six sigma black belt certification demo week2

Upload: george-farrugia

Post on 04-Apr-2018

216 views

Category:

Documents


0 download

TRANSCRIPT

  • 7/30/2019 Six Sigma Black Belt Certification Demo Week2

    1/27

    Qi2, 1995-2002

    Brought To you by

    sixsigmatutorial.com

    1Brought To you by sixsigmatutorial.com

    http://localhost/var/www/apps/conversion/tmp/scratch_8/sixsigmatutorial.comhttp://localhost/var/www/apps/conversion/tmp/scratch_8/sixsigmatutorial.com
  • 7/30/2019 Six Sigma Black Belt Certification Demo Week2

    2/27

  • 7/30/2019 Six Sigma Black Belt Certification Demo Week2

    3/27

    OVER

    PROCESSING

    INVENTORY

    MOTION

    MOVE IT OVER

    THERE UNTIL WE

    NEED IT

    CONVEYANCE

    CORRECTION

    OVER PRODUCTION

    WAITING

    The 7 MajorWastes

    Source: Ford 6 S Training Course

    3Brought To you by sixsigmatutorial.com

  • 7/30/2019 Six Sigma Black Belt Certification Demo Week2

    4/27

    EXERCISE

    Assembling two parts together

    Searching for information

    Making initial fixture setup

    Transporting materials to next station

    Capturing data once at the source

    Examining castings for defectsWalking to parts storage for assembly screws

    Storing a lot on a movable rack

    Readjusting the fixture setup

    Trouble-shooting a rejected assembly

    QC final inspection

    Leaving a form in an in-basket

    Attending a benefits meeting

    VA NVA

    For each task, select whether it is value

    added or non value added

    4Brought To you by sixsigmatutorial.com

  • 7/30/2019 Six Sigma Black Belt Certification Demo Week2

    5/27

    SETUP TIME REDUCTION

    The objective is to facilitate small-lot production,

    notto reduce setup personnel headcount/labor

    The people involved in the setup should do their ownsetup time reduction

    - Look at the setup as a process

    - Use videotape recording, with a built-in time clock(down to seconds)

    The team should follow a 4-step process:Analyze the complete videotaped setup and

    classify the time elements into 2 categories:

    1. Internaltime

    2. Externaltime

    Eliminate the external downtime

    Reduce the remaining internal setup time withbetter methods; then, practice on them

    Eliminate as many adjustments as possibleMore

    5Brought To you by sixsigmatutorial.com

  • 7/30/2019 Six Sigma Black Belt Certification Demo Week2

    6/27

    USING TESTS OF STATISTICAL SIGNIFICANCE

    It is not always possible to draw valid conclusions about a

    population just by looking at sample data. We need to

    make sure any conclusions we draw about the underlyingpopulation are truly supported by the data.

    You may ask yourself, for example,

    Is our root-cause hypothesis proven based on the data we have collected?

    Are these two groups of sample data from the same or different underlying

    populations? = ?

    Has the proportion of defectives truly changed from what it used to be? p1 p0?

    Do these two processes have the same amount of variation? = ?

    Is there any significant difference in the quality characteristics between these

    three vendors? p1 = p2= p3?

    1

    2

    2

    1 2

    2

    6Brought To you by sixsigmatutorial.com

  • 7/30/2019 Six Sigma Black Belt Certification Demo Week2

    7/27

    TWO - TAIL TESTS

    Use a two- tail test if you are testing whether a sample value(statistic) is different (whether bigger or smaller) than anestablished (generally a population) value

    a The risk is divided in twoand placed on both ends of adistribution of test statistics:

    in bothits previous value or is significantly different than another

    of the existing or reference distribution0).the right and left tails (the Rejection Regions, for rejecting H

    populations value, the critical region will be divided

    To test whether a statistic has changed significantly from

    curve.

    0

    a

    H : value 1 = value 2

    H : value 1value 2

    2

    The Acceptance Region,

    for accepting H 0

    2

    7Brought To you by sixsigmatutorial.com

  • 7/30/2019 Six Sigma Black Belt Certification Demo Week2

    8/27

    DECISION TREE ON 2 MEANS TESTS

    with known variance

    or with n > 30

    Compare sample

    mean to a specified

    value

    with unknown &

    assumed equal

    variances, and

    n 30

    Compare means

    from 2 samples

    Compare sample

    mean to a specifiedvalue

    Compare means

    of paired sets ofdata (e.g., on mult.

    samples from 2

    sources)

    with unknown &

    not assumed equal

    variances, & n 30

    Compare means

    from 2 samples

    The Non-Parametric alternatives to the 2 Means tests are the Sign Test on 2 Medians,

    the Rank Sum Test (Wilcoxon) on 2 Means, and the Mann-Whitney U Test. p. 68

    p. 42

    p. 47 p. 51

    p. 54

    p. 58

    Back to Means Test Exercise

    8Brought To you by sixsigmatutorial.com

  • 7/30/2019 Six Sigma Black Belt Certification Demo Week2

    9/27

    = 9.37

    The AcceptanceRegion, for

    accepting H0

    + 1.645x

    = 10.11

    The Rejection Region,for rejecting H0 (5%)

    +2.71x

    =10.59

    Location of thetest statistic

    .

    Z TEST ON TWO MEANS (Contd.)

    The actual p value, i.e., the probability that if H0 is true, we will

    observe the statistic x deviating by chance from the parameter being

    tested ( = 9.37) by a greater degree than is observed, is .0034.

    This is the area under the curve to the right of Z = 2.71.to Z Table

    9Brought To you by sixsigmatutorial.com

  • 7/30/2019 Six Sigma Black Belt Certification Demo Week2

    10/27

    t TEST ON TWO MEANS

    (With Unknown but Assumed Equal Variances, & n 30)

    Purpose:

    To compare the mean from a sample to the known orspecified mean of a population to see if the difference

    between them should be considered significant (at a

    selected a level of risk)

    Assumptions:

    1. Both distributions are normal, and the samples areindependent and randomly taken

    2. 1 = 2 (unknown but assumed equal; a test on thetwo variances is often applied first in order to assess

    this assumption).

    example next

    10Brought To you by sixsigmatutorial.com

  • 7/30/2019 Six Sigma Black Belt Certification Demo Week2

    11/27

    MINITAB EXERCISE APPLYING TESTS ON 2 MEANS (Contd.)

    2. File > Open Worksheet \Data\exh_stat.mtw (no 1-sample tests in Excel)

    Test to see if the likely population mean material thickness (Values inCol. 1) is less than the historical average of 5.0 mils. Use a 90%

    confidence level. is not known. Also, create a indiv. value plot of the

    data as part of the test.

    What is the attained significance level (p-value) ?

    Does this suggest that the new population mean

    is less than 5.0?

    What does the plot show?

    .017

    Yes

    The C.I. does not contain H0 (5.0).

    Minitab

    to next t test

    11Brought To you by sixsigmatutorial.com

  • 7/30/2019 Six Sigma Black Belt Certification Demo Week2

    12/27

    ANOVA TEST ON 3 MEANS

    Purpose:

    To compare the means from 3 or more samples to see if the

    difference between them should be considered significant

    (at a selected a level of risk)

    Assumptions:

    We use an F test to see whether the 3 or more sample means

    underlying populations are different. This test requires that we

    assume homogeneous variances and that the measurements

    were taken randomly from normally distributed populations.

    If the sample sizes are equal, slight departures from the

    assumption of equal variances are acceptable.

    Example:

    Three molding machines are being studied to see if their average molded

    part dimensions are the same, with the following samples:

    Machine A: .450, .458, .449, .451, .455; B: .456, .461, .458, .451, .463C: .464, .467, .463, .459, .462 (see next slide for Minitab instructions)

    13Brought To you by sixsigmatutorial.com

  • 7/30/2019 Six Sigma Black Belt Certification Demo Week2

    13/27

    In Minitab:

    Enter the 15 values in a column; Enter the machine ID letters

    (A, B, or C) in another column to match up with the response data

    Stats\ANOVA\One-way

    In the 1-way Analysis of Variance dialog box:

    Response: select the column that contains the 15 test data.

    Factor: select the column that contains the machine ID letters.Graphs: select boxplot of data; click on OK

    What does the output tell you?

    Save the worksheet for use later with ANOM & multiple variances test

    The boxplot shows machines A and C have different means. The small p value

    in the ANOVA shows there is a statistically significant difference between at least2 of the machines. The C.I.s for machines A and C do not overlap.

    14Brought To you by sixsigmatutorial.com

  • 7/30/2019 Six Sigma Black Belt Certification Demo Week2

    14/27

    DETERMINING SAMPLE SIZE

    When estimating the population mean forvariabledata from a

    normaldistribution,

    n =

    where = the table Z value for the desired

    level of confidence

    = the standard deviation of the normal

    population from which we are to

    select our sample, and

    e = the specified amount of error between

    and ; the allowable differencebetween them

    a

    2

    2

    Z

    e

    a2

    Z

    x

    We will be 1 - % confident that our estimate of will

    be off by no more than e; there is a % chance that

    we think is not between e and it actually is

    a

    a

    x

    15Brought To you by sixsigmatutorial.com

  • 7/30/2019 Six Sigma Black Belt Certification Demo Week2

    15/27

    DETERMINING SAMPLE SIZE (Contd.)

    Example

    If an initial sample of 32 product samples yielded a mean ( ) of 24.3 mm

    and a standard deviation (S) of 1.0 mm, what total sample size do we

    need to be 95% confident that our estimate of will be off by .1 mm orless (i.e., it is within 24.2 and 24.4)? As long as the initial sample size is

    30, we can use the samples S as an estimate of. The Z value for a .95

    area under the curve (two-tail) is 1.96.

    x

    Note: The beta risk defaults to 50% when it is not included

    in the sample size calculation.

    21.96x1.0

    n 384.2 385.1

    to Z Table

    16Brought To you by sixsigmatutorial.com

  • 7/30/2019 Six Sigma Black Belt Certification Demo Week2

    16/27

    Column A B CCondition 1 2 3 Data

    1 1 1 1

    2 1 2 2

    3 2 1 2

    4 2 2 1

    L4 (23)

    This design can be used to study 3 factors at 2 levels

    each but not all combinations are tested . . .

    Qi2, 2002

    Experimental

    Runs

    Factors

    Responses

    Levels

    17Brought To you by sixsigmatutorial.com

  • 7/30/2019 Six Sigma Black Belt Certification Demo Week2

    17/27

    CALCULATING MAIN EFFECTS (Contd.)

    To begin the analysis of the data, you need to isolate (or separate) the data

    that corresponds to each level of each factor:

    Column A B C Response

    Condition 1 2 3 Data (Yi)

    1 1 1 1 3

    2 1 2 2 5

    3 2 1 2 6

    4 2 2 1 4

    L4 (23)

    The average of A at level 1 ( ) is the average of the data from

    experimental run conditions 1 and 2 (Why?):1

    A

    1A =

    Cond.1 Cond.2

    n

    = = 4.0

    3 5

    2

    19Brought To you by sixsigmatutorial.com

  • 7/30/2019 Six Sigma Black Belt Certification Demo Week2

    18/27

    A1 A2 B1 B2 C1 C2 X

    Y

    6

    4

    2

    PLOTTING OF MAIN EFFECT RESPONSE GRAPHS

    Now, in order to visualize our main effect calculations, we plot

    them on response graphs:

    Plot the response graphs for factors B and C above, too.

    If we wanted a smaller-is-better response, which level is

    better for each factor? ans.

    Which factor is the most significant? The least?

    Why?

    A1 B1 or B2 C1

    C B

    Size of the delta between the levels (slope)

    4.0

    5.0

    Note how we plot all 3

    responses on one graph

    to make the interpretation

    more convenient

    20Brought To you by sixsigmatutorial.com

  • 7/30/2019 Six Sigma Black Belt Certification Demo Week2

    19/27

    low oven temp (B1)

    X

    slow conveyor spd fast conveyor spd

    A1 A2

    30

    20

    10

    40

    Y

    Cure

    Quality

    An interaction exists when a change in the level of one factor

    effects how another factor influences the measured output.

    For example, if Conveyor Speed is slower (A1), the effect of Oven Temperature, when

    changed from high (B2) to low (B1), has a positive effect on the quality of cure; but when

    Conveyor Speed is faster (A2), the effect of Oven Temperature changing from high to low

    is negative on the quality of cure.

    WHAT IS AN INTERACTION?

    Qi2, 2002

    + -

    high oven temp (B2)

    21Brought To you by sixsigmatutorial.com

  • 7/30/2019 Six Sigma Black Belt Certification Demo Week2

    20/27

    1 3 2 5 4 7 6

    2 1 6 7 4 5

    3 7 6 5 4

    4 1 2 3

    1 2 3 4 5 6 7

    6 1

    7

    35 2

    A

    BA X B

    Triangular Table for the L4 and L8 2-level OAs:

    USE OF TRIANGULAR TABLES

    FOR DESIGNING INTERACTIONS

    How would you assign C X D in addition to A X B?

    Qi2, 2002

    Cols

    If we want to study or isolate an

    interaction, we need to assign it to

    a column in an array (instead of

    assigning a main effect). Thistable was developed by Dr.

    Taguchi to help assign interactions

    to the correct columns in an L4

    and in an L8. This table shows the

    interacting column relationships.

    The bold numbers along the top

    and along the diagonal are columnnumbers to which the main effects

    (e.g., A and B) would be assigned.

    The numbers in the inside of the

    table are the columns to which the

    interaction (AXB, or AB) would be

    assigned.

    Once you assign the main effects to

    their columns (5 and 6 for example),

    then their interaction can only go in

    one place (column 3), according to

    where they intersect in this table.

    Note: you cannot find another set of 3 interacting columns that do not use

    columns 3, 5, or 6. Try it.

    back to interaction effect proof 22Brought To you by sixsigmatutorial.com

  • 7/30/2019 Six Sigma Black Belt Certification Demo Week2

    21/27

    Column B C C A D E F

    Condition 1 2 3 4 5 6 7 Data

    1 1 1 1 1 1 1 1

    2 1 1 1 2 2 2 2

    3 1 2 2 1 1 2 2

    4 1 2 2 2 2 1 1

    5 2 1 2 1 2 1 26 2 1 2 2 1 2 1

    7 2 2 1 1 2 2 1

    8 2 2 1 2 1 1 2

    B

    X

    ANALYSIS OF AN INTERACTION EXAMPLE (Contd.)

    Next, we average the response data for each of the four factor-level

    combinations of the interaction.

    38

    42

    23

    17

    21

    24

    15

    19

    B1C2: B2C1: B2C2:(23+17)/2 =20.0 (21+24)/2=22.5 (15 + 19)/2 = 17.0

    The level average ofB1C1 is (38 + 42) 2 = 40.0.

    What are the level averages (& formulas) forB1C2, B2C1, and B2C2?

    B1C1 bar is calculated

    by averaging together

    the data for conditions 1

    and 2, where, and only

    where, both factors B

    and C were tested at

    level 1.

    Next we will graph these 4 plot points for the interaction.

    23Brought To you by sixsigmatutorial.com

  • 7/30/2019 Six Sigma Black Belt Certification Demo Week2

    22/27

    DESIGN LAYOUT L8 (27) X F2

    Column A B C D E F G Response Data

    Condition 1 2 3 4 5 6 7 R/M Lot 1 R/M Lot 2

    1 1 1 1 1 1 1 1 Y1, Y2, Y3 Y4, Y5, Y62 1 1 1 2 2 2 2

    3 1 2 2 1 1 2 2

    4 1 2 2 2 2 1 1

    5 2 1 2 1 2 1 2

    6 2 1 2 2 1 2 1

    7 2 2 1 1 2 2 1

    8 2 2 1 2 1 1 2 Y43, Y44, Y45 Y46, Y47, Y48

    Is this is a 1-, 2-, or 3-way design?

    Why?

    How might it be used?

    2-way

    2 groups 1 inside OA and 1 outside factor

    parameter design

    24Brought To you by sixsigmatutorial.com

  • 7/30/2019 Six Sigma Black Belt Certification Demo Week2

    23/27

    Column

    Condition 1 2 3 4 5 6 7

    1 1 1 1 1 1 1 1

    2 1 1 1 2 2 2 2

    3 1 2 2 1 1 2 2

    4 1 2 2 2 2 1 1

    5 2 1 2 1 2 1 2

    6 2 1 2 2 1 2 1

    7 2 2 1 1 2 2 18 2 2 1 2 1 1 2

    RANDOMIZATION

    It is generally recommended to run your

    conditions in random order. The reasonwhy will now be demonstrated. Lets say

    you decide to run in standard order, which

    is 1, 2, 3, 4, and so on. And lets further

    say that you can only run the first 4

    conditions in the cool, dry morning and the

    last 4 in the hot, humid afternoon. Next,

    you collect data and analyze it, to find outthat the first column was the most

    significant. Can you be sure it was the

    factor assigned to that column or the

    ambient conditions that caused the

    changes in the measured response?

    Running in a random order would minimize(but not eliminate) that risk of an unwanted

    factor coinciding with the levels of a

    column. Hopefully, the noise would be

    scattered among the levels of the factors.

    25Brought To you by sixsigmatutorial.com

  • 7/30/2019 Six Sigma Black Belt Certification Demo Week2

    24/27

    GRAND, EXPECTED AND LEVEL AVERAGES

    32.0

    29.0

    28.5

    25.5

    27.5

    30.0

    B2

    B2

    A1 C1A1

    C1

    A1 A2 B1 B2 C1 C2

    = 33.5

    T

    n= 28.75

    1. Graphical Solution:

    2. Calculated Solution:

    (e.g., AOPT. = A1 - )T

    n

    = 28.75 + .25 + 3.25 + 1.25 = 33.5

    = + (AOPT. - ) + (BOPT. - ) + (COPT. - ) + T

    n

    T

    n

    T

    n

    T

    n

    26Brought To you by sixsigmatutorial.com

  • 7/30/2019 Six Sigma Black Belt Certification Demo Week2

    25/27

    A 2 2.00 1.00

    B 2 .67 .33

    C 2 50.67 25.33

    D 2 8.67 4.33

    (e) 0

    e2 0

    Total 8 62.00

    ANOVA BEFORE POOLING

    Sum of Mean F Percent

    Factor df Squares Squares Ratio Contrib.

    S V F

    4 2.67

    At this point we have no way to estimate the denominator of the F ratio, the within

    variation. For this we will have to pool one or more of the least significant columns

    into error to come up with an error estimate. Note that experimental error refers tounexplained variation, not any kind of a mistake. For this analysis, lets pool factors

    A and B into pooled error, (e). We will pool both the degrees of freedom and the

    sum of squares for the insignificant column(s).

    27Brought To you by sixsigmatutorial.com

  • 7/30/2019 Six Sigma Black Belt Certification Demo Week2

    26/27

    USING STATIC SIGNAL-TO-NOISE RATIOS TO MEASURE VARIATION

    Among many available Static S/N ratios are three of the most common:

    1. Nominal is Best (Case I), the most common S/N ratio

    N = +10 log10 (MSD)N

    where (MSD)N =m e

    e

    (S V )1

    n V

    28Brought To you by sixsigmatutorial.com

  • 7/30/2019 Six Sigma Black Belt Certification Demo Week2

    27/27

    INTRODUCTION TO DESIGN OF EXPERIMENTS

    Lesson 6 Intermediate Analysis of Results II

    Objectives

    Attendees will be able to

    - Calculate ANOVA values for a simple design

    - Design and analyze an experiment using parameter design

    - Calculate static S/N ratios for a parameter design

    - Identify and use tuning factors in a 2-step analysis

    - Use response tables

    29Brought To you by sixsigmatutorial.com