Automated Visual Inspection

Embed Size (px)

Citation preview

  • 8/11/2019 Automated Visual Inspection

    1/24

    IEE 572 DESIGN OF ENGINEERING EXPERIMENTS

    FALL 2000

    FINAL REPORT

    TERM PROJECT

    AN EXPERIMENTAL DESIGN FOR IMPROVING THE ACCURATE

    CLASSIFICATION OF THE IMAGES OF PRESENT AND ABSENT

    COMPONENTS IN PRINTED CIRCUIT BOARDS WHEN INSPECTED

    BY AN AUTOMATED VISUAL INSPECTION SYSTEM

    By

    PRAVEEN BABU

    SREENIVASULA REDDY KATHA

    LUIS MAR

    0

  • 8/11/2019 Automated Visual Inspection

    2/24

    INTRODUCTION

    The current miniaturization of the assembly surface mounted devices (SMD) in

    Printed Circuit Boards (PCB) makes it a difficult task for human insectors to

    determine their correct ositionin! and even their resence in the board" This task

    becomes even more difficult #hen the roduction cycle time for the PCBs is

    small$ leavin! the insectors #ith not enou!h time to accurately comlete their

    %ob"

    Fi!"# $& Picture of a PCB" 'ote the size ofComonents comared to a ten cents coin"

    The tyical solution to this roblem is the introduction of an utomated isual

    *nsection (*) system to relace the human insectors" The imminent

    advanta!e of these systems over human insectors is their caability of reliably

    insectin! a lar!e number of small comonents at a fi+ed level of erformance for

    lon! eriods of time ,-." /i!ure describes a tyical PCB assembly line&

    -

  • 8/11/2019 Automated Visual Inspection

    3/24

    bare PCB enters the !lue disenser #here !lue is laced to the ositions in the

    board #here comonents #ill later be mounted" The ne+t oeration is the

    lacement of the SMD to the board" This oeration is erformed by chi1shooters

    #hich oerate in tandem (lacement machine)" The comonents are fed to the

    chi1shooter by a feeder$ #hich is an array of rolls containin! the comonents to

    be laced in the PCB" 2nce the comonents have been fed to the chi1shooter it

    laces them on the board usin! vacuum nozzles" fter the comonents have been

    laced on the board the * machine insects the resence and correct lacement

    of the comonents" ,.

    The 3lectronics ssembly 4aboratory (34) at rizona State 5niversity is

    develoin! an inte!rated 6uality environment for SMD assembly" This

    environment consists of several modules that interact #ith each other to determine

    in real time the 6uality of the assembly rocess for SMDs and hence the 6uality of

    the roduct" These modules are&

    1D

    *nsectionSystem

    7lueDisenser

    Machine

    Placement

    Machine

    /eeder

    'ozzle

    (Piette)

    PCB

    Fi!"# 2% PCB ssembly 4ine

  • 8/11/2019 Automated Visual Inspection

    4/24

    *nsection llocation Module (*M)

    8uality Monitorin! Module (8MM)

    *nsection Module (*M)

    The *M instructs the different insection systems as to #hich elements to insect

    considerin! the total available insection time$ the insection time re6uired by

    each comonent and the needs of the 8MM amon! others"

    The 8MM uses the information rovided by the *M to determine if the SMD

    assembly rocess is in state of statistical 6uality control"

    The insection module (/i!ure 9) is in char!e of the ac6uisition$ in real time$ of

    the rocess information" *t consists of four Pulini+ TM ::0 monochromatic

    cameras #ith a resolution of ;-0+:

  • 8/11/2019 Automated Visual Inspection

    5/24

    Fi!"# && Side vie# of * system

    PROBLEM DEFINITION AND FACTORS TO BE CONSIDERED

    The ima!es rendered by the insection system are the basis for the decision

    makin! of the entire 6uality control system" *n order for these ima!es to be useful

    they must rovide sufficient level of detail so that the location$ insection and

    control chart lottin! al!orithms can make a decision" This decision consists of

    determinin! #hether the comonents are resent or absent and if they are resent$

    determinin! their location and an!le of rotation #ith resect to an ideal osition

    #ithin the board" The 6uality of the ima!e rovided by the cameras in every

    insection routine deends !reatly on three factors&

    *ris oenin! of lenses"

    n!le of illumination of the 43Ds"

    :

    Set of cameras

    Set of 43Ds

    Servomotors

    PCB

    Conveyor

  • 8/11/2019 Automated Visual Inspection

    6/24

  • 8/11/2019 Automated Visual Inspection

    7/24

    Fi!"# 5(isto!rams of resent and absent oulations of comonents

    >hen a comonent cannot be classified none of the 6uality control techni6ues can

    be alied to decide #hether the rocess is in or out of statistical control" This

    creates a !eneral failure of the system"

    SELECTION OF RESPONSE VARIABLE

    The values obtained from the classification al!orithm for resent and absent

    oulations

    A

    0

    50

    100

    150

    200

    250

    -10

    10

    30

    50 70 90 110

    130

    150

    170

    190

    210

    230

    250

    More

    Present

    Absent

    0

    50

    100

    150

    200

    250

    Present

    Absent

    2verlain!

    ofoulations

    2verlain!

    of

    oulations

  • 8/11/2019 Automated Visual Inspection

    8/24

    are calculated usin! three different inuts$ #hich are ener!y$ correlation and

    diffusion all of #hich are obtained from the ima!e rendered by the insection" /or

    the urose of this e+eriment and in comliance #ith the re6uest of the *

    system administrator$ #e focused on the ener!y factor" fter an insection routine

    is erformed$ a reort file$ #hich contains information re!ardin! ener!y$ is

    !enerated" Tables - and sho# a samle of such reorts"

    The comonent that #as used as basis for the e+eriment #as the most common

    comonent in the PCB$ #hich is the

  • 8/11/2019 Automated Visual Inspection

    9/24

  • 8/11/2019 Automated Visual Inspection

    10/24

    The numbers !iven in the ran!e and levels of factor G- refer to sto numbers on

    the ;mm focal distance T lenses that are used by each of the cameras in the

    vision system #here the lo#er the sto number the !rater the oenin! of the iris"

    ccordin! to the * administrator the assumtion of linearity in the factor effects

    is not clear secially in the *ris oenin! factor because #hen !oin! from one sto

    number to the other the oenin! varies accordin! to a constant (:;'/

  • 8/11/2019 Automated Visual Inspection

    11/24

    Since the number of factors of interest for the e+eriment is 9$ #e have adoted 9

    desi!n$ #ith 9 relications (refer to Table :) and : center oints"

    The blockin! techni6ue #as not considered to this e+eriment because none of the

    factors (or the e+eriment itself) #as affected by any nuisance source of variation"

    The PCB boards belon! to the same batch$ no human interaction is related to the

    insection routine that renders the ima!e of the oulation of comonents #ithin

    each PCB"

    The run order #as determined by Desi!n 3+ert soft#are$ the comlete test matri+

    is sho#n in Table 9&

    T-=# && Test Matri+

    -0

    Factor 1 Factor 2 Factor 3 Response

    Std Run Block A:ris Opening B:Angle! "egrees C:Current! Amps O#erlap! pi$els

    27 1 "lo#$ 1 2&2 !7&5 0&5

    22 2 "lo#$ 1 2&8 90 0&!

    3 "lo#$ 1 2&8 5 0&3

    17 "lo#$ 1 2&8 5 0&!

    1 5 "lo#$ 1 1&! 5 0&!

    8 ! "lo#$ 1 1&! 90 0&3

    12 7 "lo#$ 1 2&8 90 0&31 8 "lo#$ 1 1&! 5 0&3

    21 9 "lo#$ 1 1&! 90 0&!

    2! 10 "lo#$ 1 2&2 !7&5 0&5

    19 11 "lo#$ 1 1&! 90 0&!

    9 12 "lo#$ 1 1&! 90 0&3

    ! 13 "lo#$ 1 2&8 5 0&3

    1! 1 "lo#$ 1 2&8 5 0&!

    3 15 "lo#$ 1 1&! 5 0&3

    2 1! "lo#$ 1 1&! 5 0&3

    11 17 "lo#$ 1 2&8 90 0&3

    7 18 "lo#$ 1 1&! 90 0&3

    2 19 "lo#$ 1 2&8 90 0&!

    23 20 "lo#$ 1 2&8 90 0&!

    18 21 "lo#$ 1 2&8 5 0&!

    5 22 "lo#$ 1 2&8 5 0&3

    25 23 "lo#$ 1 2&2 !7&5 0&5

    10 2 "lo#$ 1 2&8 90 0&3

    20 25 "lo#$ 1 1&! 90 0&!

    13 2! "lo#$ 1 1&! 5 0&!

    15 27 "lo#$ 1 1&! 5 0&!

    28 28 "lo#$ 1 2&2 !7&5 0&5

  • 8/11/2019 Automated Visual Inspection

    12/24

    The results table for a number of relicatesE9 is sho#n in Table :&

    T-=# '& Hesults table

    The result table sho#s FF"A as the ercenta!e alha level of all effects at

    standard deviations from the mean$ #hich is sufficiently accurate in terms of this

    e+eriment"

    PERFORMING THE EXPERIMENT

    The run order (random) and combination of factor levels rovided by desi!n

    e+ert #as follo#ed and the results rendered are sho#n in Table ; in the

    endi+" >hile runnin! the e+eriment the most difficult factor to vary #as the

    current of the 43D anels because of the lack of accessibility resented by the

    layout of the machine" >hile chan!in! the *ris oenin! an ad%ustment to the focus

    of the camera #as also re6uired to rovide a clear ima!e in the monitor$ this

    ad%ustment (or the check for a roer ima!e) #as re6uired each time the iris

    oenin! #as chan!ed from one level to the ne+t" The variation of the an!le factor

    --

    %o&er ( alp)a le#el *or e**ect o*

    Term StdErr++ ,F Ri-S.uared 1/2 Std0 "e#0 1 Std0 "e#0 2 Std0 "e#0

    A 0&201 1 0 21&1 ' !3&3 ' 99&! '

    " 0&201 1 0 21&1 ' !3&3 ' 99&! '

    C 0&201 1 0 21&1 ' !3&3 ' 99&! '

    A" 0&201 1 0 21&1 ' !3&3 ' 99&! '

    AC 0&201 1 0 21&1 ' !3&3 ' 99&! '

    "C 0&201 1 0 21&1 ' !3&3 ' 99&! '

    A"C 0&201 1 0 21&1 ' !3&3 ' 99&! '

  • 8/11/2019 Automated Visual Inspection

    13/24

    #as the one that did not resent any ma%or difficulty" The comutation of the

    resonse variable in each run #as erformed by the method described earlier in

    this reort"

    STATISTICAL ANALYSIS OF THE DATA

    The statistical analysis of the resonse variable and the interretation of the results

    derived form this analysis #as erformed by usin! the Desi!n 3+ertJ soft#are"

    Tables ; to < as #ell as !rahs - to are resented in the endi+" Table ;$ Table

    A sho# the Hun matri+ #ith resonses and Desi!n summary resectively"

    The hi!hest order interaction term #as assumed as the error term" The half normal

    lot (7rah-) clearly sho#s that $ C$ C are the si!nificant factors" Takin! the

    si!nificant terms as the main effects and the remainin! terms as error terms refines

    the model" The de!rees of freedom for the model are 9$ for curvature - and for the

    error are :"

    The analysis of variance table (Table ) rovided by Desi!n 3+ert J sho#s the

    values$ the / values$ De!rees of freedom$ Sum of s6uares and Mean s6uare terms"

    The hi!h value of the model /1value is stron! evidence that it is si!nificant"

    ccordin! to the Desi!n 3+ert J reort$ there is only a 0"0-K chance that a

    LModel /1alueL this lar!e could occur due to noise" *t is imortant to mention

    that as sho#n by 7rah ; (lot of residuals s" current) the assumtion of

    homo!eneity of variance seems not to hold" This haens even after data

    transformation has been alied to the data (S6uare root transformation #ith

    -

  • 8/11/2019 Automated Visual Inspection

    14/24

    constant kE as su!!ested by Desi!n 3+ertJ #hen observations seem to follo# a

    Poisson distribution)" o#ever$ due to the hi!h value of the / ratioE

    Model

    MS

    MS

    (;-" 9)$ the effects of non1constant variance #ill be very unlikely to alter the

    results obtained by the e+eriment" More over$ since the e+eriment #as a

    balanced fi+ed effects model the results are only sli!htly affected #hen the

    assumtion of constant variance is violated ,:."

    The analysis also found that there is evidence of si!nificant second1order

    curvature in the resonse as measured by difference bet#een the avera!e of the

    center oints and the avera!e of the factorial oints over the re!ion of e+loration"

    He!ression 36uation obtained from Desi!n 3+ertJ is as sho#n belo#"

    -9

    (inal E)uation in *er+s of Co,e, (a#tors

    .)rt/erla 2&004

    3&!71378879-1&192!518 6 A

    1&330008522 6 C

    -0&3!7902103 6 A 6 C

    (inal E)uation in *er+s of A#tual (a#tors

    .)rt/erla 2&004

    0&00!80250

    -0&179315 6 ris ening

    17&85988!01 6 Current

    -&08780118 ris ening 6 Current

  • 8/11/2019 Automated Visual Inspection

    15/24

    ccordin! to Desi!n 3+ertJ the lack of fit is not si!nificant relative to the ure

    error" This result su!!ests that this model is ade6uately fitted by the re!ression

    e6uation"

    Hesiduals and Model de6uacy&

    The 7rah sho#s the lot of 'ormal robability vs studentized residuals "The

    assumtion of normality is satisfied after the s6uare root transformation">e have

    considered the s6uare root transformation since the observations in this model

    follo#s the oisson distribution as our resonse deals #ith countin! the number of

    comonents on the PCB board and this is also su!!ested by Desi!n 3+ertJ

    B2C2 lot" The 7rahs 9 and : sho#s the lots of residuals vs" redicted

    values and residuals vs" run order" These !rahs sho# that the indeendence

    assumtion on the errors has not been violated and also sho#s that there is no non1

    constant variance" 7rahs A and are the lots of resonse surfaces" These !rahs

    su!!ests that the resonse increases #ith the increase of the increase in Current

    and *ris oenin! "The interaction bet#een these si!nificant factors can be seen

    from the curvature of the resonse surface"

    The H1S6uared value of 0"FFAF sho#s that the roortion of the variability in the

    data is mostly e+lained by the model" lso$ the LPred H1S6uaredL and theLd% H1

    S6uaredL values of 0"FF;A and 0"FFAA resectively are in reasonable a!!reement

    #ith each other" Because the Lde6 PrecisionL ratio (#hich measures the si!nal to

    noise ratio) is --A"000 this model can be used to navi!ate the desi!n sace ,:."The

    -:

  • 8/11/2019 Automated Visual Inspection

    16/24

    PH3SS value of 0"9A su!!ests that the model for the e+eriment can be used in a

    ne# e+eriment to redict the resonse confidently"

    CONCLUSIONS AND RECOMMENDATIONS

    The identification of the li!ht intensity of the 43D anels (current factor)$ the iris

    oenin! of the lenses of the cameras and their interaction as the si!nificant factors

    in controllin! the clarity of the ima!e used by the * system$ #ill lead to an

    increase of accuracy and efficiency in the calibration rocess of the insection

    machine"

    The ma+imum value obtained for the difference in mean ener!y value for resent

    and absent oulations (:"< i+els) #as obtained #hen the *ris factor #as in the

    hi!h level (-"A ste numbers) the Current factor #as in the hi!h level (0"A ms)

    and the an!le factor #as in the lo# level (:;N)" *t is then recommended to run the

    insection machine at these levels" The si!nificance of the non1linearity bet#een

    the hi!h and lo# levels of the factors (esecially iris oenin!) needs to be taken

    into account #hen ad%ustin! the levels in future research to otimize the resonse

    variable defined in this e+eriment and as su!!ested by 7rahs A and "

    -;

  • 8/11/2019 Automated Visual Inspection

    17/24

    REFERENCES

    -" MuOoz$ 4"$ et al$Multivariate On-line Quality Monitoring of SMD Assembly$

    >orkin! Paer$ The 5niversity of Te+as t 3l Paso$-FFF

    " illalobos$ " and erduzco$ "$ Integration of Quality and rocess lanning

    Activities in SMD Assembly$ >orkin! Paer$ The 5niversity of Te+as at 3l

    Paso$-FF

    9" rellano$ M" and illalobos$ "$ !ector "lassification of SMD Images$ >orkin!

    Paer$ rizona State 5niversity$-FFF

    :" Mont!omery$ D"C" (000)$ Design and Analysis of E#periments$;th3dn$ ohn

    >iley = Sons$ 'e# Qork"

    -A

  • 8/11/2019 Automated Visual Inspection

    18/24

    A#.>i)

    -

  • 8/11/2019 Automated Visual Inspection

    19/24

    Table ;& Hun matri+ #ith resonse"

    -ig? V(

    nter#et 3&!71378879 1 0&021008112 3&!27920289 3&71837

    A- ris en -1&192!518 1 0&021008112 -1&23592377 -1&1901 1

    C-Current 1&330008522 1 0&021008112 1&28!59932 1&373!7 1

    AC -0&3!7902103 1 0&021008112 -0&113!0!93 -0&32 1

    Center Poin 0&75552183 1 0&055582239 0&!0535!1 0&870505 1

    (inal E)uation in *er+s of Co,e, (a#tors

    .)rt/erla 2&004

    3&!71378879

    -1&192!518 6 A

    1&330008522 6 C

    -0&3!7902103 6 A 6 C

    (inal E)uation in *er+s of A#tual (a#tors

    .)rt/erla 2&004

    0&00!80250

    -0&179315 6 ris ening

    17&85988!01 6 Current

    -&08780118 ris ening 6 Current

  • 8/11/2019 Automated Visual Inspection

    21/24

    Table alf :or+al lot

    GEffe#tG

    0&00 0&!7 1&33 2&00 2&!!

    0

    20

    0

    !0

    70

    80

    85

    90

    95

    97

    99

    A

    C

    ACalf 'ormalK Probability

    7rah-& alf 'ormal Plot of the factor effects

  • 8/11/2019 Automated Visual Inspection

    22/24

    -

    DE. 7:-EFPE R*Plot.)rt /1erl a3 2 &00 4

    .tu,en t i He, R es i,ua ls

    :or+al'

    2robability

    :or +a l lo t o f re s i,u a ls

    -2& 72 -1&! -0&21 1&05 2&31

    1

    5

    10

    20

    30

    50

    70

    80

    9095

    99

    7rah & 'ormal Probability Plot of Hesiduals

    DE. 7:-EFPE R*Plot.)rt /1erl a3 2 &00 4

    P r e , i # t e ,

    .tu,entiHe,Resi,uals

    Re s i,u a ls s & Pr e , i# te ,

    -3&00

    -1&50

    0&00

    1&50

    3&00

    1&52 2&78 &0 5&30 !&5!

    7rah 9& Hesiduals vs" Predicted

  • 8/11/2019 Automated Visual Inspection

    23/24

    DE . : -EFPER *P lot.)r t/er la2& 004

    R un : u+be r

    .tu,entiHe,Resi,uals

    Re s i, u a ls s & R u n

    -3&00

    -1&50

    0&00

    1&50

    3&00

    1 10 19 28

    D E. :-E FP ER* Plot.)r t/ er l a2&004

    C u r r en t

    .tu,entiHe,Resi,uals

    Re s i,u a ls s & Cu rr en t

    -3&00

    -1&50

    0&00

    1&50

    3&00

    0&30 0&37 0&5 0&52 0&!0

    7rah ;& Hesiduals vs" Current

    7rah :& Hesiduals vs" Hun 'umber

  • 8/11/2019 Automated Visual Inspection

    24/24

    DE . : -EFPER*Pl ot

    .)r t/er la2& 00 4FA r is ening