IE 3265R. Lindeke, Ph. D.

Quality Management in POM – Part 2

Topics Managing a Quality System

Total Quality Management (TQM) Achieving Quality in a System

Look early and often 6 Sigma – an approach & a technique Make it a part of the process

The Customers Voice in Total Quality Management QFD and the House of Quality

• Quality Engineering– Loss Function– Quality Studies– Experimental Approaches

• T.M.; FMEA; Shainin

Taguchi’s Loss Function

• Taguchi defines Quality Level of a product as the Total Loss incurred by society due to failure of a product to perform as desired when it deviates from the delivered target performance levels.

• This includes costs associated with poor performance, operating costs (which changes as a product ages) and any added expenses due to harmful side effects of the product in use

Exploring the Taguchi Method

• Considering the Loss Function, it is quantifiable– Larger is Better:

– Smaller is Better:

– Nominal is Best:

21( )L y ky

2( )L y ky

2( )

:

m is the target of the

process specification

L y k y m

where

Considering the Cost of Loss

• k in the L(y) equation is found from:

020

0

0

is cost of repair or replace

a product and must include

loss due to unavailability

during repair

is the functional limit on

y of a product where it would

fail to perform its function

half the

Ak

A

time

Loss Function Example: (nominal is best)

• We can define a processes average loss as:

• s is process (product) Standard Deviation

• ybar is process (product) mean

22L k s y m

Example cont.

• A0 is $2 (a very low number of this type!) found by estimating that the loss is 10% of the $20 product cost when a part is exactly 8.55 or 8.45 units

• Process specification is: 8.5+.05 units

• Historically: ybar = 8.492 and s = 0.016

Example Cont.

• Average Loss:

• If we make 250,000 units a year• Annual Loss is $64,000

2 22

2 0.016 8.492 8.500.05

800 .00032 $0.256

L

L

Fixing it• Shift the Mean to

nominal

• Reduce variation (s = 0.01)

• Fix Both!

22800 .016 0 $0.2048

Annual Loss is $51200 about 20% reduction

L

22800 .010 .008 $0.1312

Annual Loss is $32800 about 50% reduction

L

22800 .010 0 $0.08

Annual Loss is $20000 about 66% reduction

L

Taguchi Methods

• Help companies to perform the Quality Fix!– Quality problems are due to Noises in the product or

process system– Noise is any undesirable effect that increases

variability

• Conduct extensive Problem Analyses• Employ Inter-disciplinary Teams• Perform Designed Experimental Analyses• Evaluate Experiments using ANOVA and Signal-

to noise techniques

Defining the Taguchi Approach –

• The Point Then Is To Produce Processes Or Products The Are ROBUST AGAINST NOISES

• Don’t spend the money to eliminate all noise, build designs (product and process) that can perform as desired – low variability – in the presence of noise!

• WE SAY:ROBUSTNESS = HIGH QUALITY

Defining the Taguchi Approach –

• Noise Factors Cause Functional Variation

• They Fall Into Three “Classes”– 1. Outer Noise – Environmental Conditions– 2. Inner Noise – Lifetime Deterioration– 3. Between Product Noise – Piece To Piece

Variation

Taguchi Method isStep-by-

Step:

Defining the Taguchi Approach

• TO RELIABLY MEET OUR DESIGN GOALS MEANS: DESIGNING QUALITY IN!

• We find that Taguchi considered THREE LEVELS OF DESIGN:– level 1: SYSTEM DESIGN– level 2: PARAMETER DESIGN– level 3: TOLERANCE DESIGN

Defining the Taguchi Approach – SYSTEM DESIGN:

• All About Innovation – New Ideas, Techniques, Philosophies

• Application Of Science And Engineering Knowledge

• Includes Selection Of:– Materials– Processes– Tentative Parameter Values

Defining the Taguchi Approach – Parameter Design:

• Tests For Levels Of Parameter Values

• Selects "Best Levels" For Operating Parameters to be Least Sensitive to Noises

• Develops Processes Or Products That Are Robust

• A Key Step To Increasing Quality Without Increased Cost

Defining the Taguchi Approach – Tolerance Design:

• A "Last Resort" Improvement Step• Identifies Parameters Having the

greatest Influence On Output Variation

• Tightens Tolerances On These Parameters

• Typically Means Increases In Cost

Selecting Parameters for Study and Control

• Select The Quality Characteristic

• Define The Measurement Technique

• Ennumerate, Consider, And Select The Independent Variables And Interactions

• Brainstorming• Shainin’s technique where they are determined by

looking at the products• FMEA – failure mode and effects analysis

Preliminary Steps in Improvement Studies

• To Adequately Address The Problem At Hand We Must: 1. Understand Its Relationship With The Goals

We Are Trying To Achieve2. Explore/Review Past Performance compare

to desired Solutions 3. Prepare An 80/20 Or Pareto Chart Of These

Past Events4. Develop A "Process Control" Chart -- This

Helps To Better See The Relationship between Potential Control And Noise Factors

• A Wise Person Can Say: A Problem Well Defined Is Already Nearly Solved!!

Going Down the Improvement Road

• Start By Generating The Problem Candidates List:– Brainstorm The Product Or Process

• Develop Cause And Effects (Ishikawa) Diagrams

– Using Process Flow Charts To Stimulate Ideas

– Develop Pareto Charts For Quality Problems

DEVELOPING A Cause-and-Effect Diagram:

1. Construct A Straight Horizontal Line (Right Facing)

2. Write Quality Characteristic At Right

3. Draw 45° Lines From Main Horizontal (4 Or 5) For Major Categories: Manpower, Materials, Machines, Methods And Environment

4. Add Possible Causes By Connecting Horizontal Lines To 45° "Main Cause" Rays

5. Add More Detailed Potential Causes Using Angled Rays To Horizontal Possible Cause Lines

Generic Fishbone C&E Diagram

Methods Manpower

Materials Machines

EffectunderStudy

Environment

Main Causes

Primary Cause Primary Cause

2nd Cause 2nd Cause

2nd Cause

Building the ‘Experiment’ Working From a Cause & Effect Diagram

Fine GrainedChemical Yield

Raw Material

Reaction

TransportationMoistureContent

Catalyzer

CrystallizationPackage

Over Weight

Shortage of Weight

DischargeMethod

Sol. A Conc.

Sol. B Temp.

pHTime

Stir RPM

Sol APour Speed

Quality

Type

Quantity

Spillage

Road

Container

Cover

Time

Concentration

Temperature

Weight Size

Maint. OfBalance

Accuracy ofBalance

Operator

Type ofBalance

Method ofWeighing

‘Mother Crystal’

SteamPress.

SteamFlow

RPM ofDryer

Charge SpeedWet Powder

Temperature

Designing A Useful Experiment

• Taguchi methods use a cookbook approach!! Building Experiments for selected factors on the C&E Diagram

• Selection is from a discrete set of ‘Orthogonal Arrays’

• Note: an orthogonal array (OA) is a special fractional factorial design that allows study of main factors and 2-way interactions

T.M. Summary

• Taguchi methods (TM) are product or process improvement techniques that use DOE methods for improvements

• A set of cookbook designs are available – and they can be modified to build a rich set of studies (beyond what we have seen in MP labs!)

• TM requires a commitment to complete studies and the discipline to continue in the face of setbacks (as do all quality improvement methods!)

Simplified DOE

• Shainin Tools – these are a series of steps to logically identify the root causes of variation

• These tools are simple to implement, statistically powerful and practical

• Initial Step is to sample product (over time) and examine the sample lots for variability to identify causative factors – this step is called the multi-vari chart approach

• Shainin refers to root cause factors as the “Red X”, “Pink X”, and “Pink-Pink X” causes

ComponentsSearch

Multi-variCharts

PairedComparisons

Variables Search

Full Factorials

B vs. C

Scatter Plots

20 - 100 Variables

5 - 20 Variables

4 or LessVariables

Validation

Optimization

Shainin’s ‘Experimental Approaches’ to Quality Variability Control:

Shainin Ideas – exploring further

• Red X – the primary cause of variation

• Pink X – the secondary causes of variation

• Pink-Pink X significant but minor causes of variation (a factor that still must be controlled!)

• Any other factors should be substituted by lower cost solutions (wider tolerance, cheaper material, etc.)

Basis of Shainin’s Quality Improvement Approaches

• As Shainin Said: “Don’t ask the engineers, they don’t know, ask the parts”

• Contrast with Brainstorming approach of Taguchi Method

• Multi-Vari is designed to identify the likely home of the Red X factors – not necessarily the factors themselves

• Shainin suggests that we look into three source of variation regimes:• Positional• Cyclical• Temporal

Does the mean shift in time or between products or is the product (alone) showing the variability?

Positional Variations:

• These are variation within a given unit (of production)

• Like porosity in castings – or cracks• Or across a unit with many parts – like a

transmission, turbine or circuit board

• Could be variations by location in batch loading processes

• Cavity to cavity variation in plastic injection molding, etc.

• Various tele-marketers at a fund raiser

• Variation from machine-to-machine, person-to-person or plant-to-plant

Cyclical Variation

• Variation between consecutive units drawn from a process (consider calls on a software help line)

• Variation AMONG groups of units

• Batch-to Batch Variations

• Lot-to-lot variations

Temporal Variations

• Variations from hour-to-hour

• Variation shift-to-shift

• Variations from day-to-day

• Variation from week-to-week

Components Search – the prerequisites

• The technique is applicable (primarily) in ass’bly operations where good units and bad units are found

• Performance (output) must be measurable and repeatable

• Units must be capable of disassembly and reassembly without significant change in original performance

• There must be at least 2 assemblies or units – one good, one bad

The procedure:

• Select the good and bad unit• Determine the quantitative parameter

by which to measure the units• Dissemble the good unit – reassemble

and measure it again. Disassemble and reassemble then measure the bad units again. If the difference D between good and bad exceeds the d difference (within units) by 5:1, a significant and repeatable difference between good and bad units is established

Procedure (cont.)

• Based on engineering judgment, rank the likely component problems, within a unit, in descending order of perceived importance.

• Switch the top ranked component from the good unit to the bad unit or assembly with the corresponding component in the bad assembly going to the good assembly. Measure the 2 (reassembled) units.

• If there is no change: the good unit stays good bad stays bad, the top guessed component (A) is unimportant – go on to component B

• If there is a partial change in the two measurements A is not the only important variable. A could be a Pink X family. Go on to Component B

• If there is a complete reversal in outputs of the assemblies, A could be in the Red X family. There is no further need for components search.

Procedure (cont.)

• Regardless of which of the three outcomes above are observed, restore component A to the original units to assure original conditions are repeated. Then, repeat the previous 2 steps for the next most important components: B, C, D, etc. if each swap leads to ‘no’ or ‘partial’ change

• Ultimately, the Red X family will be ID’d (on complete reversal) or two or more Pink X or pale Pink X families if only partial reversals are observed

Procedure (cont.)

• With the important variables identified, a ‘capping run’ with the variables banded together as good or bad assemblies must be used to verify their importance

• Finally, a factorial matrix, using data generated during the search, is drawn to determine, quantitatively, main effects and interactive effects.

Paired Comparisons

• This is a technique like components search – but when products do not lend themselves to disassembly (perhaps it is a component in a component search!)

• Requires that there be several Good and Bad units that can be compared

• Requires that a suitable parameter can be identified to distinguish Good from Bad

Steps in Paired Comparison

1. Randomly select one “Good” and one “Bad” unit – call it pair one

2. Observe the differences between the 2 units – these can be visual, dimensional, electrical, mechanical, chemical, etc. Observe using appropriate means (eye, optical or electron microscopic, X-ray, Spectrographic, tests-to-failure, etc)

3. Select a 2nd pair, observe and note as with pair 1.

4. Repeat with additional pairs until a pattern of repeatability is observed between “goods & bads”

Reviewing:

• The previous (three methods) are ones that followed directly from Shainin’s “talk to the animals (products)” approach

• In each, before we began actively specifying the DOE parameters, we collect as much information as we can from good or bad products

• As stated by one user: “The product solution was sought for over 18 months, we talked to engineers & designers; we talked to engineering managers, even product suppliers – all without a successful solution, but we never talked to the parts. With the component search technique we identified the problem in just 3 days”

Taking the Next step: Variables Search

• The objective is to – Pinpoint the Red X, Pink X and one to three (more) critical

interacting variables – Its possible that the ‘Red X’ is due to strong interactions between

two or more variables– Finally we are still trying to separate the important variables from

unimportant ones

• Variables search is a way to get statistically significant results without executing a large number of experimental runs (achieving knowledge at reduced cost)

• It has been shown the this binary comparison technique (on 5 to 15 variables) can be successful in 20, 22, 24 or 26 runs vs. 256, 512, 1024, etc. runs using traditional DOE

Variables Search is a 2 stage process:

1. List the important input variables as chosen by engineering judgment (in descending order of ability to influence output)

2. Assign 2 levels to each factor – a best and worst level (within reasonable bounds)

3. Run 2 experiments, one with all factors at best levels, the second with all factors at worst levels. Run two replications sets

4. Apply the D:d 5:1 rule (as above)5. If the 5:1 ratio is exceeded, the Red X is

captured in the factor set tested.

STAGE 1:

Stage 1 (cont):

6. If the ratio is less than 5:1, the right factors are not chosen or 1 or more factors have been reversed between “best” & “worst” levels. Disappointing, but not fatal!

a. If the wrong factors were chosen – in opinion of design team – decide on new factors and rerun Stage 1

b. If the team believes it has the correct factors included, but some have reversed levels, run B vs. C tests on each suspicious factor to see if factor levels are in fact reversed

c. One could try the selected factors (4 at a time) using full factorial experiments – could be prone to failure too if interacting factors are separated during testing!

Moving on to Stage 2:1. Run an experiment with AW (a at worst level) and the

rest of factors at best levels (RB)a) If there is no change in best results in Stage 1 step 3, factor A is

in fact unimportantb) If there is a partial change from best results – toward Worst

results – A is not the only important factor. A could be Pink Xc) If a complete reversal in Best to Worst results in Stage 1 step 3,

A is the Red X

2. Run a second test with AB and RW

a) If no change from Worst results in Stage 1 the top factor A is further confirmed as unimportant

b) If there is a partial change in the worst results in Stage 1 – toward Best results – A is further confirmed as a possible Pink X factor

c) If a complete reversal – Best results in Stage 1 are approximated, A is reconfirmed as the Red X

Continuing Stage 2:

3. Perform the same component search swap of step 1 & 2 for the rest of the factors to separate important from unimportant factors

4. If no single Red X factor, but two or three Pink X factors are found, perform a capping or validation experiment with the Pink X’s at the best levels (remaining factors at their worst levels). The results should approximate the best results of Step 3, Stage 1.

5. Run a second capping experiment with Pink’s at worst level, the rest at Best level – should approx. the worst results in Step 3, Stage 1.

Variables Search Example:Press Brake Operation

• A press brake was showing high variability with poor CPK • The Press Brake was viewed as a “Black Magic” operation

– the worked sometimes then went bad ‘for no reason’• Causes of the operational variability were hotly debated,

Issues included:– Raw Sheet metal

• Thickness• Hardness

– Press Brake Factors (some which are difficult or impossible to control)

• The company investigated new P. Brakes but observed no realistic and reliable improvements– Even high cost automated brakes sometimes produced poor

results!

A Variables Search was Performed

• Goal was to consistently achieve a .005” tolerance (or closer!)

• 6 Factors were chosen:– A. Punch/Die Alignment – B: ‘Aligned’, W: ‘not Specially

Aligned’– B. Metal Thickness – B: ‘Thick’, W: ‘Thin’– C. Metal Hardness – B: ‘Hard’, W: ‘Soft’– D. Metal Bow – B: ‘Flat’, W: ‘Bowed’– E. Ram Storage – B: ‘Coin Form’, W: ‘Air Form’– F. Holding Material – B: ‘Level’, W: ‘Angle’

• Results reported in “Process Widths” which is twice tolerance, in 0.001” units

Results:

STAGE 1 Process Width (x.001)

All Best All Worst

Initial 4 47

Rep 1 4 61

D = 50; d = 7 D:d 7:1 (> 5:1) so a significant repeatable difference; Red X (or Pink X’s) captured as a factor

Continuing to Stage 2Test Comb. Results Conclusion

1 AWRB 3A. not Important

2 ABRW 102

3 BWRB 5B. Not Important

4 BBRW 47

5 CWRB 7C. Not Important

6 CBRW 72

7 DWRB 23 Pink X: Interaction w/ other factor(s)8 DBRW 30

9 EWRB 7???

10 EBRW 20

11 FWRB 73Prob. Red X + Interaction

12 FBRW 18

Cap Run DW FW RB 70Complete Reversal Effected

Cap Run DB FB RW 4

Factorial Analysis: D & F

D Best D Worst

F Best 4, 4, 3, 5, 7,

7, 4

Avg: 4.9

23, 18

Avg: 20.5

Row Sum: 25.4

F Worst 73, 20

Avg: 51.5

47, 102, 61

47, 72, 70,

20; Avg: 57.8

Row Sum: 109.3

Diagonal Sum: 72

Column Sum: 56.4

Column Sum: 78.3

Diagonal Sum: 62.7

Factorial Analysis:

20.5 51.8 4.9 51.5 78.3 56.4

2 210.95

51.5 57.8 4.9 20.5 109.3 25.4

2 241.95

72 62.7

24.7

1 2D. Sum D. Sum (interaction)

2

D

F

DF

Factorial Analysis:

• Factor G is Red X: It has a 41.9 main effect on the process spread

• Factor D is a Pink X with 10.9 main effect on process spread

• Their interaction is minor with a contribution of 4.9 to process spread

• With D & F controlled, using a holding fixture to assure level and reduction in bowing (but with hardness and thickness tolerances open up leading to reduced raw metal costs) the process spread was reduced to 0.004” (.002) much better than the original target of .005” with an observed CPK of 2.5!

Introduction to Failure Mode and Effects Analysis (FMEA)

• Tool used to systematically evaluate a product, process, or system

• Developed in 1950’s by US Navy, for use with flight control systems

• Today it’s used in several industries, in many applications – products – processes – equipment– software– service

• Conducted on new or existing products/processes• Presentation focuses on FMEA for existing process

Benefits of FMEA

• Collects all potential issues into one document– Can serve as troubleshooting guide– Is valuable resource for new employees at the process

• Provides analytical assessment of process risk– Prioritizes potential problems at process– Total process risk can be summarized, and compared to other

processes to better allocate resources• Serves as baseline for future improvement at process

– Actions resulting in improvements can be documented– Personnel responsible for improvements can gain recognition– Controls can be effectively implemented – Example: Horizontal Bond Process: FM’s improved by 40%;

causes improved by 37%. Overall risk in half in about 3 months.

FMEA Development• Assemble a team of people familiar with

process• Brainstorm process/product related defects

(Failure Modes) • List Effects, Causes, and Current Controls

for each failure mode• Assign ratings (1-10) for Severity,

Occurrence, and Detection for each failure mode– 1 is best, 10 is worst

• Determine Risk Priority Number (RPN) for each failure mode– Calculated as Severity x Occurrence x Detection

Typical FMEA Evaluation Sheet

Capturing The Essence of FMEA

• The FMEA is a tool to systematically evaluate a process or product

• Use this methodology to:– Prioritize which processes/ parameters/

characteristics to work on (Plan)– Take action to improve process (Do)– Implement controls to verify/validate

process (Check)– Update FMEA scores, and start focusing

on next highest FM or cause (Act Plan)