguchi experiment for smallest diameter deviation
be found in [1]:
Similarly, Fig. 13 shows the experiment data for the
smallest diameter deviation. The response of each control-
" !# lable factor on the smallest diameter deviation and the S/N
1 X4
h ¼ 10 log y2 ratio are shown in Fig. 14. Figure 15 is the graphic display
n j¼1 ij of the response to smallest diameter deviation. The quality
characteristic is also the-smaller-the-better for the smallest
diameter deviation. The optimal setting combination for
where, minimum smallest diameter deviation is A1-B2-C1-D3,
which means small tip size, a feed rate of 93 in/min, a
η S/N ratio,
yij individual response for each trial run,
n the number of runs due to noise factors; in this case,
n=4.
The bevel response table (Fig. 11) shows the mean
response of the variable (bevel) from each controllable
factor at each level in the Taguchi experimental design.
Each level of the controllable factors has three response
values in the orthogonal array (Fig. 10) and is calculated by
averaging these three values. The response table for the S/N
ratio (also in Fig. 11) was similarly obtained, using the S/N
ratios in the orthogonal array. With the response values and Fig. 14 Response of controllable factors to smallest diameter
the S/N ratio values, the response table can be used to deviation
Int J Adv Manuf Technol (2009) 41:760–769 767

S/N Ratio S/N Ratio


Factor A: Tip Size Deviation
Factor B: Feed Rate (ipm) Deviation
40 0.080
50 0.050 0.075
48 38 0.070
46 0.065
0.000 36 0.060
44

Deviation
42 0.055
34 0.050

Deviation

S/N Ratio
40
S/N Ratio

-0.050 0.045
38 32 0.040
36 0.035
34 -0.100 30 0.030
32 0.025
28 0.020
30
-0.150 0.015
28 26 0.010
26 0.005
24 -0.200 24 0.000
sm md lg 83 93 103

S/N Ratio S/N Ratio


Factor C: Voltage (Volts) Deviation Factor D: Amperage (Amps) Deviation
50 0.040 60 0.040
48 58
56
46 0.030 54 0.030
44 52
50
42 0.020 48 0.020

Deviation

Deviation
S/N Ratio
S/N Ratio

40 46
38 44
0.010 42 0.010
36 40
34 38
32 0.000 36 0.000
34
30 32
28 --0.010 30 -0.010
26 28
26
24 --0.020 24 -0.020
100 105 110 43 53 63

Fig. 15 Response graph for smallest diameter deviation

voltage of 100 V, and an amperage of 63A. The S/N ratio three times, while level 2 occurred once. Therefore, level 3
responses render the same parameter setting combination. for factor D is chosen as the overall optimal setting
combination. With this rule, the overall optimal setting for
4.2.3 Setting selection the plasma table is A1-B2-C1-D3, as indicated in the bottom
row of Table 1. The setting A1-B2-C1-D3 means the small
According to Figs. 12 and 14, bevel and smallest diameter tip size, a feed rate of 93 in/min, a voltage of 100 V, and
deviation as well as their S/N ratios have different optimal amperage of 63A.
setting combinations. These are summarized in the first four
rows of Table 1. For example, bevel S/N ratio response 4.2.4 T-test
gave level 2 (53A) for factor D (amperage), while all other
responses (bevel, smallest diameter deviation and its S/N To examine the effect of noise factors on the response
ratio) chose level 3(63A). However, Company A needs variables, a t-test was conducted for each noise factor. For
only one overall optimal setting combination for its plasma
table. Therefore, the appropriate optimal setting combina-
Table 2 T-test for air pressures effect on smallest diameter deviation
tion is the level of each factor having the largest number of
occurrences. For example, level 3 of factor D occurred Air pressure (PSI) 45 60

Mean 0.019 0.030


Table 1 Optimal setting combination Variance 2.4791 e-4 3.0446 e-4
Observations 18 18
Selection Criteria Combination
df 34
Bevel A1 B1 C2 D3 Difference in mean 0.011
Bevel S/N ratio A1 B2 C1 D2 Std. Error (difference between the means) 0.0055
Smallest diameter deviation A1 B2 C1 D3 t Stat 2.0358
Smallest diameter deviation S/N ratio A1 B2 C1 D3 t Critical two-tail (alpha=0.005) 2.7284
Overall optimal setting A1 B2 C1 D3 P(T <=t) one-tail 0.0496
768 Int J Adv Manuf Technol (2009) 41:760–769

Table 3 T-test for pierce times effect on smallest diameter deviation Table 5 T-test for effect of pierce time on bevel

Pierce time (s) 0.7 1.4 Pierce time (s) 0.7 1.4

Mean 0.023 0.026 Mean 43.944 48.056


Variance 2.94 e-4 3.18 e-4 Variance 107.4673 290.9967
Observations 18 18 Observations 18 18

Df 34 Df 34
Difference in mean 0.003 Difference in mean 4.112
Std. Error (difference between the means) 0.0058 Std. Error (difference between the means) 4.7050
t Stat 0.6575 t Stat −0.8738
t Critical one-tail (alpha=0.005) 2.7284 t Critical one-tail (alpha=0.005) 2.7284
P(T <=t) one-tail 0.5153 P(T <=t) one-tail 0.3884

air pressure, a t-test determines if the two levels (45 psi and or equivalently P (T <= t) one-tail=0.2139 > alpha=0.005.
60 psi) will have a significant effect on smallest diameter Therefore, the air pressure does not significantly affect the
deviation. The hypothesis is stated as bevel magnitude. From Table 5, Abs (t Stat)=0.8738 < t
critical two-tail (alpha=0.005)=2.7284, or equivalently P
(T <= t) one-tail=0.3884 > alpha=0.005. Therefore, the
H0 : μ45 ¼ μ60 pierce time does not significantly affect the small radius.
H1 : μ45 6¼ μ60
4.3 Implementation
The t-test result is shown in Table 2. From the test result
it can be seen that the Abs (t Stat)=2.0358 < t critical two- After the overall optimal setting combination for the plasma
tail (alpha=0.005)=2.7284, or equivalently P (T <= t) one- cutter was identified, a confirmation run was conducted.
tail=0.0496 > alpha=0.005. The statistical conclusion is With this optimal setting (A1-B2-C1-D3), 30 cuts were made
that the difference in mean from the two air pressure levels to test the smallest diameter deviation and the bevel
is not significant, so the hypothesis Ho can not be rejected. magnitude. The hardware passed easily through all 30 cuts,
Therefore, we can conclude that air pressure does not meaning no rework was needed. In addition, as the results
significantly affect the smallest diameter deviation. shown in Table 6 indicate, the mean value and standard
Similarly, three t-tests were conducted for the effects of deviation of the confirmation run data were smaller than
pierce time to smallest diameter deviation (Table 3), air those before Taguchi design was conducted.
pressure to bevel (Table 4), and pierce time to bevel
(Table 5). From Table 3, it can be seen that the Abs 4.4 Control
(t Stat)=0.6575 < t critical two-tail (alpha=0.005)=2.7284,
or equivalently P (T <= t) one-tail=0.5153 > alpha=0.005. The optimal setting combination was sent to Company A’s
Therefore, pierce time does not significantly affect the production department. The Six Sigma team is now trying
small radius. From Table 4, it can be seen that the Abs to uncover other possible causes of unacceptable deviation
(t Stat)=1.2667 < t critical two-tail (alpha=0.005)=2.7284, so that other Taguchi experiments can be conducted for
continuous improvement of the process. For example, if
defects occur later following the optimal condition, the Six
Table 4 T-test for effect of air pressure on bevel
Sigma team will follow the DMAIC procedure (Fig. 1) to
Air pressure (PSI) 45 60 pursue the next cycle of process improvement. Taguchi

Mean 43.056 48.944


Variance 110.5 278.5 Table 6 Confirmation run data
Observations 18 18
Before Taguchi Conformation
Df 34 design run
Difference in mean 5.888
Std. Error (difference between the means) 4.6491 Smallest diameter μ=0.024 μ=0.006
t Stat −1.2667 deviation б=0.0173 б=0.0032
t Critical two-tail (alpha=0.005) 2.7284 Bevel μ=46 μ=35.2
P(T <=t) one-tail 0.2139 б=14.0672 б=4.5166
Int J Adv Manuf Technol (2009) 41:760–769 769

parameter design for optimal condition could then be Acknowledgement This project was partially funded by the Iowa
Center for Industrial Research and Service, through a grant from the
executed again.
Department of Commerce NIST Manufacturing Extension Partnership.

References
5 Conclusion
1. Fowlkes WY, Clyde CM (1995) Engineering methods for robust
Six Sigma and lean manufacturing are powerful strategies product design using Taguchi methods in technology and product
for transforming an enterprise and escalating its competi- development. Addison-Wesley, Reading
2. Ghani JA, Choudhury IA, Hassan HH (2004) Application of
tiveness. The fact that Six Sigma and lean manufacturing
Taguchi method in the optimization of end milling parameters. J
can successfully save time and cut costs are important Mater Process Technol 145(1):84–92 Jan 1
considerations, especially for small- and medium-sized 3. Zhang JZ, Chen JC, Kirby ED (2007) Surface roughness
enterprises with limited resources. This paper presented optimization in an end-milling operation using the Taguchi design
method. J Mater Process Technol 184(1–3):233–239 Apr 12
the application of the Taguchi method to optimize the
4. Davim JP, Reis P (2005) Damage and dimensional precision on
roundness of the holes cut by an aging plasma-cutting milling carbon fiber-reinforced plastics using design experiments.
machine. Using the orthogonal array in the experiment J Mater Process Technol 160(2):160–167 Mar 20
design for four factors, the Taguchi method reduced the 5. Kirby ED, Zhang Z, Chen JC, Chen J (2006) Optimizing surface
finish in a turning operation using the Taguchi parameter design
experiment set-up from 81 parameter combination settings
method. Int J Adv Manuf Technol 30(11–12):1021–1029 October
(34) in DOE to an L9 setting. With two noise factors 6. Palanikumar K, Karthikeyan R (2006) Optimal machining con-
included in the Taguchi experiment design, 36 total experi- ditions for turning of particulate metal matrix composites using
ments are conducted. The optimal setting combination Taguchi and response surface methodologies. Mach Sci Technol
10(4):417–433 Dec 1
received from Taguchi experiment design is A1B2C1D3
7. Deng C-S, Chin J-H (2005) Hole roundness in deep-hole drilling
(small for tip size, 93 in/min for feed rate, 100 V for voltage, as analysed by Taguchi methods. Int J Adv Manuf Technol 25(5–
and 63A for amperage), which maintains the existing feed 6):420–426 March
rate for productivity and improves quality of products. The 8. Tsao CC (2007) Taguchi analysis of drilling quality associated
with core drill in drilling of composite material. Int J Adv Manuf
optimal setting combination gave no defects from the 30
Technol 32(9–10):877–884 April
plasma-cut holes in the confirmation run. In addition, the 9. Bagci E, Ozcelik B (2006) Analysis of temperature changes on the
recommended setting combination was well received by twist drill under different drilling conditions based on Taguchi
Company A, and the problem with defects situation has been method during dry drilling of Al 7075-T651. Int J Adv Manuf
Technol 29(7–8):629–636 July
much improved. With the reduced time and cost, Taguchi
10. Tsao CC, Hocheng H (2004) Taguchi analysis of delamination
experiment design has again demonstrated its effectiveness associated with various drill bits in drilling of composite material.
in achieving Six Sigma and lean paradigm. Int J Mach Tools Manuf 44(10):1085–1090 August

You might also like