control charts for non-normal process

Upload: apoorv

Post on 06-Apr-2018

224 views

Category:

Documents


0 download

TRANSCRIPT

  • 8/3/2019 Control Charts for Non-normal Process

    1/46

    CONTROL CHARTS FOR

    NON-NORMALPROCESSES

    P-CHARTS

    np-CHARTS

    C-CHARTS

    U-CHARTS

  • 8/3/2019 Control Charts for Non-normal Process

    2/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    WHAT GENERATES ND OUTPUT?

    IF AN EVENT IS THE RESULT OF A RELATIVELY LARGENUMBER OFSMALL, CHANCE, INDEPENDENTINFLUENCES, THEN ITS OUTPUT WILL BE ND.

    MANY PROCESSES ARE ND BECAUSE:

    WE HAVE WORKED HARD TO ELIMINATE THE VERY LARGEINFLUENCES, THUS ONLY A RELATIVELY LARGE NUMBEROF SMALL, INDEPENDENT INFLUENCES REMAIN.

    WHAT IF A PROCESS IS NOT NORMALLY DISTRIBUTED?

    THAT IS OUR FOCUS HERE!

  • 8/3/2019 Control Charts for Non-normal Process

    3/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    Statistical Sampling--Data

    Attribute (Discrete, Go no-go information)

    Defectives--refers to the acceptability of product

    across a range of characteristics.

    Defects--refers to the number of defects per unit--

    may be higher than the number of defectives.

    Variable (Continuous)

    Usually measured by the mean and the standard

    deviation.

  • 8/3/2019 Control Charts for Non-normal Process

    4/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    THE BINOMIAL DISTRIBUTION

    Attribute (Go no-go information)

    We monitor the number of defectives over time.

    The relevant population parameter beingcontrolled is the population proportion Pie (P)

    We want to assure that the population

    proportion remains in control.

    We want to make sure the populationproportion does not become defective.

    The critical assumption is that Pie remains

    constant over time.

  • 8/3/2019 Control Charts for Non-normal Process

    5/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    DISTRIBUTION OF SAMPLE PROPORTIONS

    POP IS NOT ND

    = .98

    SAMPLE LOOKS

    LIKE POP,

    P= .99

    DIST. OF SAMPLE

    PS IS ND

  • 8/3/2019 Control Charts for Non-normal Process

    6/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    P-CHARTS

    Require large samples nu30.

    When population proportion is known:

    P=T s Z T(1 - T)/n

    When population proportion is unknown:

    _ _ _P=P s Z P(1 - P)/n

    Where P-Baris an estimate ofT

  • 8/3/2019 Control Charts for Non-normal Process

    7/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    Constructing a p-Chart

    No. of defective Monitors

    Sample n Defectives

    1 100 4

    2 100 2

    3 100 5

    4 100 35 100 6

    6 100 4

    7 100 3

    8 100 8

    9 100 1

    10 100 2

    11 100 3

    12 100 2

    13 100 2

    14 100 8

    15 100 3

  • 8/3/2019 Control Charts for Non-normal Process

    8/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    p =Total Number of Defectives

    Total Number ofObservations

    = p (1- p)

    npS

    UCL = p + Z

    LCL = p - Z

    p

    p

    s

    s

    P-CHART FORMULAS

  • 8/3/2019 Control Charts for Non-normal Process

    9/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    1. Calculate the sample proportion,

    p, for each sample.

    Defectives p

    4 0.04

    2 0.02

    5 0.05

    3 0.03

    6 0.06

    4 0.04

    3 0.03

    7 0.07

    1 0.01

    2 0.023 0.03

    2 0.02

    2 0.02

    8 0.08

    3 0.03

    Stephen A. DeLurgio and The McGraw-Hill Companies, Inc., 1998Irwin/McGraw-Hill

  • 8/3/2019 Control Charts for Non-normal Process

    10/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    2. Calculate the average of the sample proportions.

    0.037=1500

    55=p

    3. Calculate the standard deviation of the

    sample proportion

    .0188=

    100

    .037)-.037(1=

    n

    )p-(1p

    =

    sp

    The McGraw-Hill Companies, Inc.,Irwin/McGraw-Hill

  • 8/3/2019 Control Charts for Non-normal Process

    11/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    4. Calculate the control limits.

    3(.0188).037 sUCL = 0.093

    LCL = -0.0197 (or 0)

    p

    p

    sZ-p=LCL

    sZ+p=UCL

    The McGraw-Hill Companies, Inc.,Irwin/McGraw-Hill

  • 8/3/2019 Control Charts for Non-normal Process

    12/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    p-Chart (Continued)

    5. Plot the individual sample proportions, the average

    of the proportions, and the control limits

    You will be asked to duplicate these results using SPSS.

    What do you infer from the following control chart?

  • 8/3/2019 Control Charts for Non-normal Process

    13/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    Control Chart: PCHART

    Sigma level: 3

    15.00

    14.00

    13.00

    12.00

    11.00

    10.00

    9.00

    8.00

    7.00

    6.00

    5.00

    4.00

    3.00

    2.00

    1.00

    Pro

    portionNonconforming

    .10

    .08

    .06

    .04

    .02

    0.00

    PCHART

    UCL = .0942

    Center = .0373

    LCL = .0000

  • 8/3/2019 Control Charts for Non-normal Process

    14/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    Np-CHART

    Sometimes we express binomial occurrences in units.

    The formula is simply:

    Mean = nP-bar (e.g., 100*.037 = 3.7

    Were n = number in sample

    P = best estimate of the population proportion

    Sigma(np) = Sqrt(nP-bar(1-P-bar)

  • 8/3/2019 Control Charts for Non-normal Process

    15/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    DISTRIBUTION OF SAMPLE DEFECTS - nP

    POP IS NOT ND

    n = 100*.98 = 98

    SAMPLE LOOKS

    LIKE POP,

    P= 100*.99 = 99

    DIST. OF SAMPLE

    nPS IS ND

  • 8/3/2019 Control Charts for Non-normal Process

    16/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    nP TheorynP-CHARTS

    Require large samples nu30:

    When population proportion is known:

    nP = nT s Z nT(1 - T)When population proportion is unknown:

    _ _ _

    nP = nP s Z nP(1 - P)Where P-Baris an estimate ofT

  • 8/3/2019 Control Charts for Non-normal Process

    17/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    Constructing an np-Chart Step 1,

    Sample n Defectives = np

    1 100 4

    2 100 2

    3 100 5

    4 100 35 100 6

    6 100 4

    7 100 3

    8 100 8

    9 100 1

    10 100 211 100 3

    12 100 2

    13 100 2

    14 100 8

    15 100 3

  • 8/3/2019 Control Charts for Non-normal Process

    18/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    Calculate nP-bar

    nsObservatioSample

    DefectivesofNumberTotal=p

    )p1(pn=Snp

    np

    np

    sZ-pn=LCL

    sZ+pn=UCL

  • 8/3/2019 Control Charts for Non-normal Process

    19/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    2. Calculate the average of the sample proportions.

    3.7=150055*100=pn

    3. Calculate the standard deviation of the

    sample proportion

    1.89=.037)-3.7(1

    =

    )p-(1np=snp

    The McGraw-Hill Companies, Inc.,Irwin/McGraw-Hill

  • 8/3/2019 Control Charts for Non-normal Process

    20/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    4. Calculate the control limits.

    3(1.896)3.73sUCL = 9.42

    LCL =0

    np

    np

    sZ-pn=LCL

    sZ+pn=UCL

    The McGraw-Hill Companies, Inc.,Irwin/McGraw-Hill

  • 8/3/2019 Control Charts for Non-normal Process

    21/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    p-Chart (Continued)

    5. Plot the individual sample proportions, the average

    of the proportions, and the control limits

    You will be asked to duplicate these results using SPSS.

    What do you infer from the following np-chart?

  • 8/3/2019 Control Charts for Non-normal Process

    22/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    Control Chart: PCHART

    Sigma level: 3

    15.

    00

    14.00

    13.

    00

    12.00

    11.

    00

    10.00

    9.

    00

    8.00

    7.

    00

    6.00

    5.

    00

    4.00

    3.

    00

    2.00

    1.

    00

    NumberNonconfo

    rming

    10

    8

    6

    4

    2

    0

    PCHART

    UCL = 9.4207

    Center = 3.7333

    LCL = .0000

  • 8/3/2019 Control Charts for Non-normal Process

    23/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    DEFECTS PER UNIT OF TIME

    OR SPACE C Charts

    Frequently, there are processes that yield

    distributions that follow a Poisson Distribution.

    The Poisson Distribution is a discrete distribution

    which takes on the values X = 0, 1, 2, 3,...

    It models the events per unit of time or space.

    Determined by its mean,

    ___

    C

    ___

    C

  • 8/3/2019 Control Charts for Non-normal Process

    24/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    C-CHARTS POISSON DIST.

    In Quality Control the Poisson Distributionmeasures:

    Number of blemishes or defects per unit oftime or space such as:

    Blemishes per sqft. on painted panel.

    Defective ICs perWafer.

    Number of defects per sqft. on furnituresurface.

    Number of typos per page in your paper.

  • 8/3/2019 Control Charts for Non-normal Process

    25/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

  • 8/3/2019 Control Charts for Non-normal Process

    26/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    Consider Defective ICs on a Wafer

    How about 1 Giga Bit Chips

  • 8/3/2019 Control Charts for Non-normal Process

    27/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    http://www.austin.cc.tx.us/HongXiao/overview/history-semi/sld015.htm

  • 8/3/2019 Control Charts for Non-normal Process

    28/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    http://www.austin.cc.tx.us/HongXiao/

    overview/history-semi/sld016.htm

  • 8/3/2019 Control Charts for Non-normal Process

    29/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    POISSON

    GENERATING PROCESS

    Probability of an occurrence is very very low.

    No. of possible points is very very high.

    Prob. remains constant.

    Defined completely by its mean.

    Variance = Mean

    Standard Deviation = Sqrt(Mean) =sqrt( )

    A Skewed distribution to the right.

    High probability of low number, very low, but afinite probability of high number.

    ___

    C

    ___

    C

  • 8/3/2019 Control Charts for Non-normal Process

    30/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    POISSON AND EXPONENTIAL

    If the time or space between events follows

    an exponential distribution, then the rate of

    occurrence of the event will likely follow a

    Poisson.

    That is, the Poisson and Exponential

    Distribution are inverses of each other.

  • 8/3/2019 Control Charts for Non-normal Process

    31/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    Source:http://info.bio.cmu.edu/Course

    s/03438/PBC97Poisson/PoissonPage.

    html#distribution

  • 8/3/2019 Control Charts for Non-normal Process

    32/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    C-Chart

    Mean =

    UCL = + 3 sqrt( )

    LCL = - 3 sqrt( )

    Consider an example, a company measures

    the number of defects per square foot of

    expensive floor tile.

    Service example, mistakes made per hour per

    call center worker.

    ___

    C

    ___

    C

    ___

    C

    ___

    C

    ___

    C

  • 8/3/2019 Control Charts for Non-normal Process

    33/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    SAMPLE DEFECTS/SQFT

    1 5

    2 4

    3 7

    4 65 8

    6 5

    7 6

    8 5

    9 16

    10 10

    11 912 7

    13 8

    14 11

    15 9

    16 5

    17 7

    18 619 10

    20 8

    21 9

    22 9

    23 7

    24 5

    25 7

  • 8/3/2019 Control Charts for Non-normal Process

    34/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    SPSS DESCRIPTIVE STATISTICS

    Statistics

    .

    .

    .

    a

    .

    .

    .

    a

    ss ng

    N

    ean

    e an

    o e

    .

    ev a on

    ar ance

    um

    u p e mo es ex s . e sma es va ue s s own.

  • 8/3/2019 Control Charts for Non-normal Process

    35/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    SPSS DESCRIPTIVE STATISTICS

    POISSON

    . . .

    . . .

    . . .

    . . .

    . . .

    . . .

    . . .

    . . .

    . . .

    . .

    .

    .

    .

    .

    .

    .

    .

    .

    .

    o a

    Valid

    Cumulative

  • 8/3/2019 Control Charts for Non-normal Process

    36/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    SPSS DESCRIPTIVE STATISTICS

    POISSON

    16.014.012.010.08.06.04.0

    10

    8

    6

    4

    2

    0

    Std. Dev = 2.57

    Mean = 7.6

    N = 25.00

  • 8/3/2019 Control Charts for Non-normal Process

    37/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    C-Chart What Do You Infer?

    Control Chart: POISSON

    Sigma level: 3

    25.00

    23.00

    21.00

    19.00

    17.00

    15.00

    13.00

    11.00

    9.00

    7.00

    5.00

    3.00

    1.00

    Nonconform

    ities

    20

    10

    0

    POISSON

    UCL = 15.

    8086

    Center = 7.5600

    LCL = .0000

  • 8/3/2019 Control Charts for Non-normal Process

    38/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    U-CHART Nonconformities

    Per Unit

    Used in same situation as c-Chart but when

    sample sizes vary.

    This means area of time or space varies.

    Consider situation with different size panels

    or furniture.

    Service: Mistakes made for employees in a

    call center, but actual phone time varies.

    We want to control defects per square foot.

  • 8/3/2019 Control Charts for Non-normal Process

    39/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    SAMPLE SQMETERS NONCONF

    1 200 5

    2 300 14

    3 250 8

    4 150 8

    5 250 12

    6 100 6

    7 200 20

    8 150 10

    9 150 6

    10 250 10

    11 300 9

    12 250 16

    13 200 12

    14 250 10

    15 100 6

    16 200 8

    17 200 5

    18 100 5

    19 300 14

    20 200 8

    MEANS 205 9.6

  • 8/3/2019 Control Charts for Non-normal Process

    40/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    Whats the problem here?

    NONCONFO

    20.017.515.012.510.07.55.0

    7

    6

    5

    4

    3

    2

    1

    0

    Std.

    Dev = 4.

    06Mean = 9.6

    N = 20.00

  • 8/3/2019 Control Charts for Non-normal Process

    41/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    NONCONFORMITY/100 SQM

    Looks more like a Poisson

    NONP100

    20.018.016.014.012.010.08.06.0

    6

    5

    4

    3

    2

    1

    0

    Std. Dev = 3.56

    Mean = 9.9

    N = 20.00

  • 8/3/2019 Control Charts for Non-normal Process

    42/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    u-Chart for Carpet Data

    Mitra page 354

    Control Chart: NONCONFO

    Sigma level: 3

    20.00

    19.00

    18.00

    17.00

    16.00

    15.00

    14.00

    13.00

    12.00

    11.00

    10.00

    9.00

    8.00

    7.00

    6.00

    5.00

    4.00

    3.00

    2.00

    1.00

    Fractionof

    Nonconformities

    .12

    .10

    .08

    .06

    .04

    .02

    0.00

    NONCONFO

    UCL

    Center = .0468

    LCL

  • 8/3/2019 Control Charts for Non-normal Process

    43/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    So:http://www.mathsrevision.net/alev

    el/statistics/normal_distribution2.php

    Poisson Approximation

    The normal distribution can also be used to

    approximate the Poisson distribution for largevalues of C (the mean of the Poisson

    distribution).

    If X ~ Po(C) then for large values of C, X ~

    N(C, C) approximately. This last statementdenotes that X is ND with mean of C and

    variance of C.

  • 8/3/2019 Control Charts for Non-normal Process

    44/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    ND APPROXIMATIONS FOR

    BINOMIAL

    For Binomial, the higher the value of N andthe closer p is to .5, the better the NDapproximates the Binomial. (See link on next

    slide.) When n*p nd n*(1-p) are muchgreater than 5 then the ND approximates theBinomial Dist.

    For large n (say n > 20) and p not too near 0

    or 1 (say 0.05 < p < 0.95) the distributionapproximately follows the Normal distribution.

  • 8/3/2019 Control Charts for Non-normal Process

    45/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    SIMULATING APPROXIMATIONS

    Fun with statistics:

    Normal Approximation of Binomial

    http://www.ruf.rice.edu/~lane/stat_sim/binom_demo.html

  • 8/3/2019 Control Charts for Non-normal Process

    46/46

    Stephen A. DeLurgio and MGraw-Hill, 2004 all rights reserved.

    So:http://www.mathsrevision.net/alev

    el/statistics/normal_distribution2.php

    Binomial Approximation

    The normal distribution can be used as an

    approximation to the binomial distribution, under

    certain circumstances, namely:

    If X ~ B(n, p) and if n is large and/or p is close to ,

    then X is approximately N(np, npq)

    (where q = 1 - p).

    In some cases, working out a problem using the

    Normal distribution may be easier than using a

    Binomial.