Choose Language Hide Translation Bar

Laney P’ and U’ Control Charts New in JMP® 17 (2023-EU-PO-1210)

Traditionally, control charts for attribute data (p-charts and u-charts) assume the data is either binomial or Poisson and that the mean is constant over time. Rarely is this assumption true in practice. David Laney developed a technique that solves the problem so that control charts work well whether the mean parameter is stable or not. This talk will both walk through the evolution of the Laney P’ and U’ charts and show examples of how best to apply the charts.

 

 

Hi,  I'm  Annie  Dudley  Zengi,  the  developer  for  control  charts  in  JMP .  In  version  17,  I  added  the  Laney  P'  Prime  and  U'  Prime  Control  Charts.

Just  a  little  reminder  to  everybody,  the  control  charts  are  intended  to  show  the  stability  of  your  process,  how  well  you  can  reliably  make  the  same  part  of  the  same  size  over  and  over  again  over  time.

Let's  think  about,  in  order  to  do  that,  you  need  to  have  a  constant  variance.  An  attribute  chart,  either  Binomial  or  Poisson,  assume  a  constant  variance.

But  what  happens  if  your  variance  is  not  constant?  What  happens  if  the  temperature  changes  or  the  humidity  level  changes,  or  the  voltage  changes,  or  all  kinds  of  little  things  change  in  your  process,  and  you  can't  rely  on  the  variance  being  constant  over  time?

This  is  called  over dispersion,  or  in  rare  instances,  under dispersion.  A  one  parameter  distribution,  the  Binomial  and/ or  the  Poisson  can  model  over dispersion  or  under dispersion.

Now,  Laney  recognizes  this  and  proposed  that  we  normalize  the  data,  account  for  varying  subgroup  sizes,  and  then  compute  a  moving  range.  One  thing  to  keep  in  mind  with  Laney  control  charts  is  the  over dispersion  and  under dispersion  happen  typically  when  you  have  either  large  lot  sizes  or  you  have  unequal  lot  sizes.  It  causes  the  traditional  attribute  charts  to  incorrectly  characterize  your  process.

Let's  take  a  look  at  a  sample  data  that  has,  although  these  lot  sizes  are  not  terribly  large,  it  does  have  some  that  are  unequal  in  size.  We'll  open  up  washers  and  go  to  Control  Chart  Builder.  We're  going  to  model  our  number  of  parts  defective.

I  will  change  my  chart  type  to  attribute  because  this  is  only  available  in  the  attribute  control  charts.  I'll  pull  in  my  lot  size  and  the  lot.  Then  recall  that  lot  size  2  is  the  varying  lot  size.  I'm  going  to  pull  this  down  into  the  lot  size  grouping  here.

Now,  we  notice  we  have  varying  control  limits  and  we  have  a  couple  of  points  out  of  control.   I'm  going  to  turn  on  our  test  beyond  limits.  We  see  we  have  two  points  out  of  control.

Now,  we're  looking  at  a  Poisson  chart.  Also  understand  that  Laney's  P'  Prime  and  U'  Prime  control  charts  only  apply  when  the  statistic  is  a  proportion .  Next,  we  need  to  change  our  statistic  to  a  proportion.  Again,  we  notice  that  those  same  two  points  are  beyond  the  limits.

Now,  if  we  change  our  Sigma  so  that  we  can  choose  a  Laney  P'  Prime,  we  notice  that  now  that  we  change  it  to  proportion.  We  have  four  options  here.  Not  only  just  the  regular  binomial  and  Poisson,  we  also  have  the  Laney  P'  Prime  and  Laney  U'  Prime.

Now,  we're  looking  at  a  U'  chart  right  now  of  number  of  defective, so  I'm  going  to  change  it  to  a  Laney   U' Prime  just  for  consistency .  You  notice  now  those  two  points  that  were  out  of  control  are  now  in  control.  It's  not  indicating  that  our  process  is  unstable  at  this  point.

We'll  take  a  little  look  at  the  formulas.  There's  our  standardized  values.  We  compute  a  moving  range  of  those  standardized  values.  We  sum  them  up  and  then  we  compute  an  average .  That  Sigma  sub  Z  is  what  is  inserted  into  the  formula  for  both  the   P' Prime  limits  and  the   U' Prime  limits.

These  look  just  like  the  formulas  for  the  U'  and  the  P'  charts,  only  with  that  Sigma  sub  Z  in  there .  That  Sigma  sub  Z  approaches  one  as  over dispersion  goes  away .  Keep  in  mind  that   U' Prime  and   P' Prime  charts  will  match  your  U'  and  P'  charts,  respectively,  if  there  is  no  over dispersion,  and  that  this  is  available  when  the  chart  type  is  attribute  chart  and  the  statistic  is  proportion.

Thank  you  very  much  for  your  time.  I  hope  you  found  this  talk  helpful,  and  feel  free  to  reach  out  to  me  with  any  questions.  Thank  you.