ME371: Manufacturing

Systems
Dr. Supratik Mukhopadhyay

Quality Engineering

Lecture#1
 Why is ‘quality’ important ?
▪ Quality is one of the main deciding factors for choosing products or services from
competing sets.
▪ Understanding and improving quality leads to success and growth of business,
competitiveness and reliable customer base.
As per the American Society for Quality (ASQ), quality of a product or a service can be
defined from two different perspectives :
(a) The characteristics of a product or service that bear on it’s ability to satisfy stated or
implied needs [The value perspective] , also known as ‘fitness for use’ (Joseph Juran).
(b) A product or service that is free from deficiencies and defects [The conformance
S. Mukhopadhyay Lecture#1 ME371
perspective].
 Meaning of quality
▪ David Garvin of Harvard Business School found that most definitions of quality were
1. Transcendent: Intuitively understood, impossible to communicate, e.g. beauty or
love.
2. Product-based: Quality is an attribute of a product.
3. User-based: If the customer is satisfied, the product has good quality.
4. Manufacturing-based: If the product conforms to specifications, it has good quality.
5. Value-based: If the product is perceived having good value, it has good quality.

S. Mukhopadhyay Lecture#1 ME371

 Measures of quality
▪ Based on this, Garvin identified eight dimensions of quality:
― Performance: Brands can be ranked objectively on individual aspects of basic product performance.
― Features: Additional characteristics that enhance the appeal of the product or service to the user.
― Reliability: How long the product can go between failures or need for maintenance.
― Durability: Useful life of a product. How the product will hold up against extended or extreme use.
― Conformance: Was the product made, or the service performed as per the specifications?
― Aesthetics: How well the product appeals to the senses.
― Serviceabilty: How quickly the product can be repaired or service restored, after it falters.
― Perceived quality: What is the reputation of the product or service?

S. Mukhopadhyay Lecture#1 ME371

 Example

S. Mukhopadhyay Lecture#1 ME371

*Bozarth
 The ‘conformance’ perspective of quality
▪ The ‘conformance’ perspective of quality focusses on whether a product was
manufactured, or service was performed as intended.
▪ Conformance quality is typically evaluated by measuring the actual product or service
against some pre-established standards, e.g. number of defects in a car ,or number of
mistakes in the tax return (see previous example).
To fully address the ‘value’ and ‘conformance’ perspectives on quality, an organization must:
― Understand what dimensions of quality are important to the users.
― Develop products and services that will meet user’s requirements.
― Put in place business processes that are capable to meet specifications driven by user’s requirements.
― Verify that the business processes are indeed meeting the specifications.
S. Mukhopadhyay Lecture#1 ME371
 Total cost of quality

Maintaining quality incurs cost. Joseph Juran, a pioneer in Quality Engineering suggested
that there are four quality related costs:
▪ Internal failure costs
▪ External Failure costs
▪ Appraisal costs
▪ Prevention costs

S. Mukhopadhyay Lecture#1 ME371

 Total cost of quality

Internal failure costs

▪ These costs are caused by defects that occur prior to the delivery to the customer. This
includes money spent on repairing or reworking defective products , as well as the time
wasted on these activities.

External failure costs

▪ Costs incurred by defects that are not detected until the product or service reaches the
customer. These costs are large, and include warranty costs, loss of future business, and
in some situations, legal actions.

S. Mukhopadhyay Lecture#1 ME371

 Total cost of quality

Appraisal costs
▪ Appraisal costs are costs the company incurs in assessing it’s quality levels . Typical
appraisal costs are costs associated with sampling of products, inspection costs and
customer survey costs.

Prevention costs
▪ The costs a company incurs to prevent defects in a product or service to begin with, such
as costs of employee training, supplier certification efforts, investment in new processes,
equipment maintenance etc .

S. Mukhopadhyay Lecture#1 ME371

 Total cost of quality curve: Traditional view

Total quality cost = Failure costs + Appraisal costs + Prevention costs

▪ As defect levels increase, failure

costs increase .
▪ Appraisal cost is insensitive to the
level of defects.
▪ Prevention cost will increase , to
minimize the level of defect in the
product /service.
S.
AsMukhopadhyay Lecture#1
per the traditional view, there is an optimum ME371
defect level at which the cost is minimum!
 Total cost of quality curve: Zero defects view

▪ In reality, as defect level decreases,

there is little or no need to inspect
products -> Appraisal cost
decreases.
▪ Prevention costs held steady or even
decrease as managers and
employees become more skillful at
identifying and resolving problems..

AsMukhopadhyay
S. per this view, there is the cost is minimum at zero defect level.
Lecture#1 ME371
 Total Quality Management (TQM)

▪ Total Quality Management (TQM) is a managerial approach in which an entire

organization is managed so that it excels in all quality dimensions that are important
to customers.
▪ According to the International Organization for Standards, TQM is defined as: ‘A
management approach for an organization, centered on quality, based on the
participation of all its members and aiming at long-term success through customer
satisfaction and benefits to all members of the organization and to the society’.

S. Mukhopadhyay Lecture#1 ME371

 Total Quality Management (TQM)

▪ TQM is a broad business philosophy centered around seven core ideas or principles:
― Customer focus.
― Leadership involvement.
― Continuous improvement.
― Employee empowerment.
― Quality assurance.
― Supplier partnerships.
― Strategic quality plan.
S. Mukhopadhyay Lecture#1 ME371
 Total Quality Management (TQM)

▪ Customer focus: TQM demands that employees are willing to place them in customer
shoes. They need to know well the customer requirements. If this is not met, the
company risks alienating the customers.
▪ Leadership involvement: For a successful TQM implementation, the top management
needs to be involved. To inspire and guide managers, W Edwards Deming (famous
American Business theorist and economist) presented ‘fourteen points for
management’ : a set of guidelines for managers to follow.

S. Mukhopadhyay Lecture#1 ME371

 Total Quality Management (TQM)

▪ Deming’s 14 points:
1. Demonstrate consistency of purpose towards product improvement.
2. Adopt the new philosophy [of continuous improvement]
3. Use statistical methods of inspection rather than mass inspection.
4. End awarding business on the basis of price tag.
5. Find and work continually on problems.
6. Institute modern methods of training.
7. Institute modern methods of supervision.
8. Drive out fear, promote a company-oriented attitude.
9. Break down barriers between departments.
10. Eliminate numerical goals without providing methods how to achieve them. W Edwards Deming*
11. Eliminate standards prescribing numerical quotas.
12. Eliminate barriers that stand between the hourly worker and his right to pride of
workmanship.
13. Institute a program for education and training.
S. Mukhopadhyay
14.
Lecture#1
Create a corporate and management structure that will promote the above 13 points.
ME371
Image courtesy: https://deming.org
 Total Quality Management (TQM)

▪ Continuous improvement: A principle of TQM that assumes there will always be room for
improvement, no matter how well an organization is doing.
▪ Employee empowerment: This relates to giving employees the responsibility, authority,
training, and tools necessary to manage quality.
▪ Quality assurance: The specific actions firms take to ensure that products, services, and
processes meet the quality requirements of their customers.
▪ Supplier partnerships: TQM efforts should be extended to supply chain partners.
▪ Strategic quality plan: An organizational plan that provides the vision, guidance and
measurements to drive the quality effort forward and shift the organization’s course when
S. Mukhopadhyay Lecture#1 ME371
necessary.
 Statistical Quality Control (SQC)
Recall that, in order to ensure quality, an organization must verify that the business
processes are indeed meeting the specifications. Statistical Quality Control is directly aimed
at this issue, that makes sure that the business’s current processes are meeting the
specifications.
Statistical Quality Control is the application of statistical tools to Quality Control.
Some definitions:
Population (N): Population refers to the entire set of elements (products or services in our
case) whose attributes are to be measured.
Sample (n): This is a subset of the population. It is a representative portion of the population
S. Mukhopadhyay
from ME371
which data is collected for analysis to draw inferences about the entire population.
 Statistical Quality Control (SQC)
Upper tolerance limit (UTL): It is the upper limit of the acceptable range of values for some
measure of interest.
Lower tolerance limit (UTL): It is the lower limit of the acceptable range of values for some
measure of interest.

σ𝑛
1 𝑋𝑖
Mean : 𝜇 = where, 𝑥𝑖 => ith observation.
𝑛

σ𝑛
1 𝑋−𝑋𝑖
2
Standard deviation: 𝜎 =
𝑛−1

S. Mukhopadhyay ME371
 Process Capability Ratio (𝐶𝑝 )
▪ No two products would have identical measure of attributes, even if they have the same
design specifications, and they come from the same processes.
▪ This natural variation of processes should be small enough to generate products that meet
the required standards.
▪ A process in statistical control does not necessarily meet the design specifications.
▪ Process capability ratio is a measure of the relationship between the natural variation of
the process and the design specifications.

S. Mukhopadhyay ME371
 Process Capability Ratio (𝐶𝑝 )

Distribution of values of a part characteristic of interest at four times during process operation:
At t0, process is in statistical control;
At t1, process mean has increased;
At t2, process standard deviation has increased;
S. Mukhopadhyay ME371
*Groover
At t3, both process mean and standard deviation have increased
 Process Capability Ratio (𝐶𝑝 )
Assume that the outcome of the process is normally distributed.
1 1 𝑥−𝜇 2
−2 𝜎
Probability density function: 𝑓 𝑥 = 𝑒
2𝜋𝜎

Let 𝑋 be a certain measured value from a randomly distributed variable 𝑥

The probability that this value lies between 𝜇 ± 3𝜎 is:
𝜇+3𝜎 1 𝑥−𝜇 2
1 −
𝑃(𝜇 − 3𝜎 ≤ 𝑋 ≤ 𝜇 + 3𝜎) = න 𝑒 2 𝜎 𝑑𝑥 = 0.997
𝜇−3𝜎 2𝜋𝜎

S. Mukhopadhyay ME371
 Process Capability Ratio (𝐶𝑝 )
Process capability ratio is a mathematical determination of the capability of a process to
meet a pre-specified quality standard.

𝑈𝑇𝐿 − 𝐿𝑇𝐿
𝐶𝑝 =
6𝜎

UTL (Upper tolerance limit): The highest acceptable value for some measure of interest.
LTL (Lower tolerance limit): The lowest acceptable value for some measure of interest.

If 𝐶𝑝 ≥ 1, the process is capable of producing within acceptable tolerance limits.

Wider tolerance limits and/or smaller values of 𝜎 will result in higher 𝐶𝑝 values, while
S.narrower
Mukhopadhyay
tolerance limits and/or larger s values will have the opposite result. ME371
 Process Capability Ratio (𝐶𝑃 )

▪ Suppose the difference between the

upper and lower tolerance limit (UTL-
LTL) happens to be exactly 6𝜎 .
▪ This means the process is capable of
producing within the specified
tolerance limits 99.7% of the time and
𝐶𝑝 = 1.
▪ However, if the specified limits are
tighter than 6𝜎 , then 𝐶𝑝 < 1,
S. Mukhopadhyay ME371
 Process Capability Ratio (𝐶𝑝𝑘 )

▪ In some cases, the process mean is not exactly centered on the target value.
▪ In this case, a process capability index 𝐶𝑝𝑘 is established. 𝐶𝑝𝑘 = negative number

𝜇 − 𝐿𝑇𝐿 𝑈𝑇𝐿 − 𝜇 LTL UTL

𝐶𝑝𝑘 = min ,
3𝜎 3𝜎 𝐶𝑝𝑘 = zero

▪ A capable process must have a 𝐶𝑝𝑘 at least equal to 1.

𝐶𝑝𝑘 = between 0 and 1

𝐶𝑝𝑘 = 1

𝐶𝑝𝑘 > 1
S. Mukhopadhyay ME371
 Six Sigma quality

Six Sigma quality: A level of quality that indicates the process is ‘well-controlled’. The
term is usually associated with ‘Motorola’ which named one of it’s key operational
initiatives as Six Sigma quality.
At a Six Sigma quality level, the process variability is reduced to such an extent that
𝑈𝑇𝐿 − 𝐿𝑇𝐿
𝐶𝑝 = ≥2
6𝜎

This in effect amounts to squeezing in twelve standard

deviations within the specification limits.
S. Mukhopadhyay ME371
 Examples

S. Mukhopadhyay ME371
 Examples

S. Mukhopadhyay ME371
*Bozarth
 Examples

S. Mukhopadhyay ME371
*Bozarth
 Control charts

▪ In contrast to the process capability ratio (𝐶𝑝 ) and index (𝐶𝑝𝑘 ), control charts are
specialized run charts that help organizations track changes in key measures over time.
▪ By using control charts, an organization can quickly determine whether a process is “in
control” and take appropriate action if it is not.

Sampling: The idea of sampling is that businesses do not have to to examine every
process outcome to assess how well a process is doing. Instead, they can use carefully
selected samples to get a fairly good idea of how well a process is working.

S. Mukhopadhyay ME371
*Bozarth
 Control charts

A good sample is one in which:

▪ Every outcome has an equal chance of being selected into the sample. This is typically
accomplished by taking a random sample from the entire population.
▪ The sample size is large enough to not be unduly biased by any single observation.
Variable types:
▪ Continuous variable: A variable that can be measured on a continuous scale, e.g.
length, height, temperature.
▪ Attribute: A characteristic of an outcome or item that is accounted for by its presence
or absence, such as “defective” versus “good” or “late” versus “on-time.”
S. Mukhopadhyay ME371
 Control charts

Sample average: A key measure that represents the central tendency of a process
outcome extracted from a group of samples.

σ𝑛𝑖=1 𝑋𝑖
𝑋ത =
𝑛
𝑛 = number of observations in a sample
𝑋𝑖 = measured outcome for the 𝑖th observation.

Sample range: A key measure that represents the variation of a process outcome
extracted from a group of samples.

Range R = (highest value in the sample) – (lowest value in the sample)

S. Mukhopadhyay ME371
 Control charts

Proportion: Because attributes refer to the presence or absence of a particular

characteristic, the variable of interest is the proportion of the sample with the
characteristic. The proportion for a sample is measure that refers to the presence or
absence of a particular characteristic:

σ𝑛𝑖=1 𝑎𝑖
𝑝=
𝑛
Here ,
𝑛 = number of observations in a sample
𝑎𝑖 = 0 if attribute is not present for the 𝑖th observation and 1 if it is.
S. Mukhopadhyay ME371
 Example

S. Mukhopadhyay ME371
 Example

S. Mukhopadhyay ME371
 Control limits

▪ Control charts are specialized run-charts that help organizations track changes of key
measures over time.
▪ A control chart has a center line showing the expected value for a sample measure, as
well upper and lower control limits.
▪ Control limits are derived using statistical techniques. They are calculated so that if a
sample result falls inside the control limits, the process is considered “in control.” If a
sample result falls outside the control limits, the process is considered “out of control.”

S. Mukhopadhyay ME371
 Process of setting up control charts
Regardless of the variable type (whether continuous variable or attribute), the process for
setting up control chart is the same:
▪ Take 𝑚 samples of size 𝑛 each while the process is in control.
▪ Use the sample results to set up the control chart, using the tables or formulas provided.
▪ Continue to take samples of size 𝑛 and plot them against the control charts.
▪ Interpret the results and take appropriate action.

Control charts should not be employed until the process is capable of providing acceptable
performance on a regular basis.
Control charts, by themselves, will not result in improved quality levels. Rather, control charts
S. Mukhopadhyay ME371
are used to catch quality problems early, before they get out of hand => Appraisal activity
 Control chart for variables: 𝑋ത and 𝑅 chart

following slide

following slide

S. Mukhopadhyay ME371
 A2, D3 and D4

S. Mukhopadhyay ME371
Example

ME371
*Bozarth
Example

ME371
Example:Continued

ME371
Example

ME371
*Groover
Example

𝑥ҧ chart

𝑅 chart

ME371
 Control chart for attributes: 𝑝 chart

▪ When the measure of interest is an attribute, firms use 𝒑 charts to track the
sample proportions.
▪ Examples include fraction of nonconforming parts in a sample, proportion of
plastic molded parts that have flash, number of defects per automobile, and
number of flaws in a roll of sheet steel.
▪ As with 𝑋ത and 𝑅 charts, a 𝑝 chart has upper and lower control limits. If a sample
𝑝 value falls outside these limits, management should immediately investigate
to determine whether or not the underlying process has somehow changed.

ME371
𝑝 chart calculations

ME371
Example

ME371
*Groover
Example

ME371
*Bozarth
Control chart as a feedback loop

ME371
*Groover
 Acceptance sampling

Even under the best circumstances, defects can occur and be sent on to the customer. Companies must, therefore,
have some way to determine whether an incoming lot of material or products is of acceptable quality and to take
action based on the results. One way to determine the quality levels is through 100% inspection (i.e., inspection of
each and every item). While this may be necessary in some critical circumstances (e.g., donated blood), it has
drawbacks.
First, 100% inspection can be extremely expensive and time-consuming, especially if there are hundreds or even
thousands of items to inspect. Moreover, some quality inspection requires that goods be destroyed or otherwise
used up in order to be tested. Wooden matches are a good example. When 100% inspection is not an option,
companies depend on acceptance sampling to determine whether an incoming lot of items meets specifications.
APICS (Americal Production and Inventory Control Society) defines acceptance sampling as “the process of sampling
a portion of goods for inspection rather than examining the entire lot. The entire lot may be accepted or rejected
based on the sample even though the specific units in the lot are better or worse than the sample. ME371
 Acceptance sampling: Some terminologies

Acceptable Quality Level (AQL): A term used in acceptance sampling to indicate a

cutoff value that represents the maximum defect level at which a consumer would
always accept a lot.
Lot tolerance percent defective (LTPD): A term used in acceptance sampling to indicate
the highest defect level a consumer is willing to “tolerate.”
Consumer’s risk (𝜷): A term used in acceptance sampling to indicate the probability of
accepting a lot with quality worse than the LTPD level.
Producer’s risk (𝜶): A term used in acceptance sampling to indicate the probability of
rejecting a lot with quality better than the AQL level.
ME371
Example

ME371
 Operating Characteristics

Operating Characteristics curve (OC): For a given sampling plan, the Operating
Characteristics curve (OC curve) gives the probability that a batch will be accepted as a
function of possible fraction defect rates that might exist in the batch .
In effect, the OC curve indicates the degree of protection provided by the sampling
plan for various quality levels of incoming lots. If the incoming batch has a high quality
level, then the probability of acceptance is high. If the quality level of the incoming
batch is poor, then the probability of acceptance is low.

ME371
Operating Characteristics

ME371
 Operating Characteristics

▪ As per the OC curve, there is

an 80% chance that a lot will
be accepted that is 90%
defect free.
▪ Only 5% chance that it will
accept a lot that is ~ 60%
defect free.
▪ For AQL 5%, the probability
of rejecting a good lot is
around 8%.
▪ More importantly, the
probability of accepting a lot
that does not meet LTPD is ~
15% ME371
 OC curve with modified sampling plan

▪ A new sampling plan with 𝑛 = 20,

𝑐 = 2 is selected.
▪ A larger sample is more
representative of entire lot →
steeper OC curve.
▪ Producer’s risk falls to around 7%,
consumers risk falls to less than 5%!

Larger the sample size, lower the producers and consumers risk. ME371
 Taguchi methods in Quality Engineering

▪ Genichi Taguchi (1924-2012) was a Japanese Engineer and

Statistician.
▪ Taguchi spent much of his professional life on researching ways
to improve the quality of manufactured products.
▪ Taguchi methods have been accepted by many western
Genichi Taguchi*
statisticians as valuable addition to the body of knowledge for
controlling and improving quality.
▪ Taguchi’s concepts have been adopted by many companies:
Toyota, Fujifilm, Bell labs, Ford, Xerox, Boeing and ITT.
ME371
Image Courtesy: Wikipedia
 Taguchi methods in Quality Engineering

▪ After World War II, the Japanese Telephone System was badly damaged and
dysfunctional.
▪ Taguchi was appointed as Head of the newly formed Electrical Communications
Laboratories (ECL) of the Nippon telephone and Telegraph Company.
▪ Taguchi spent twelve years there developing methods for improving quality and
reliability.
▪ During the 1950s, he was a visiting professor at the Indian Statistical Instiute (ISI). There
he was introduced to the ‘orthogonal arrays’ technique, which he later extensively
applied to develop his ‘Design of Experiments (DOE)’.
ME371
 Taguchi methods in Quality Engineering

▪ The major contributions that Taguchi has made to the Quality Improvement programme
(also known as ‘Taguchi methods’) are:
1. Robust design concept.
2. The Quality loss function.
3. Design of Experiments (DoE) using Orthogonal Array (OA)

ME371
 Taguchi’s Robust design

▪ Taguchi’s definition of a robust design is: “a product whose performance is minimally

sensitive to factors causing variability”.
▪ There are variations that are difficult or impossible to control which affect the function
of a process or product.
― Taguchi calls these as noise factors.
▪ The goal is not to eliminate the noise, but to design the product or process in such a
way that the desired performance is minimally affected by the noise (build quality into
the product, rather than inspecting for it).

ME371
Basic idea of Robust design

ROBUSTNESS → QUALITY

Variability due to noise ↓

Quality ↑ Cost ↓

ME371
 Taguchi’s Robust design

Three level’s of noise factors can be identified:

▪ Unit-to-unit noise factors: Inherent random variations in the process or product caused by variability in
raw materials, machinery, and human participation. They are associated with a production process that is
in statistical control.
▪ Internal noise factors: These sources of variation are internal to the product or process. They include
(1) time-dependent factors such as wear of mechanical components, spoilage of raw materials, and
fatigue of metal parts; and
(2) operational errors, such as improper settings on the product or machine tool.
▪ External noise factors: An external noise factor is a source of variation that is external to the product or
process, such as outside temperature, humidity, raw material supply, and input voltage. External noise
factors are generally more difficult to control.
ME371
 Taguchi’s Robust design
A robust design is one in which the function and
performance of the product or process are relatively
Examples of Robust Design*
insensitive to variations in any of the noise factors.
▪ In product design, robustness means that the
product can maintain consistent performance with
minimal disturbance due to variations in
uncontrollable factors in its operating
environment.
▪ In process design, robustness means that the
process continues to produce good product with
minimal effect from uncontrollable variations in its
operating environment. ME371
*Groover
 Taguchi’s loss function

▪ Taguchi defines poor quality as ‘The loss a product costs society from the time the
product is released for shipment’.
▪ Loss includes :
― costs to operate.
― failure to function.
― maintenance and repair costs.
― customer dissatisfaction.
― injuries/accidents caused by poor design etc.

ME371
 Taguchi’s loss function

▪ In the traditional approach of quality control,

tolerance limits (UTL and LTL) are defined Target value
such that:
― Any product within that tolerance range is
acceptable.
― If the product is outside the tolerance limit, it
is not acceptable.
▪ This approach results in the
Loss function: Traditional approach
discontinuous/stepped loss function.
ME371
 Taguchi’s loss function

▪ In reality, the loss function is not discrete, but continuous, i.e. the quality of any
product or service starts to fall off in a continuous manner as soon as the measure
of interest drifts from the target value, such as
― The temperature of a freshly served cup of coffee.
― The length of a pair of pants.
― The amount of medicine in a capsule.
▪ Taguchi’s quadratic loss function reflects this idea that any deviation from the target
value incurs some failure cost.

ME371
 Taguchi’s quadratic loss function

▪ Taguchi proposed a loss function as:

𝐿 𝑥 = 𝑘(𝑥 − 𝑁)2
Target value
where, 𝐿 𝑥 is the loss function, 𝑥 is the quality characteristic of
interest, 𝑁 is the nominal target value, 𝑘 is a constant of
proportionality.
At some level of deviation (𝑥2 −𝑁) = −(𝑥1 − 𝑁), the loss will be Loss function: Taguchi approach
prohibitive, that will require to scrap or rework the product . This
level identifies one possible way of specifying the tolerance limit
for the dimension. But even within these limits, there is also a
ME371
loss, as suggested by the grey shaded area.
Example

ME371
*Groover
Cost of alternative tolerances

ME371
*Groover
Cost of alternative tolerances

ME371
*Groover
 General model for a process or system
Controllable factors
▪ A product is manufactured by passing input(s)
𝑥1 𝑥2 𝑥𝑛
through a process to generate certain output(s).
▪ There are certain desirable quality of the output(s): ⋯
larger the better, smaller the better or nominal
Inputs Output 𝑦
(target value) the better. Process
▪ Many ‘factors’ influence the quality of tge output.
▪ Factors operate at different levels ⋯
― Combinations of levels need to be decided to achieve
optimum output quality 𝑧1 𝑧2 𝑧𝑛
Uncontrollable factorsME371
 Examples of Engineering optimization

▪ Design of aircraft structures for minimum weight.

▪ Design of pumps, turbines and heat transfer equipment for maximum efficiency.
▪ Selection of machining conditions in metal cutting processes for minimum production cost.
▪ Optimum design of electrical networks.
▪ Controlling the waiting and idle times in production lines to reduce costs.
▪ Optimising fertiliser combination for maximising crop yield.

ME371
The Taguchi roadmap to achieve quality

*www.6sigma.us
ME371
 Design of Experiments (DoE)

▪ The term “Design of Experiments

(DoE)”, also known as experimental
design, was coined by Ronald Fisher in
the 1920s.
▪ The purpose of DoE is to efficiently
plan, conduct, analyze, and interpret
experiments to optimize a process or
product by systematically varying
factors and identifying their effects. *https://safetyculture.com
ME371
 Strategy of experimentation

▪ ‘Best guess’ experiments

― Used a lot
― More successful than one might expect, but it has disadvantages.

▪ One-factor-at-a-time (OFAT)
― Sometimes associated with the ‘scientific’ or ‘engineering’ method.
― Devastated by interaction,

▪ Statistically designed experiment

― Based on Fischer’s factorial concept.

ME371
 Factorial design (FD)

▪ Factorial experiment is an experiment whose design consists of two or more factor

each with different possible values or ‘levels’.
― Factor’s can be quantitative (numbers) or qualitative.
― They may be names rather than numbers like: Method 1, site A, present or absent
etc.
▪ FD technique introduced by Fisher in 1926.
▪ Factorial design mainly applied in optimization techniques.

ME371
 Full Factorial design

▪ The total number of experiments required to run all possible combinations of all the levels
for each of the factors.
▪ If there are 𝑘 factors, and there are 𝑛 levels associated with each factors, the total number
𝑘
of experiments required to include all possible combinations of the factors is: 𝑛 .
X1

22 FD => 2 factors, 2 levels, 4 runs.

23 FD => 3 factors, 2 levels, 8 runs.
X2
32 FD => 2 factors, 3 levels, 9 runs.
33 FD => 3 factors, 3 levels, 27 runs.
ME371
2 factors X1 and X2, 2 levels: High (+) and Low (-)
 Full Factorial design

-++ +++
+-+
--+

X3
-+- ++-

X2

--- +--
X1

3 factors X1, X2 and X3, 2 levels: High (+) and Low (-) ME371
 Advantages and Limitations

▪ Advantages:

― It is able to provide comprehensive insights into the system under investigation,

since it evaluates the outcome under all possible factor combinations.

― Unlike some experimental designs that ignore interaction between factors, the full
factorial design explicitly accounts for interaction between factors.

▪ Limitations:

― Resource intensive, as the number of factors increase, the number of required runs
grow exponentially , 10 factors at 2 levels → 210 experiments!

― Requires large sample sizes for statistical validity and reliable estimates. ME371
 Fractional factorial design

▪ In this approach, experiments are conducted only on a selected subset or ‘fraction’ of the
runs in the full factorial design.
▪ Fractional factorial designs are a good choice if resources are limited or if the number of
factors in a design experiment is large.
▪ Fractional factorial designs uses a subset of a full factorial design, so some of the main
effects and 2-way interactions cannot be separated from the effects of other higher-order
interactions. Usually experimenters assume the higher-order effects are negligible to
achieve information about main effects and low-order interactions with fewer runs.
▪ Example : A ‘half-fractional factorial design’ conducts 2𝑘−1 experiments for a two-level k
ME371
factor experiment.
 Fractional factorial design

Suppose a 23 two-level Full-Factorial design table is given showing runs and observations:
.
X1 X2 X3 Y
1 -1 -1 -1 Y1=33
2 +1 -1 -1 Y2=63
3 -1 +1 -1 Y3=41
4 +1 +1 -1 Y4=57
5 -1 -1 +1 Y5=57
6 +1 -1 +1 Y6=51
7 -1 +1 +1 Y7=59
8 +1 +1 +1 Y8=53

ME371
 Fractional factorial design

▪ From this table, we are able to compute all effects,

such as main effects, first order interaction effects
etc.
▪ Suppose we want to compute the main effect ‘C1’
of the factor X1

Average response of all factors when X1 is at ‘high’ –

Average response of all factors when X1 is at ‘low’
1 1
𝐶1 = 𝑌2 + 𝑌4 + 𝑌6 + 𝑌8 − 𝑌1 + 𝑌3 + 𝑌5 + 𝑌7
4 4
1 1 ME371
𝐶1 = 63 + 57 + 51 + 53 − 33 + 41 + 57 + 59 = 8.5
4 4
 Fractional factorial design

▪ Suppose now, we only have resources to conduct

four runs (half-fractional factorial design).
▪ Is it still possible to get the main effect C1 of X1?
▪ The answer is yes, and there are even different
choices of the four runs that will accomplish this.
▪ For example, let us do experiments 1,4,6, and 7
only.

ME371
Fractional factorial design

-++ +++
+-+
--+

X3
-+- ++-

X2

--- +--
X1

ME371
 Fractional factorial design

▪ With the reduced experimental runs, we can still

compute C1 as:
1 1
𝐶1 = 𝑌4 + 𝑌6 − 𝑌1 + 𝑌7
2 2
1 1
𝐶1 = 57 + 51 − 33 + 59 = 8
2 2

▪ Similarly, we could compute C2, the

effect X2 as:
1 1
𝐶2 = 𝑌4 + 𝑌7 − 𝑌1 + 𝑌6
2 2
1 1 ME371
𝐶2 = 57 + 59 − 33 + 51 = 16
2 2
 Advantages and Limitations

▪ Advantages:

― Reduced resource requirement

― Obtain the main effect of parameters with reduced set of experiments.

▪ Limitations:

― Interaction between factors and higher order effects may not be captured!
.

ME371
 Orthogonal arrays

▪ In Mathematics, an Orthogonal Array is a “table” or array whose entries come from a

fixed finite set of symbols (example :{1,2,…v}) arranged in such a way that there is an
integer t so that for every selection of t columns of the table, all ordered t-tuples of the
symbols, formed by taking entries in each row restricted to these columns, appear the
same number of times.
▪ The number t is called the strength of the orthogonal array.

ME371
 Example

▪ The example at right is that of an orthogonal array with symbol set

{1,2} and strength 2.
1 1 1
▪ The four ordered pairs (2-tuples) formed by the rows restricted to
the first and third columns :(1,1), (2,1), (1,2) and (2,2), are all the 2 2 1

possible ordered pairs of the two element set and each appears
1 2 2
exactly once.
2 1 2
▪ The second and third columns would give, (1,1), (2,1), (2,2) and
(1,2); again, all possible ordered pairs each appearing once.
▪ The same statement would hold had the first and second columns
been used. ME371
 Example

▪ The example at right is that of an

orthogonal array with symbol set 0 0 0 0
0 0 1 1
{0,1} and strength 3. 0 1 0 1
Check for yourself! 0 1 1 0
1 0 0 1
1 0 1 0
1 1 0 0
1 1 1 1

ME371
 Taguchi’s Orthogonal arrays

▪ Taguchi used this concept of Orthogonal Arrays (OA) to efficiently design the
experiment matrix for his DoE program.
▪ It is a highly fractional factorial design that allows to consider a selected subset of
combinations of multiple factors at multiple levels.
▪ Due to the balanced nature of orthogonal arrays, it is ensured that all levels of all
factors appear equal number of times, avoiding any bias.

ME371
 Taguchi’s Orthogonal arrays

▪ Taguchi used this concept of Orthogonal Arrays (OA) to efficiently design the
experiment matrix for his DoE program.
▪ It is a highly fractional factorial design that allows to consider a selected subset of
combinations of multiple factors at multiple levels.
▪ Due to the balanced nature of orthogonal arrays, it is ensured that all levels of all
factors appear equal number of times, avoiding any bias.

Notation of Taguchi’s OA: Lruns (Levelsfactors) e.g. L4 (23), L8 (27), L9 (34),

ME371
 Example

3 factors

L4 (23)
ME371
2 levels (designated ‘1’ and ‘2’)
4 runs
 Example (two level designs)

L8 (27)

L12 (211)

L16 (215) ME371

 Example (three level designs)

L9 (34)

L27 (313) ME371

 Advantage of Taguchi’s OA

As compared to full factorial experiments, Taguchi’s OA provides same results in

lesser ‘run’s (experiments)

Number of factors Number of levels Number of experiments

Full factorial Taguchi
3 2 8 4
7 2 128 8
15 2 32,768 16
4 3 81 9
13 3 1,594,323 27

ME371
 Steps in Taguchi’s DoE

▪ Decide the important input process parameters and their levels (by pilot
experimentation and/or literature review).
▪ Select the appropriate OA and assign the parameters to it’s various column’s.
▪ Conduct experiments for the levels given in each row RANDOMLY and note
down the values of the response parameter. Repeat each experiment three (or
more) times for statistical significance.
▪ Study factor effects and find out the optimum combination of input
parameters. Calculate the corresponding best value of the response
characteristic.
▪ Conduct confirmation experiment (if required) to verify the DoE outcome.
▪ Perform Analysis of variance (ANOVA) to find out the significance of various
parameters and their relative contribution.

ME371
 Some definitions

▪ Designed experiments are called matrix experiments and individual experiment

constituting one row of the orthogonal array is called run.
▪ Parameters are called factors and settings of these factors are called levels.
▪ The response variable is called the dependent variable, as it’s value is assumed to
depend on the values of the factors.
▪ Factors are also called treatment, a term which comes from agricultural application,
where Taguchi’s DoE was first applied.
▪ Degrees of freedom of a 𝑛 level factor is 𝑛 − 1, e.g. DOF of a 3 level factor is 2.

ME371
 Some definitions

▪ Designed experiments are called matrix experiments and individual experiment

constituting one row of the orthogonal array is called run.
▪ Parameters are called factors and settings of these factors are called levels.
▪ The response variable is called the dependent variable, as it’s value is assumed to
depend on the values of the factors.
▪ Factors are also called treatment, a term which comes from agricultural application,
where Taguchi’s DoE was first applied.
▪ Degrees of freedom of a 𝑛 level factor is 𝑛 − 1, e.g. DOF of a 3 level factor is 2.

ME371
 Example: Optimizing EDM process outcome

▪ Discharge current.
▪ Sparking voltage
▪ Type of flushing
Factors
▪ Type of dielectric
▪ Pulse on-time
▪ Pulse off-time
▪ Polarity
▪ Micro-hardness Dependent variable/Response
▪ Surface roughness
ME371
 Example: Optimizing EDM process outcome

▪ Three levels of discharge current are chosen: 2A,4A and 6A, because any nonlinearity
in the response characteristic can only be studied with a minimum 3 levels.
▪ Three levels of pulse-on time : 5,10 and 20 𝜇sec.
▪ Three levels of pulse-off time: 38, 57, and 85 𝜇sec.
▪ Response Name: Micro-hardness
▪ Response Type: Higher-the-better
▪ Units: HV (Vicker’s hardness number)
▪ Response Name: Surface Roughness, 𝑅𝑎
▪ Response Type: Lower-the-better
▪ Units: Microns ME371
 Example: Optimizing EDM process outcome

▪ We have 3 factors at 3 levels. We choose L9 orthogonal array. But L9 allows for 4

factors. We can leave the 4th column blank. The orthogonality is not disturbed by this.
▪ Each row of the orthogonal array represents the set of values (levels) of the input
factors with which a particular experiment has to be continued. In L9,a total of 9
experiments are required for the study.
▪ Taguchi strongly recommends that rows must be selected at random, and each
experiment must be repeated at least 3 times for statistical significance.

ME371
 Example: Optimizing EDM process outcome

▪ For discharge current:

2A ≡ 1, 4A ≡ 2, 6A ≡ 3
▪ For Pulse on-time:
5𝜇sec ≡ 1, 10𝜇sec ≡ 2, 20𝜇sec ≡ 3
▪ For Pulse off-time:
38𝜇sec ≡ 1, 57𝜇sec ≡ 2, 85𝜇sec ≡ 3

L9

ME371
Example: Optimizing EDM process outcome

Experiment Columns
number Discharge Pulse on-time Pulse off-time Unassigned
current
1 2A 5 𝜇sec 38 𝜇sec
2 2A 10 𝜇sec 57 𝜇sec
3 2A 20 𝜇sec 85 𝜇sec
4 4A 5 𝜇sec 57 𝜇sec
5 4A 10 𝜇sec 85 𝜇sec
6 4A 20 𝜇sec 38 𝜇sec
7 6A 5 𝜇sec 85 𝜇sec
8 6A 10 𝜇sec 38 𝜇sec
9 6A 20 𝜇sec 57 𝜇sec

ME371
 Signal to Noise (S/N) ratio

𝑛
S/N ratio for “larger the better”: 1 1
𝑆𝑁𝐿 = −10 log ෍ 2
𝑛 𝑦𝑖
𝑖=1

𝑛
1
S/N ratio for “smaller the better”: 𝑆𝑁𝑆 = −10 log ෍ 𝑦𝑖2
𝑛
𝑖=1

▪ S/N ratio indicates the actual effect of the factors over the effect of noise.
▪ Our aim is to look for maximum S/N ratio. Maximizing the S/N ratio will
make the design robust.

ME371
 Experimental results
Experiment Micro-hardness (HV) S/N Ratio Surface roughness Ra (microns) S/N ratio
No. R1 R2 R3 (dB) (dB)
R1 R2 R3

1 617 598 607 55.666 2.57 2.68 2.76 -8.534

2 595 582 572 55.310 2.63 2.71 2.75 -8.618
3 576 585 598 55.360 3.75 3.91 3.82 -11.618
4 653 658 662 56.360 2.93 3.12 2.98 -9.574
5 651 645 638 56.186 3.83 3.76 3.68 -11.497
6 598 611 623 55.712 5.17 5.28 5.05 -14.266
7 691 685 672 56.682 4.52 4.67 4.38 -13.112
8 608 620 633 55.849 5.37 5.53 5.58 -14.798
9 640 618 622 55.938 5.63 5.72 5.95 -15.221
Total 5629 5602 5627 503.063 36.40 37.38 36.95 -107.278
MH=overall mean of micro- Mean SR=overall mean surface Mean
hardness = 624.37 𝑚=55.896 roughness = 4.101 𝑚=- ME371
11.9198
 Calculating effect of factors
𝜂1 + 𝜂2 + 𝜂3
Effect of factor A at level 1 => 𝑚𝐴1 =
3
where, 𝜂𝑖 is the S/N ratio of 𝑖 th row of OA
𝜂4 + 𝜂5 + 𝜂6
Effect of factor A at level 2 => 𝑚𝐴2 =
3
𝜂1 + 𝜂4 + 𝜂7
Effect of factor B at level 1 => 𝑚𝐵1 =
3
𝜂2 + 𝜂5 + 𝜂8
Effect of factor B at level 2 => 𝑚𝐵2 =
3

ME371
 Calculating effect of factors

Factor levels Peak current (A) Pulse on-time Pulse off-time

Raw data S/N data Raw data S/N data Raw data S/N data
L1 592.22 55.445 649.22 56.236 612.78 55.743
L2 637.67 56.086 616.00 55.782 622.44 55.869
L3 643.22 56.156 607.85 55.670 637.89 56.076

ME371
Effect of factors on Micro hardness and it’s S/N ratio

ME371
Effect of factors on Surface roughness and it’s S/N ratio

ME371
 Rank of the factors on influencing Micro-hardness

Factors L1 L2 L3 Max-Min Rank

Peak current 55.445 56.086 56.156 0.711 1
Pulse-on time 56.236 55.782 55.670 0.566 2
Pulse-off time 55.743 55.869 56.076 0.333 3

ME371
 Calculated value of optimum S/N ratio
▪ The optimum condition of the three process parameters (factors) for micro-hardness is:
A3B1C3.
▪ The theoretical value of 𝜂 under the optimum conditions : 𝜂𝑜𝑝𝑡 is given by:
𝜂𝑜𝑝𝑡 = 𝑚 + 𝑚𝐴3 − 𝑚 + 𝑚𝐵1 − 𝑚 + 𝑚𝐶3 − 𝑚
▪ The corresponding optimum value of ‘higher-the-better’ type response characteristic is:
𝜂𝑜𝑝𝑡
2 − 10
𝑦𝑜𝑝𝑡 = 1/ 10
▪ For the lower the better type characteristic:

𝜂𝑜𝑝𝑡
2
𝑦𝑜𝑝𝑡 = 10− 10
▪ In the present case, optimum micro hardness = 682.08 HV
ME371
 ANOVA analysis for percentage contribution

▪ ANOVA (Analysis of Variance) is a statistical technique for testing the hypothesis that there is
no difference between two or more population means
(Null hypothesis : 𝜇𝐴 = 𝜇𝐵 = 𝜇𝐶 … ,Alternate hypothesis is that this is not true).
▪ It is basically a procedure for testing the difference among different groups of data for
homogeneity
▪ “The essence of ANOVA is that the total amount of variation in a set of data is broken down
into two types, the amount that can be attributed to chance and the amount which can be
attributed to specific causes.”
▪ ANOVA analysis is paired with DoE to determine if a factor has an effect on a dependent
variable. ME371
 Steps

▪ Calculate the Sum of squares for each factor/treatment: 𝑆𝑆𝑡 = ෍ 𝑛 𝑋ഥ𝑖 − 𝐺𝑀 2

Grand mean
Number of repetitions

Therefore, Individual sample means

𝑆𝑆𝐴 = 3(𝑚𝐴1 − 𝑚)2 +3(𝑚𝐴2 − 𝑚)2 +3(𝑚𝐴3 − 𝑚)2

𝑆𝑆𝐵 = 3(𝑚𝐵1 − 𝑚)2 +3(𝑚𝐵2 − 𝑚)2 +3(𝑚𝐵3 − 𝑚)2

𝑆𝑆𝐶 = 3(𝑚𝐶1 − 𝑚)2 +3(𝑚𝐶2 − 𝑚)2 +3(𝑚𝐶3 − 𝑚)2

ME371
 Steps

▪ Calculate the Degrees Of Freedom corresponding to factor levels: 𝐷𝐹𝑡 = 𝐿 − 1

Number of levels of a factor

Therefore,

𝐷𝐹𝐴 = 3 − 1 = 2

𝐷𝐹𝐵 = 3 − 1 = 2

𝐷𝐹𝐶 = 3 − 1 = 2

ME371
 Steps
Sum of squares of a factor
𝑆𝑆𝑡
▪ Calculate the Mean Squares of the treatment/factors: 𝑀𝑆𝑡 =
𝐷𝐹𝑡

DOF of a factor

Therefore,

𝑀𝑆𝐴 = 𝑆𝑆𝐴 /𝐷𝐹𝐴

𝑀𝑆𝐵 = 𝑆𝑆𝐵 /𝐷𝐹𝐵

𝑀𝑆𝐶 = 𝑆𝑆𝐶 /𝐷𝐹𝐶

ME371
 Steps

▪ Calculate the Sum of squares of the error: 𝑆𝑆𝑒 = ෍ 𝑋𝑖 − 𝑋ഥ𝑖 2

Individual sample means

Individual values
Therefore,

2
𝑆𝑆𝐸 = σ 𝜂𝑖 − 𝑚

ME371
 Steps

▪ Calculate the Degrees of Freedom of the error:

𝐷𝐹𝑒 = 𝑁 − 1 − ෍ 𝐿 − 1

Total number of observations Total DOF of factors

Therefore, in the present case

𝑆𝑆𝐸 = 9 − 1 − 3 × 3 − 1 = 2

ME371
 Steps
Sum of squares of a factor
𝑆𝑆𝑒
▪ Calculate the Mean Squares of the error: 𝑀𝑆𝑒 =
𝐷𝐹𝑒

DOF of a factor

ME371
 Steps

Mean Squares of the treatment

𝑀𝑆𝑡
▪ Calculate the F value: 𝐹𝑡 =
𝑀𝑆𝑒
Mean Squares of the error

ME371
 Steps

ANOVA table

Factors/ Sum of squares DOF Mean Squares F ratio P%

Treatments (Percentage
contribution)
Factor A 𝑆𝑆𝐴 𝐷𝐹𝐴 𝑀𝑆𝐴 𝐹𝐴 𝑆𝑆𝐴
× 100
𝑆𝑆𝑡𝑜𝑡
Factor B 𝑆𝑆𝐵 𝐷𝐹𝐵 𝑀𝑆𝐵 𝐹𝐵 𝑆𝑆𝐵
× 100
𝑆𝑆𝑡𝑜𝑡
Factor C 𝑆𝑆𝐶 𝐷𝐹𝐶 𝑀𝑆𝐶 𝐹𝐶 𝑆𝑆𝐶
× 100
𝑆𝑆𝑡𝑜𝑡
Error 𝑆𝑆𝑒 𝐷𝐹𝑒 𝑀𝑆𝑒 - -
Total

ME371
 Steps

For the present example ANOVA table

Factors/ Sum of squares DOF Mean Squares F ratio P%

Treatments (Percentage
contribution)
Factor A 2.0121 2 1.0061 1176.98 70.27
Factor B 0.6281 2 0.3141 367.40 21.93
Factor C 0.2215 2 0.1108 129.59 7.74
Error 0.0017 2 0.0009 - 0.06
Total 2.8634 8

Tabulated value at F ratio at 95% confidence level = 19. If F_factor > this value, effect of factor is
significant, otherwise insignificant.
ME371
F table

ME371