9 views

Uploaded by Ryan Nguyen

Part 1: Quality Assessment using Quality Loss Functio

Part 1: Quality Assessment using Quality Loss Functio

© All Rights Reserved

- Surveying by Dr. Ramachandra
- Plabstat Manual
- Thera-band Elastic Band Tension Reference
- Allometric Equation for Biomass Estimation of Cedrus Deodara (Deodar) in Northern Pakistan
- Linest
- Output Sig Iik
- generalized additive model
- Central Composite Design for the Optimization of r Absorbance
- vvxc
- Output Yulianto 25 Maret 2019
- Dm 26785792
- ANOVA (QUALITY MANAGEMENT)
- ch13
- Brm Power Point by Nila
- Download Fullpapers Akk60199870a9full
- exemplu_regresie
- Caro 2013
- ANOVA 4 regression
- 11E Chapter 13
- Repeated Measures

You are on page 1of 76

Mechanical and Aerospace Engineering

Columbia, Missouri, USA

Quality By Design

Part 1: Quality Assessment using Quality Loss Function

Fall 2016

By

Prof. Dr. Eng. Ahmed Sherif El-Gizawy

Professor and Director Tech Development Center

University of Missouri, USA

Presentation Method

CBL involves the interactive, participantcentered exploration of realistic and

specific situations.

Realistic case studies are presented to

stimulate seminar discussion and

collaborative analysis and understanding

using tools and techniques presented in

the workshop.

CBL is used effectively in all professional

teaching (Medicine, Law and Business).

Quality by Design (QBD) is one of the most powerful quality improvement tools.

and processes that are robust.

and process designs.

Products or services with specification on the target values give best performance .

rather than ideal performance.

The focus in the present seminar is on the application of the quadratic loss function

to quantify improvement opportunities in engineering industry.

Costs Associated with Product Life Cycle

(fabrication cost)

(costs of repair & replacement

due to quality loss)

It is important to quantify quality loss associated with different

product and process designs.

The old method of measuring quality by fraction defective

(GO/NOGO method) is misleading.

Products with specification on the target values give best

performance.

Example: Color density in TV set.

4

Sony - USA

Sony - Japan

different system designs.

The old method of measuring quality by fraction

defective (GO/NOGO method) is misleading.

Systems with outputs on the target values give best

performance.

5

quality characteristics (responses, performance and outputs) of the

system

Noise Factors

They are the uncontrolled factors that cause Quality

characteristics to deviate from their target value.

1.

2.

3.

Inner Noise: Deterioration of parts, or variation on material

properties

Between Product Noise: Variation between different machines or

shifts

Robustness

Product and process designs that are insensitive to noise factors

are Robust.

Minimum Variability

System Design

It involves the development of a prototype design and determination of

materials, parts, components and assembly system.

Parameter Design

In this step, the levels (values) of controllable factors are selected to

minimize the effect of noise factors on the functional characteristics

of the product or the quality characteristics of the process.

Tolerance Design

To determine the allowable variation of the controlled parameters

without changing the quality.

10

11

Noise Factors

They are the uncontrolled factors that cause Quality characteristics to

deviate from their target value.

Robustness

System designs that are insensitive to noise factors are Robust.

12

Measurement of Quality

During Design/Development

Quality Loss

Function

Signal To

Noise Ratio

Parameters

Efficient

Experimentation

Using Design of

Experiments

(ANOVA)

Analysis of

Variances

13

m - 0

Loss is Proportional to the Square

Deviation From Target Value

m + 0

Loss is proportional to the square of the deviation from the target

value

L(y) = k(y-m)2

k=

0 = Tolerance (allowance)

A0 = Quality Loss at y = y0 = (m + 0)

k = Constant

(y - m) is deviation from the target value

15

Variations of the

Quality loss function

(QLF)

16

The nominal value (target) of a power supply circuit of a TV

set is m = 115 volts. When the voltage exceeds the range:

115 20 volts

The average cost of repairing is $100

a.

b.

resulted in producing circuits with output = 110 volts,

evaluate the quality of the production

17

L(y)

Loss ($)

Nominal the Best

0 = $100.00

0 = $100.00

0 = 20 V

95

L = ( )2

k = 02 =

0

100

20 2

115

0

135

y = voltage

= 0.25

= 0.25 ( 115) 2

18

Summary

22

breaker base is:

y = % shrinkage of the plastic base

(y 0)

cost of $80.

Two different plastics were tested. The following table shows the %

shrinkage values of the castings. Assuming the same cost in both cases,

evaluate the results and make recommendations.

Material

Data (% shrinkage)

A

0.28 0.24 0.33 0.30 0.18 0.26 0.24 0.33

B

0.08 0.12 0.07 0.03 0.09 0.06 0.05 0.03

23

Solution

Smaller the Better

k=

L= 35.55556 y2

35.55556

100

90

80

70

LOSS, $

60

50

40

30

20

10

0

0

0,2

0,4

0,6

0,8

SHRINKAGE %

1,2

1,4

1,6

1,8

Material B

k=

35.5555

L= 35.55556 y2

average B =

0.0663

Standard deviation square B = 0.0009

MSD B =

L=

0.0351

1.2468

Material A

k=

35.5555

L= 35.55556 y2

average A =

0.0729

Standard deviation square A =0.0026

MSD A =

L=

0.1239

4.4050

Conclusion

A

L=4.4050

B

L=1.2468

B ,So material B is better

An automobile manufacturer requires that the clearance between the cylinder

and the piston of a six cylinder engine be 3 (-2)(+7). Defect loss for each

cylinder and piston assembly is $200, and the monthly production is 50000

units. Data showing deviation from the target value for the first two months of

production are shown below. What are the quality levels during these two

months? What is the improvement, if any of the quality level?

Month

1

Deviations

-2

-2

-2

-1

-1

-2

-1

-1

-1

-2

-2

-2

-1

-3

cylinder engine be 3 (-2)(+7).

Solution

Nominal the Best

Unsymmetrical

m=3

For (y-m)>0

K=200/7^2= 4.081633

L= 4.081633(y-3)^2

For (y-m)<0

K=200/2^2= 50

L= 50 (y-3)^2

For (y-m)>0

K=200/7^2= 4.081633

L= 4.081633 (y-3)^2

For (y-m)<0

K=200/2^2= 50

L= 50 (y-3)^2

500

450

400

350

300

250

Chui1

200

150

100

50

0

-8

-6

-4

-2

Y-m

Results

Month1

Month2

Conclusion

There is improvement in the process during month 2

because the total loss of quality in month 2 is less than

loss of quality in month 1

The strength of an adhesive is usually determined by the kilograms force

(kgf) needed to break apart specimens joined by the adhesive. Two types

of adhesives, S_1 andS_2, which cost $50 and $60 per unit weight,

respectively, are to be compared. The lower specification limits

is 5 kgf

for the breaking force. The out-of-specification units are discarded,

resulting in a loss of $70 per unit. The annual production rate is 120,000

units. Sixteen units were tested for each type of adhesive, and the

following data for the breaking force were obtained:

Type of adhesive

10.2

5.8

4.9

16.1

15.0

9.4

4.8

10.1

14.6

19.7

5.0

4.7

16.8

4.5

4.0

16.5

7.6

13.7

7.0

12.8

11.8

13.7

14.8

10.4

7.0

10.1

6.8

10.0

8.6

11.2

8.3

10.6

Solution

K=70/5^2= 2.8

L= 2.8 (1/y^2)

0,35

0,3

L, $

0,25

0,2

0,15

0,1

0,05

0

0

10

15

Y-m

QLF for Joint Resistance

20

25

Results

S1

S2

Total loss per unit $ 0.064

Total loss of manufacture cost

$ 7674

Material cost $50

Total loss of manufacture cost + material

cost $7724

Conclusion

Even thought S1 is cheaper than S2, but the

total loss of quality is higher than S2

Total loss of manufacture cost

$ 3827

Material cost $60

Total loss of manufacture cost + material

cost $3887

Orthogonal Arrays

N = LF

N = number of experiments

L = number of levels

F = number of factors

A. Full Factorial Design

- All levels of all factors are investigated

- Examplepage 43

- The process has 4 factors, each with

3 levels: N = (3) 4 = 81 experiments

- If a fourth level is added:

N = (4)4 = 256 experiments

38

= 10 to 15 with 3 levels to be tested.

Inefficient for large number of factors

Interactions between factors are not included

Robustness is not explored

Only a few experiments that give the most important information

about the system are selected.

O.A. defines a balanced combination of factor levels to be studied.

39

Powder Metallurgy Sintering Process

Factors

A = Temperature

B = Sintering Time

C = Atmosphere (inert gas)

Levels

900o

1200o

12 hr

Type 2

8 hr

Type 1

Number of parameters = 3

Levels = 2

3

Full factorial design requires (2) = 8 experiments

Experiment run

Response (y)

y1

2

3

4

5

6

7

8

1

1

1

2

2

2

2

1

2

2

1

1

2

2

2

1

2

1

2

1

2

y2

y3

y4

y5

y6

y7

y8

Orthogonality means that for any pair of columns, all combinations of factor levels

occur and they occur an equal number of times.

40

Experimental Design

Degrees of Freedom

It measures the amount of information that can be obtained.

know which material is the strongest?

The material factor has 3 levels (M1,

M2, M3)

Comparison can be made with

respect to M1 (M2 > M1), (M3 < M1)

3 levels 2 comparisons

M has two degrees of freedom

41

independent statements that can be made about a

factor.

L = number of levels

Degree of freedom of an Orthogonal Array:

D.F. = N 1

D.F. (L4) = 4 - 1 = 3

(L8) = 8 - 1 = 7

42

Interaction

the level of others

43

Use Degree of Freedom

(D.F.) of O.A. = D.F. of all parameters and interactions

=N1

where N is the number of experiments or the order of O.A.

expected interactions:

A x B, C x D

Select a suitable O.A.

Number of Experiments =

D.F + 1 = 11

Sources

Number of Degrees of Freedom (df)

A

2-1

=

1

B

2-1

=

1

C

2-1

=

1

D

4-1

=

3

AxB

1x1

=

1

CxD

1x3

=

3

D.F.

=

10

44

Assigning Interactions

Interactions are located in specific columns to avoid confounding.

A = Temperature

Column 1

B = Sintering Time

Column 2

C = The Type of Environment will be Ignored

A x B = Interaction Between Temperature and

Time is Located in Column 3

45

Exp. #

1

2

3

4

AxB

1

1

2

2

1

2

1

2

1

2

2

1

If C is significant; therefore,

C and A x B are Confound.

y1

y2

y3

y4

Linear Graphs

assignment of factors & interactions

to an orthogonal array.

Factors are assigned to dots

the textbook

Sintering Process Using L4

factors is assigned to the line

connecting the two corresponding

dots

Factors may be assigned to lines,

not used for interactions.

46

Interaction Table

Interaction table is presented for each orthogonal

array to determine which column must assign

interaction.

Interaction column number is found where row (the

first factor) intersects column (the second factor).

The three column combination is interchangeable.

47

(a) () orthogonal array

Column

Expt.

No

1

2

3

4

5

6

7

8

Column

1

1

1

1

1

2

2

2

2

2

1

1

2

2

1

1

2

2

3

1

1

2

2

2

2

1

1

4

1

2

1

2

1

2

1

2

5

1

2

1

2

2

1

2

1

6

1

2

2

1

1

2

2

1

7

1

2

2

1

2

1

1

2

Factor Assignment

Column

1

2

3

4

5

6

7

1

(1)

2

3

(2)

3

2

1

(3)

4

5

6

7

(4)

5

4

7

6

1

(5)

6

7

4

5

2

3

(6)

7

6

5

4

3

2

1

(7)

which the interaction between every pair of columns

is confounded.

48

The experimental results are analyzed to determine the following:

a. Optimum process or product parameters

b. The expected response at the optimum combination of

parameters

c. The contribution of each factor process or product response

Method of Analysis

1. Evaluation of Quality Characteristic (Q.L.F.)

2. Main Effects (factor effects)

3. Interactions

4. Analysis of Variance (ANOVA)

5. Signal to Noise Ratio

49

Main Effects

Estimation of Factor Effects

= Response (measurement of quality characteristic y or

signal to noise ratio S/N)

Overall mean value of response:

n = number of experiments

Add the results of the experiments in which the factor and level occur

and divide by the number of these experiments.

Response Graph

Plotting main effects illustrates the results of each factor and level

and helps in selecting the optimum condition.

50

1.

Quality Characteristic y

Smaller the better

Larger the better

Nominal the best

2.

Optimum condition corresponds to maximum S/N

A model to estimate the expected results at the optimum

condition can be expressed by the following additive model:

= + + + + +

m = overall mean

= ( ), = ( )

e = error in repeatability of

51

Data collected from experiments are analyzed by ANOVA

to the variation of the results from the effective mean

Variance (data variability) can be decomposed into different

effects:

1.

2.

Amount of variation due to error

from the mean value

52

(ANOVA, contd)

Grand total sum of squares = = = sum of squares due to mean +

total sum of square

Sum of squares due to mean = n*

where: n = number of experiments

m = overall mean value

Total sum of squares = = ( )

= Sum of squares due to a factor x

= Total sum of squares of the deviation from the mean value for all

levels of that factor

(= ( ) + ( )+ )

= mean square due to factor =

(where = d.f of that factor = L 1)

Percentage contribution =

53

(ANOVA, contd)

Errors

If only one experiment is conducted at each experimental

condition, the error term is identically zero for each experiment.

In most cases, all columns are used for studying the process factors and

therefore no degree of freedom is left to estimate error variance.

(pooling method)

2

Error variance = =

corresponding to the factors having the lowest mean square.

54

(ANOVA, contd)

F calculated =

the random effect due to error

The calculated F is compared with F table at a desired level

of confidence (95% is often used)

1 value on F table = d.f of the factor

2 value on F table = d.f of error

F table is the value at the intersection of 1 and 2

If F calculated is greater than F table, the factor is significant

to that level of confidence

55

F distribution table

56

The Signal to Noise (S/N) ratio is an

objective measure of quality and of the effect of

noise factors on quality.

(S/N) takes both the mean and variation into

account.

Robustness is measured by (S/N) ratio.

(S/N) is similar to the quality loss function.

Higher S/N means higher performance and

smaller loss measured by Q.L.F.

L(Y) = K (MSD)

(S/N) = = -10 log (MSD)

57

- Smaller the better

L(Y) = K(MSD) = K

= -10 log

= -10 log ( + )

= mean =

= variance =

L(Y) = K(MSD) = K

S/N = = -10 log

(correct)

=( )

58

L(Y) = K(MSD) = K (

= K [( ) + ]

Where: t = target value

= mean value

= 10 log ( )

= mean =

= variance =

=( )

59

(P Diagram)

A block diagram representation of the parameters that influence the

quality characteristic or response of the product or the process.

60

Questions:

1. Which input affects the 2. What is the relationship

output parameter y?

between the important

inputs and the output

parameter y?

3. How can y be

controlled?

Input

raw materials,

components,

and subassemblies

Measurement

Evaluation

Control

Product

Process

Output product

y = Quality

characteristic

61

into three categories:

Planning the Experiment

1.

2.

3.

4.

optimized

Identification of the Control Factors and their alternate levels

Identification of the Noise Factors and testing conditions

Design the Matrix Experiments

5.

6.

7.

Determine optimum levels of the Control Factors and predict performance under

these levels (regular analysis and S/N analysis)

Conduct verification experiment and plan future actions

62

Exercise:

In order to optimize the process capability of an injection molding

process, the following factors were considered.

Control Factors:

A: Cool Time

B: Mold Temperature

C: Melt Temperature

D: Holding Pressure

E: Injection Speed

F: Holding Time

G: Injection Pressure

Noise Factors

H: Material Batch

I: Machine Type

J: Ambient Temperature

Level 1

A1

B1

C1

D1

E1

F1

G1

Level 2

A2

B2

C2

D2

E2

F2

G2

Level 1

Level 2

H1

I1

J1

H2

I2

J2

63

Quality Characteristic:

Percentage shrinkage

(smaller the better)

inner array and L4 for the outer array.

1. Calculate S/N ratios

2. Construct the response table and graphs

3. Determine the optimum process condition

64

Results

Control Factor

A

1

2

3

4

5

1

1

1

1

2

1

1

2

2

1

1

1

2

2

2

1

2

1

2

1

1

2

1

2

2

1

2

2

1

1

1

2

2

1

2

Noise Factors

1

1

2

1

2

1

1

2

2

2.20

2.10

2.30

0.30

2.50

2.70

0.50

3.10

0.40

2.00

1.90

1.80

3.00

3.10

3.00

2.10

4.20

1.00

3.10

7

8

2

2

2

2

1

1

1

2

2

1

2

1

1

2

4.00

2.00

1.90

1.90

4.60

1.90

2.20

1.80

H

I

J

2

2

1

2.30

0.30

2.80

2.00

3.00

L(Y) = K(MSD) = K

= -10 log

= -10 log ( + )

= mean =

= variance = =( )

Exercise Solution

Control Factor

1

2

3

4

5

6

7

8

1

1

1

1

2

2

2

2

1

1

2

2

1

1

2

2

1

1

2

2

2

2

1

1

1

2

1

2

1

2

1

2

1

2

1

2

2

1

2

1

1

2

2

1

1

2

2

1

1

2

2

1

2

1

1

2

Level

Noise Factors

H

1

1

I

1

2

J

1

2

2.20 2.10

0.30 2.50

0.50 3.10

2.00 1.90

3.00 3.10

2.10 4.20

4.00 1.90

2.00 1.90

S/N

2

1

2

2.30

2.70

0.40

1.80

3.00

1.00

4.60

1.90

2

2

1

2.30

0.30

2.80

2.00

3.00

3.10

2.20

1.80

Response Table

C

D

1

2

67

Solution, contd

Control Factor

1

2

3

4

5

6

7

8

1

1

1

1

2

2

2

2

1

1

2

2

1

1

2

2

1

1

2

2

2

2

1

1

1

2

1

2

1

2

1

2

1

2

1

2

2

1

2

1

1

2

2

1

1

2

2

1

1

2

2

1

2

1

1

2

Noise Factors

H

1

1

I

1

2

J

1

2

2.20 2.10

0.30 2.50

0.50 3.10

2.00 1.90

3.00 3.10

2.10 4.20

4.00 1.90

2.00 1.90

S/N

2

1

2

2.30

2.70

0.40

1.80

3.00

1.00

4.60

1.90

2

2

1

2.30 -6.95

0.30 -5.88

2.80 -6.98

2.00 -5.70

3.00 -9.62

3.10 -9.36

2.20 -10.73

1.80 -5.58

2.225

1.45

1.7

1.925

3.025

0.0092 1.77

2.1

0.0092

0.0025

2.6

3.175

1.9

Response Table

Level

-8.82381

-7.25

-7.9145

68

69

Final Solution

= 4.477

Y = 1.68 %

71

Case statement

The results of L16 orthogonal array experiments are given in the attached

table.

Calculate S/N.

Construct the response table and draw the response graph for the strong

effects.

Estimate the response at the recommended condition.

Conduct ANOVA for the results and comment on all results.

72

Case statement

73

Case analysis

74

Case analysis

75

Case analysis

76

- Surveying by Dr. RamachandraUploaded byvrsec-rao
- Plabstat ManualUploaded byJubair Al-rashid
- Thera-band Elastic Band Tension ReferenceUploaded byLuiz Santos
- Allometric Equation for Biomass Estimation of Cedrus Deodara (Deodar) in Northern PakistanUploaded byAnwar Ali
- LinestUploaded bycharu555
- Output Sig IikUploaded byDeniapt Alamsyah
- generalized additive modelUploaded byapi-285777244
- Central Composite Design for the Optimization of r AbsorbanceUploaded byAsif Fareed Qaisrani
- vvxcUploaded byDep En
- Output Yulianto 25 Maret 2019Uploaded byIhda EL-Mahfuzah
- Dm 26785792Uploaded byAnonymous 7VPPkWS8O
- ANOVA (QUALITY MANAGEMENT)Uploaded byyttan1116
- ch13Uploaded bySaied Aly Salamah
- Brm Power Point by NilaUploaded byShelveyElmoDias
- Download Fullpapers Akk60199870a9fullUploaded byAris Sudibyo
- exemplu_regresieUploaded byLivia Ionescu
- Caro 2013Uploaded byavishekkgec
- ANOVA 4 regressionUploaded bySamrah Siddiqui
- 11E Chapter 13Uploaded byslade
- Repeated MeasuresUploaded byFarizAgyan
- CVEN2002 Week12Uploaded byKai Liu
- Guideline for b.tech final year projectUploaded byVivek chaubey
- Hw JMPUploaded byAna Alejandra Gutierrez Ortiz
- CONSORTUploaded byshaza
- NPA of Private sectorUploaded bySharad Kumar
- factorial_anova_pdf.pdfUploaded bydarwin_hua
- Empirical Final2Uploaded byKaran Jhaveri
- MAGGI 1Uploaded byParvathi Aishwarya
- Appendix_E.pdfUploaded byBahman Homayun
- pppppUploaded byAlejandro Suáres Fuentes

- GeoRUploaded byLuis Zanxex
- prmia iiUploaded byLetsogile Baloi
- AgHbk232.pdfUploaded bySima Andreea
- sb3-hw13Uploaded byRichard Liu
- Rizzi-Ortuzar-JTEP-06Uploaded byphalanx
- Assignment 1 SolutionUploaded byKen
- AnovaUploaded byRone Redundant
- [Michael Greenacre] Biplots in PracticeUploaded bypoladores
- Solutions Manual Ch13 - 2012Uploaded bythegreatll
- ch07.docUploaded byAnonymous JmeZ95P0
- Assignment Statistics Behaviour of Sikkim Manipal University Semester 1Uploaded byLavaKumarK
- Sample SizeUploaded byHandhi Montok Setiawan
- Inferential Statistics for TourismUploaded byGeorges Otieno
- stationary120312.pdfUploaded bymadan kumar
- ch04Uploaded bySarjeel Ahsan Niloy
- Residual AnalysisUploaded byfansuri80
- Population Density and Spatial Distribution of Bean Bug Chauliops Fallax Sweet and Schaeffer on Different Legume Crops Under Mid Hills of Himachal PradeshUploaded bykiran
- Gas turbine performance prognostic for condition-based maintenance.pdfUploaded byEddy Pacheco
- RomUploaded byAroel Barabbaall
- A Meta-Analysis Comparing the Effect of Vaccines Against Mycoplasma Hyopneumoniae on Daily Weight Gain in PigsUploaded byThu Y 2007
- Syllabus CollegeUploaded byromelraj
- Hull_OFOD9e_MultipleChoice_Questions_Only_Ch23.docUploaded byguystuff1234
- 1010Uploaded byipatoff
- Reliability of BIDRUploaded byFlorin Roman
- Mult RegressionUploaded byiabureid7460
- SPSS Advanced Statistics 17.0Uploaded byraufurraheem
- 2015-GBZ-QQUploaded byEdson Bonfim
- CheatsheetUploaded byसुनिल बज्राचार्य
- Tutorial 1 Jan 2012Uploaded bynashful91
- Development of Sensory TestingUploaded byEmmae Thaleen