You are on page 1of 18

SSFM-1116/Feature/me

CASE STUDY

Breakthrough Solution
DMAIC guides glass supplier in saving money and cutting down defects

by Krishan Kumar, Maruti Center for Excellence, Krishen Kumar, Asahi India Glass Ltd.,
and Sumit Yadav, Maruti Center for Excellence

Asahi India Glass Co. is a tier-one auto glass supplier for Maruti Suzuki, the largest car
manufacturer in India. To optimize process yield, the glass supplier used the define, measure,
analyze, improve and control (DMAIC) method to identify and address major pain points
within its operationsparticularly, the ME3 back-door glass models manufacturing process
that had a process yield of just 86.5%well below expectations.
The manufacturing process under scrutiny consists of three subprocesses. The
DMAIC approach was used on all three to understand the significant input variables and
optimize their settings using design of experiments (DoE).
The glass suppliers actions resulted in breakthrough improvement, and the process
yield reached 96.3% by the end of the project. The project also aimed to sustain the results,
which was well maintained for the last year.

The glass makers improvement approach can be easily used in any industry because
process yield is a function of input factors [Y = f(x)]. Optimizing these factors will
undoubtedly result in process yield improvement.

Sidebar: Approaches to Consider


Within any industry, the ever-changing, competitive marketplace requires that products be
made of the highest quality and at the lowest cost. Each and every organization must strive to
meet these important requirements.
The only way to turn a profit in such a situation is to cut down manufacturing costs by
reducing defects in a process. This is particularly true when the original equipment
manufacturers (OEM) demand reductions year after year on the price of the components from
their vendors.
As far as quality of products is concerned, OEMs also demand a detailed analysis of
countermeasures (A3 report) on defects observed in the suppliers products on its production
line or from the customers perspective. The A3 report is a powerful tool that establishes a
concrete structure to implement plan-do-check-act (PDCA) thinking to provide insight and
address the problems.1 This tool was initially developed by Toyota and is the core of the
Toyota Production System.2
Quality methods
Although there are many process improvement methods using statistics, lean Six Sigma has
become one of the cost-effective approaches.3 Six Sigma differs from conventional problem
solving because it requires proof of cause and effect before improvement action is taken
because resources for improvement actions are limited in most organizations.4 Six Sigma was

conceived and developed at Motorola in the early 1980s after Motorola realized that it was
spending 5 to 10% of its annual revenues in correcting poor quality.5
Lean Six Sigma is a process-focused approach used to fundamentally change the
method of working within an organization. Mikel Harry has said that organizations operating
at a three sigma level that marshal all their resources around Six Sigma can expect to make
one sigma shift improvement each year.6 Accordingly, organizations will experience:

A 20% margin improvement.

A 12 to 18% increase in capacity.

A 12% reduction in the number of employees.

A 10 to 30% capital reduction.


This can be a difficult target when you consider that one sigma shift in improvement

will be required every year until the target of Six Sigma is achieved. Six Sigma is bound to
improve the organizations performance levels in terms of yield, productivity and customer
performance, provided there is top management commitment and a champion leads each
project from start to finish.
Considering that Six Sigma focuses on improvement project by project, the authors
felt that the Six Sigma approach under the prevailing situation at the glass maker was likely
to give better results. Six Sigma seeks to improve the quality of process outputs by leveraging
the important inputs and minimizing variability in manufacturing and business processes.7 Six
Sigma is a business improvement strategy that focuses on improving products, processes and
profits. It enables an organization to improve performance by eliminating deficient processes
and defects in product and services.8 Various statistical toolssuch as hypothesis testing and
statistical evidences based on datahave been used for process improvement.

As noted earlier, this study portrays a simple approach that can be used across
industriesnot just auto glass manufacturersto improve process yield with minimal efforts
by optimizing input factors affecting the process. K.K., K.K and S.Y.
REFERENCES
1.

Durward K. Sobek II and Art Smalley, Understanding A3 Thinking, Productivity Press,


2008.

2.

John Shook, Managing to Learn, Lean Enterprise Institute, July 2010.

3.

Vijaya Sunder and Jiju Antony, Offshore Output, Six Sigma Forum Magazine, May
2015, pp. 8-20.

4.

Chew Jian Chieh, Six Sigma Basics: DMAIC Like Normal Problem Solving, Feb.
26, 2010, iSixSigma, www.isixsigma.com/new-to-six-sigma/dmaic/six-sigma-basicsdmaic-normal-problem-solving.

5.

Mikel J. Harry and Richard Schroeder, Six Sigma: The Breakthrough Management
Strategy Revolutionizing the Worlds Top Corporations, Doubleday, 2000.

6.

Ibid.

7.

Mikel J. Harry, Prem S. Mann, Ofelia De Hodgins, Chris Lacke and Richard Hulbert,
Practitioners Guide for Statistics and Lean Six Sigma for Process Improvements,
Wiley, 2010.

8.

Joseph De Feo and William Barnard, Juran Institutes Six Sigma Breakthrough and
Beyond: Quality Performance Breakthrough Methods, McGraw-Hill Professional,
2004.

Process flow of glass manufacturing


The three subprocesses that make up the overall glass manufacturing process are shown in
Figure 1 and are as follows:

1. Pre-process: cutting the glass to the desired shape, grinding the edges for
smoothness, washing, drying and inspecting.
2. Printing process: printing the glass surface (edges) with black ceramic ink and
passing the glass through heating zones where hot air is blown to dry the ink. Using a
sealant, the back-door glass is stuck to the vehicle body. The ink is used to protect the
sealant from damage from ultraviolet rays.
3. Bending and tempering process: heating the glass to specified temperature levels,
bending the glass to the desired shape and quenching the glass with high pressure air.
In this subprocess, the glass is heated to 680C to 720C and quenched with high air
pressure of 1,800mm of aqua. Glass is cooled, inspected and packed for dispatch.
Define phase
Asahi India purchased a Victory Technology furnace used for bending and tempering process
from a glass equipment manufacturer in 2011 in Spain. The decision was based on a
recommendation from its parent company and information that the furnace could be used for
producing both side door and back-door glass. Previously, Asahi India used multiple furnaces
that produce either side-door glasses or back-door glasses.
Unfortunately, the trials at the manufacturer could not be completed successfully for
both types of glass due to time constraints and pressure from the OEMs for more glass supply
in India. Subsequent use of this furnace in India indicated that although it was possible to
achieve a satisfactory level of yield and productivity for the side-door glasses, there was still
a big gap and fall-off in yield and productivity for the back-door glass.
Problem statement: The overall process yield for the back-door glass was 90.2%
from April 2013 to February 2014 against the budgeted target of 93.5%. This not only caused

financial losses because overall yield targets fell short of expectations, but there also were
problems in obtaining enough good glass from this furnace to meet customer requirement.
Model selection (among back-door glass) using a scoping tree is shown in Figure 2.
Among the back-door glasses, the ME3 back-door glass to be installed in the Maruti Suzuki
Alto 800 car was selected because it was the worst-performing model of glass (with lowest
process yield). The process yield of the ME3 model of back-door glass was 86.5%. Process
yield broken down by subprocess were as follows:
1. Pre-process = 98.5%.
2. Printing process = 98.5%.
3. Bending and tempering process = 89.5%.
Rolled-throughput yield (RTY) = 86.8% ()

The defects contributing to the rejections in these three subprocesses are shown in
Figure 2. A cross-functional team, which included subject matter experts (SME), was formed
and aggressive targets were set to improve the three subprocesses to the following target
levels:
1. Pre-process yield = 99.5%.
2. Printing process yield = 99.5%.
3. Bending and tempering process yield = 93.5%.
The targets were based on benchmark figures for similar types of equipment at the
organizations parent company in Japan. From there, it would be possible to achieve 99.5%
yield for the pre-process and printing process yields.
As far as the bending and tempering process, the yield of a fully mechanized furnace
in Japan was 97.5%. Considering that the furnace under study was not fully mechanized and

there was no previous experience using this type of furnace at Asahi India, the target for the
bending and tempering process yield was set as 93.5%.
The defects (all were attribute type) that needed to be reduced were selected for all
three subprocesses, as shown in Table 1. The progress of the project was reviewed with the
companys vice president of manufacturing every two weeks.
After setting the targets and selecting the major defects, it was required to map all
three subprocesses using a process flow diagram (PFD). A detailed analysis also was
performed to understand the hidden processes at each subprocess level.
The PFD of the bending and tempering process is shown in Figure 3. This shows that
there were 16 process steps in this processseven of which were value-added (VA), eight
were nonvalue-add (NVA) but necessary for the process (muda type one), and one step was
NVA (muda type two) and waste.1
After defining the problem in DMAIC process, of course the measure phase follows.
In fact, during the initial stages when the problem is being defined, you can alternate between
the define and measure phases until the problem and targets are clearly defined.
Measure phase
In the measure phase, it is necessary to check the accuracy and precision of the measurement
system before you analyze the process capability of the input and output parameters. The
measurement must be statistically controlled, which means that variation from the
measurement system is only from common causesnot special causes.2 The quality of a
measurement system is usually determined by the datas statistical properties produced over
time.3

To check variation of the measurement system, measurement system analysis (MSA)


was conducted. Because the defects under investigation were attribute type, attribute
agreement analysis was performed to understand the inspector variation. The MSA study
carried out for defect1 (roller imprint) of bending and tempering process is shown in Figure 3.
Twenty parts were selected for analysis: Five were perfectly OK, five completely not
good (NG), five marginally OK and five marginally NG. Two inspectors were selected to
participate in the study. The actual status of all 20 parts was noted as standard, and the two
inspectors were asked to give their opinions on the quality of partsOK or NG. The data
collected are shown in Table 2.
In general, values of kappa greater than 0.75 indicate good to excellent agreement
(with a maximum kappa = 1).4 The data analysis in Table 2 was done using Minitab statistical
software. Results were compared and fell within acceptable range. The resulting output is
shown in Figure 4, which clearly indicates that scores from both inspectors were within
acceptable range. A similar process was performed for other defects in all three subprocesses,
and variation was found to be within acceptable range.
After finding the measurement system to be acceptable, process capability studies for
input and output parameters were performed and showed that some of the input parameters
were having process capabilities of less than 1.33, as shown in Figure 5.
The capability analysis showed that some parameters were not in control and not
centered on the target. This resulted in poor control over output parameters. The control chart
of one of the output parameters (defect1 of bending and tempering processroller imprint) is
shown in Figure 6. The p-chart in Figure 6 depicts that there was no control over the output
parameters as well. This needed to be corrected.

Also at this stage, it was necessary to document the input parameters for the three
subprocesses, which was done using an input-output (I-O) sheet and a cause and effect
diagram. The I-O sheet for the heating-bending process (part of bending and tempering
process) is shown in Figure 7. The reason for high first-time yield (FTY) of the heatingbending process (99.7%) was that the furnaces heating and bending phase was not
accessible, and it wasnt possible to detect all the defects occurring during the heating and
bending process.
All of these defects were found during the final inspection stage. Based on the
experts process know-how, it was possible to figure out where the defects were occurring.
The FTY of 99.7% included only the glass that was broken during the bending process. To
that extent, it didnt include other defects from the bending process.
This I-O sheet highlighted that some of the processes basic conditions were not being
followed (as stated in process capability analysis). These abnormalities were corrected using
some quick-win opportunities. After correcting basic gaps in the processes, it was necessary
to filter out potential causes out of all possible causes.
A cause and effect matrix was used to filter causes having a high degree of correlation
with defects. A sample portion of a cause and effect matrix for bending and tempering
process is shown in Table 3. The highlighted causes were the ones having a high correlation
with defects.
These highly correlated input parameters were further filtered for a high degree of
occurrence using failure mode and effects analysis (FMEA). The filtering process for the
bending and tempering process is shown in Figure 8.

After filtering, we were left with 10 potential causes to explain the defects in the
bending and tempering process. A similar filtering process was performed for the pre-process
and printing process. The numbers of potential causes after filtering were:
1. Pre-process potential causes: six.
2. Printing process potential causes: seven.
3. Bending and tempering potential causes: 10.
After determining the potential causes for each subprocess, it was natural to proceed
to the analyse phase where the necessary validation of process parameters was performed.
Analyze phase
The validation of the causes is part of analyze phase. Here, we were able to further eliminate
some factors through graphical and statistical analysis. To do so, data in the running process
were collected for all the potential causes along with the defects. The data collection sheet
was designed with help from the SMEs. The data collection sheet and layout of equipment for
the bending and tempering process is shown in Table 4.
Data for 900 pieces of glass (10 glasses every 30 minutes) were taken to understand
the conditions in which the defects were being generated so suitable correlations with process
parameters could be established. Figure 9derived from the actual observations in Table
4is an illustration of the analysis performed to validate the factors responsible for the
defects.
Because the input parameters in this case were continuous and defects were discrete
variables, it was necessary to analyze using an interval plot in graphical analysis and binary
logistic regression in statistical analysis. In the data collected for cause validation in the preprocess and printing process, the input parameters were continuous as well as discrete, and

the defects were discrete (OK/NG). Therefore, it was required to use a one-proportion test to
validate the causes. Sample graphical and statistical analysis for each subprocess is shown in
Figure 9. The p-value < 0.05 highlighted in the circle meant that the cause was valid for the
defect it was tested against.
Similarly, all the potential causes for the three subprocesses were validated. Valid
causes are listed in Table 5. After discovering all of the valid causes, it was required to
optimize them. Optimization is part of the improve phase and was successfully performed in
this case study.
Improve phase
Optimization can be done using regression as well as DoE. Regression uses passive data,
meaning that you analyze data after it was collected.5 With DoE, your data collection is
active, which means you plan how and when you will collect your data in advance.10 With
DoE, you have blocking and other noise treatments to ensure the signals you see come from
your factors.6
With regression, you have limited noise control due to the passive nature of the data
collection.7 With DoE, you have more control over the measurement system accuracy than
with regression.8 Because historical data were not available in this case study, it was decided
to optimize the processes using DoE.
DoE is defined as the systematic procedure carried out under controlled conditions to
discover an unknown effect to test or establish a hypothesis, or to illustrate a known effect.9
When analyzing a process, experiments often are used to evaluate which process inputs have
a significant impact on the process output, and what the target level of those inputs should be
to achieve a desired result (output).10

Experimental design can be used at the point of greatest leverage to reduce design
costs by speeding up the design process, reducing late engineering design and process
changes, and reducing product material and labor complexity.11
In this case study, the designs were planned after consulting the SMEs. The list of
experiments and hypotheses testing conducted is shown in Table 6. One experiment from
each of the three subprocesses also is shown. Only the DoE for the bending and tempering
process is shown in detail.
Details on the three experiments follow:
1. Pre-process experimentDoE with cutting pressure, breaking off pressure, and
height of breaking off stopper to reduce breaking off chips rejection.
A three-factor, resolution III design with four blocks and 16 treatments was created to
conduct the experiment. It was decided to process 400 glasses at each treatment. The design,
along with data collected for defects and analysis, is shown in Figure 10.
Statistically, the interaction of cutting pressure and breaking off pressure should have
been neglected. On the advice of the SMEs, however, this was included because the
significance level of this parameter was about 88%. Minitab was used to optimize the
settings:

Cutting pressure: 2.2 Kgf.

Breaking off pressure: 2.0 Kgf.

Stopper height: 25 mm.

These settings were implemented on the production line, which resulted in an increase
of pre-process yield as shown in [Figure 13]. It is clear that pre-process yield has crossed the
yield target of 99.5% and is well maintained at the level.

2. Printing process experimentExperiment with new spacers (used to separate


glasses on pallet) with less hardness to reduce misprint rejection.
A sample of 34,564 pieces of glass using new spacers were processed, and the
rejection data for misprints (defect1 for the printing process) were compared with the
historical rejection data for misprints. Details of the analysis are shown in Figure 11. The pvalue 0.000 shows that misprint rejection was significantly reduced by using new spacers
with less hardness.
Similarly, other experiments were conducted and countermeasures were implemented
on the production line. The actions resulted in an increase of printing process yield as shown
in [Figure 13]. The yield graph shows that the printing process yield had crossed the target of
99.5% and was being well maintained at this level.
3. Bending and tempering process experimentDoE with main-line speed, zone-one
upper temperature, zone-one lower temperature, zone-two upper temperature,
zone-two lower temperature, zone-three upper temperature, zone-three lower
temperature to reduce curvature NG, blast-head breakage and roller-imprint
rejection:
This particular experiment was completed by optimizing the input parameters of the
bending and tempering process. The discussion among the cross-functional team members
resulted in selecting two levels of the factors, shown in Table 7.
As the number of factors in 2k factorial design increases, the number of runs required
for complete replication of the design rapidly outgrows the resources of most experimenters.12
It is not possible to accommodate so many runs in an experiment in a production facility.

Major use of fractional factorial designs occurs during the screening experiments in
which many factors are considered and used objectively to identify factors that have large
effects.13 In a seven-factor design at two levels, 128 treatments will be required to complete
one set of the experiment. We usually like to employ fractional designs that have the highest
possible resolution consistent with the degree of fractionation required. The higher the
resolution, the less restrictive the assumptions required regarding which interactions are
negligible to obtain a unique interpretation of the results.14
Ideally, a full-factorial design is the best. As an alternative to reduce the number of
treatments without compromising accuracy, a resolution IV design is generally acceptable
because no main effects are confounded with two-way interactions.
In this case, an experiment with seven factors, resolution IV, two blocks and 16
treatments was designed. It was decided to process 400 pieces of glass at each treatment. The
design is shown in Table 8.
An analysis of variance (ANOVA) table was created to identify significant factors in
the model. It is always important to examine the magnitude and direction of the factor effects
to determine which variables are likely to be important.15 ANOVA can generally be used to
confirm this interpretation.16 To create an ANOVA table using the data in Figure 19, certain
parameters needed to be calculated. The following provides sample calculations to create an
ANOVA table for blast-head breakage:

In a similar way, sum of squares, degrees of freedom, mean square, F value and p
value can be calculated for all of the parameters. The full ANOVA table for blast-head
breakage is shown in Figure 12. The p-values signify that all the terms included in the model
were significant for blast-head breakage. Also, the R2 value was 80.65, which indicates that
the terms included in the model explain 80.65% of the variation in blast-head breakage.
Analysis of data using Minitab resulted in following equation for blast-head breakage:

Similarly, equations and ANOVA table for roller imprint and curvature NG were
formulated. Applying a similar approach, the following are equations for roller imprint and
curvature NG:

At this stage, optimization of these defects was easy because the model equations had
been derived using Minitab. The objective became to reduce all three defects of the bending
and tempering process. Repeated iterations could have been done to find optimum levels of
the input parameters. Instead, Minitab was used to optimize the parameters. The optimum
settings derived were:

Main-line speed: 102 mm per second.

Zone-one lower temperature: 650 C.

Zone-two lower temperature: 650 C.

Zone-three upper temperature: 695 C.

Zone-three lower temperature: 710 C.

Implementing the earlier settings into the process resulted in breakthrough


improvement in the bending and tempering process yield, shown in Figure 13. Its clear that
the process yield had crossed the target of 93.5% and was being well maintained at a level of
97%.
Combining all three subprocesses together, there was an improvement of 9.8% in
process yield, as shown in Figure 14.
The improvement was validated based on statistical evidences of data. Daily data
were monitored to check the effectiveness of the actions taken.
Control phase
It is important to sustain implemented improvements for positive long-term business impact.
The control phase is intended to hold the gains obtained in the improve stage. Actions also
must be taken to ensure that other products or processes dont experience similar failures that
had been resolved.17
Introducing a control plan is essential for sustaining results. Table 9 shows a portion
of the control plan for the bending and tempering process, which was finalized in
consultation with the cross-functional team.
Similar control plans were developed for the pre-process and printing process, and
they have been helpful in sustaining the improvements gained. Control charts were used to
confirm the effectiveness of controls, and normality and capability analysis were performed
to check the behavior of the process. A control chart for zone-three lower temperature is
shown in Figure 24 and shows the process parameter was within control.

Also, through process capability studies, analysis showed the various process
parameters were capable. A sample is shown for zone-three lower temperature in Figure 15.
It also shows that Cpk is 1.39, which is above 1.33 and considered acceptable.
Correlation with business metric
The improvement in this case study was well appreciated by the management. Annual
recurring savings realized by this initiative was about $230,000. The project helped
management realize the value of Six Sigma for process improvement, and it decided to
embed Six Sigma in its problem-solving approach.
The study not only reduced the defects in the process, but it also exposed the team to
the importance of statistical tools in problem solving.
Although Six Sigma tools have been previously applied to glass manufacturing
processes, the overall approach can be applied in any industry in which Y = f(x) and the
objective is to optimize the input parameters to improve process yield.

NOTE AND REFERENCES


1. Value-added (VA) steps are those that change the shape of the product that the customer
is willing to pay for. VA steps are done right the first time. Similarly, nonvalue-added
(NVA) steps are those that may be necessary to complete production, but they do not
change the shape of the product. This type of NVA is called a business NVA (muda type
one) activity. The other type of NVA activity is muda type two, which means waste and
can be avoided.
2. Forrest W. Breyfogle III, Implementing Six Sigma: Smarter Solutions Using Statistical
Methods II, John Wiley & Sons, 2003.

3. Automotive Industry Action Group (AIAG), AIAG Measurement System Analysis


Reference Manual, fourth edition, 2010.
4. Ibid.
5. iSixSigma, How Is DOE Different From Regression, online forum discussion, April 8,
2008, www.isixsigma.com/topic/how-is-doe-different-from-regression.
6. Ibid.
7. Ibid.
8. Ibid.
9. Moresteam.com: The Engine Room of Continuous Improvement, Design of
Experiments, www.moresteam.com/toolbox/design-of-experiments.cfm.
10. Ibid.
11. Ibid.
12. Douglas C. Montgomery, Design and Analysis of Experiments, eighth edition, John Wiley
and Sons, 2013.
13. Ibid.
14. Ibid.
15. Ibid.
16. Ibid.
17. Matthew Barsalou and Robert Perkin, A Structured, Yet Flexible, Approach, Six Sigma
Forum Magazine, August 2015, pp. 21-25.

You might also like