Ambo University

Department of Industrial Engineering

1
 Chapter Outline
•Perception of quality
•Taguchi’s definition of quality and quality
loss function
•Planning of experiments
•Basic design principles and terminology
•Normal probability plots
•Analysis of variance (ANOVA)
•Linear regression models

2
Perception of Quality
•What does "high quality" means?

•The Old View (The Goalpost Model): A product is good if it falls within pre-set specification
limits. Anything inside the limits is "good," and anything outside is "bad."
• Problem: This ignores customer satisfaction. A part that is just barely within spec is no better than one
that is just barely out of spec, yet one is accepted and one is scrapped. A customer prefers a product that
is on target.

•The Modern View: Quality is about consistently meeting customer expectations and being on
target, not just within spec. This leads to higher customer satisfaction and lower costs.

3
 Taguchis Definition of Quality & Loss Function
• Genichi Taguchi, a Japanese engineer, revolutionized quality thinking.
• His Definition: “Quality is the loss a product imparts to society after it is shipped. “This loss includes
warranty costs, customer dissatisfaction, lost reputation, and time wasted.
• The Quality Loss Function (QLF): Taguchi mathematically defined this loss. It's a simple quadratic
equation:
L(y) = k × (y - T)²
L(y) = Financial Loss ($)
y = Actual value of the product's characteristic (e.g., diameter, weight)
T = Target value (the ideal value)
k = A constant unique to the process (a scaling factor)

4
Key aspects
•Shift From tolerance Based Quality: Unlike traditional methods that consider products within
tolerance as "good," the QLF recognizes that quality and loss are continuous variables, with loss
beginning the moment a characteristic deviates from the target.
•Economic Perspectives: The function translates quality defects into economic terms, helping
businesses understand the financial implications of even small, off-target performance.
•Quadratic Relationships: The loss increases quadratically with deviation (y-T)², meaning the
loss grows significantly as a product or process moves further from its target.
•Constant ‘K’ :This constant is crucial and is determined by the cost of failure at the specification
limit. The relationship is often expressed as k = C/d², where C is the cost of failure at the
tolerance limit and d is the distance from the target to the tolerance limit.

5
Applications and Benefits
•Process Optimization: By understanding the loss associated with variation,
manufacturers can focus on improving robustness and consistency, which helps
reduce overall costs.
•Customer-Centric Approach:The QLF highlights that customer satisfaction is
affected by deviations from the ideal, not just being outside engineering tolerances.
•Decision Making:It provides a quantitative tool for making decisions about product
design and process improvements to minimize quality costs

6
Continued
• Why it's important: The loss increases exponentially as you move away from the target. This shows that
even products within specification limits can cause significant financial loss if they are not on target.

• Simple Example:

• Baking a cake with a target baking time of 30 minutes.

• A cake baked for 30 minutes is perfect.

• A cake baked for 31 minutes is a little dry. You might still eat it, but it's not ideal (some loss).

• A cake baked for 35 minutes is burnt and must be thrown away (total loss).

• The QLF captures this increasing dissatisfaction, which translates to financial loss.

7
 Planning of Experiments (Design of Experiments - DOE)
• How do we find out what affects quality? We don't guess—we experiment!

• DOE is a structured method for determining the relationship between factors (inputs) affecting a process and the output
of that process.

• Instead of changing one thing at a time (which is inefficient and can miss important interactions), DOE lets you change
multiple factors simultaneously in a specific pattern.

Basic Terminology:
• Response (Y): The output you measure (e.g., strength, yield, time). The "y" in the Loss Function.

• Factor (X): An input variable you can control (e.g., Temperature, Pressure, Material Type).

• Level: The specific setting of a factor (e.g., Temperature: 100°C or 150°C; Material: Type A or Type B).

• Run/Trial: One instance of the experiment where you set all factors to specific levels and measure the response.
8
Basic Design Principles & Terminology
To get valid results, experiments must be planned correctly.

1. Randomization: Running trials in a random order. This helps eliminate the influence of unknown "lurking"
variables.

2. Replication: Repeating the entire experiment or specific runs. This helps you estimate the natural variation
(noise) in the process.

3. Blocking: A technique to group similar experimental units together to account for nuisance variables you can't
control (e.g., different raw material batches, different days of the week).

Common Design: A "2-Level Full Factorial" design. If you have `k` factors, you test all 2^k combinations of their
high and low levels. It's efficient and reveals interactions.

9
Normal probability plots
•What it is: A simple graphical tool to check if your data follows a normal (bell-shaped)
distribution.

•Why we use it:

1. To check if the experimental data meets the assumptions of many statistical tests.

2. To identify important factors! When we analyze a DOE, we plot the estimated effects of each
factor. Effects due to random noise will fall along a straight line. Important effects will deviate
from this line, making them easy to spot.

10
 Analysis of Variance(ANOVA)
• What it is: A statistical "calculator" that determines if the differences in your output (Y) are caused by the
factors (X's) you changed or just by random chance.
• How it works: It breaks down the total variation in your data into pieces:
• Variation due to Factor A
• Variation due to Factor B
• Variation due to the interaction of A & B(e.g., does Temperature affect the outcome differently depending on the
Material Type?)
• Variation due to Random Error (the noise)

•The Output: ANOVA gives you a p-value for each factor.

11
 Linear Regression Model
• What it is: The "magic formula." It's the equation that predicts your output (Y) based on the inputs (X's).
• Purpose: Once ANOVA tells you which factors are important, regression builds the model to predict how they
affect the response.
The Model Looks Like This:
Y = β₀ + β₁X₁ + β₂X₂ + β₃X₁X₂`
Y: The predicted Response (e.g., final strength)
β₀:The baseline or average result.
β₁:The coefficient for Factor 1 (e.g., how much strength increases for every unit increase in Temperature).
X₁: The value of Factor 1 (e.g., Temperature)
β₃X₁X₂:This is an interaction term. It means the effect of Temperature (X₁) depends on the level of
Material (X₂).

12
Quality Improvement Cycle
1. Philosophy: Adopt Taguchi's view. Your goal is to be on target, not just in spec.
2. Plan: Use DOE to efficiently test which process factors (X's) control your key output (Y).
3. Analyze: Use ANOVA and Normal Plots to identify the vital few significant factors from the
trivial many.
4. Model: Use Regression to create a prediction equation.
5. Optimize:Use the model to find the perfect settings for your factors (X's) to get the output (Y)
exactly on its target value consistently.
6. Verify: Run a confirmation experiment to prove your new settings work.

13
Any Questions ?
Thank You!

14