Showing results for 
Show  only  | Search instead for 
Did you mean: 
Choose Language Hide Translation Bar
News Flash: Gauges Aren’t Perfect! OK, You Know That. But How Much Is It Costing Your Business? - (2023-US-30MP-1480)

Jerry Fish, Systems Engineer, JMP
Jason Wiggins, Sr Systems Engineer, JMP
Brady Brady, Principal Systems Engineer, JMP


Most engineers and scientists know the value of a good measurement systems analysis (MSA). After all, our data decisions depend on having gauges that give good readings, right? But no gauge is perfect. No matter how hard we try, there are always errors associated with a gauge. That’s why MSAs are so important. They help us quantify the errors that the gauges make.


But once you have finished your MSA, what do you do next? Depending on the results, you might conclude that the gauge is “poor” and throw it away. Or you might conclude that the gauge is “good” and use it without really understanding the errors that it will still make.


What if this imperfect gauge is used in production? How often will it make Type 1 vs Type 2 errors? What do those errors cost your company, in terms of money and in terms of diminished reputation (if bad parts are shipped)? Would it be wise to invest time and/or money in improving your gauge to mitigate these errors?


These questions (and more) are answered with a new JMP add-in. To demonstrate the add-in, we present a case study that shows how to determine the percentage of good parts that are potentially being rejected by your gauge, and the number of bad parts that are being passed on to your customers. We also discuss strategies to mitigate these errors.



Welcome,  everybody. My  name  is  Jason  Wiggins.


I'm  a  Systems  Engineer  with  JMP.

For  this  presentation,  I'll  be  playing the  role  of  a  Quality  Consultant.

My  name  is  Jerry  Fish. I'm  a  Systems  Engineer  with  JMP.

I'll  be  playing  the  role  of  a Quality  Manager  at  a  manufacturing  plant.

Just  a  couple  of  references  for  this  presentation  upfront.

We're,  Jerry  and  I,

building  on  a  paper  that  we presented  at  Discovery  Europe  2023.

That's  the  paper  number.

You  can  go  to   community.j

and  search  for  that  paper.

Jerry  will  also  be  presenting  an  on-demand video  for  Discovery  on  Demand.

That's  the  paper  number  to  search  as  well.

Throughout  the  presentation, we  will  be  referencing

Dr. Donald  Wheeler's  book,  EMP  III,

Evaluating  the  Measurement  Process and  using  imperfect  data.

All  right,  Jerry,  ready  to  kick  this  off?

Absolutely,  go  for  it.

All  right,  I'm  calling  you  up. Hi,  Jerry.

Thanks  for  spending  a  few  minutes  with  me.

As  a  quality  consultant,

I  help  quality  stakeholders  like  yourself understand  and  improve  processes.

Hi,  Jason. Nice  to  meet  you.

I  need  to  let  you  know,  I  don't  have much  time  to  spend  with  you  right  now.

We've  got  an  emergency  situation  on  our

production  line  that  I  don't really  need  to  go  address.

I  know  we've  got  30  minutes  scheduled

and  I'm  interested  in  discussing  gages, but  can  we  try  to  make  this  quick?

I  understand  completely.

I'll  try  to  make  the  most of  our  time  today.

To  kick  things  off,

can  you  tell  me  a  little  bit  about your  company  and  your  quality  program?


You  probably  know  that  Acme has  built  a  reputation  with  our  customers

for  manufacturing highest  quality  products.

We're  always  concerned  with  quality.

We  have  various  gages  that  we  use to  ensure  our  quality  stays  high,

and  we've  been  doing  this  for  many  years, so  we  think  we're  pretty  good  at  it.

I'm  familiar  with  Acme's high  quality  reputation.

My  consulting  team  and  I  have  been  working

with  manufacturing  companies  like  yourself that  have  a  high  focus  on  quality

to  advance  the  use  and  effectiveness of  great  gage  studies.

One  of  the  things  we  seek  to  understand  is the  monetary  cost

associated  with  the  gages  used  to  measure process  quality  characteristics.

Have  you  quantified  how  much  any  of  your gages  are  costing  your  business?

I'm  not  sure  I  know  what  you  mean.

Well,  gages  aren't  perfect.

They  make  mistakes.

Sometimes  they'll  throw  away  good  parts and  sometimes  they'll  pass  bad  parts.

Unless  you  have  a  perfect  gage  and  no one  has  these,  mistakes  are  inevitable.

Yeah,  I  get  that,  I  suppose  so.

We've  done  traditional  gage studies  that  say  our  gages  are  good.

Well,  some  of  them  are  actually categorized as adequate  for the

AIAG  Automotive  Industry  Action  Group, which  we  use  for  our  testing.

Doesn't  that  mean  they're  okay  to  use?

There's  a  lot  more  to  the  story  than  just

using  the  good,  adequate, and  poor  AIAG  assessment.

In  fact,  we  could  have  a  long  discussion

about  the  differences  between  AIAG results  and  a  newer  method  called  EMP  3.

I  know  you  said  you  have  30  minutes.

It  turns  out  that  AIAG, we  were  trying  to  sum  it  up,

it's  just  much  more  critical of  gages  than  it  needs  to  be.

The  EMP  method,  which  is pioneered  by  a  guy  named

Dr. Wheeler,  is  a  more  realistic way  to  evaluate  gages.

I  appreciate  you  skipping  over  that  part.

We  recognize  no  gage  is  perfect.

We  also  recognize  that  using  a  poor  gage

risks  accepting  bad  parts, which  is  bad  for  my  business.

The  worse  the  gage,  the  higher  the  risk.

We've  got  a  way  that we handle  this  problem.

Oh,  what's  that?

Well,  we  instruct  our  manufacturing  group

to  use  inspection  limits  that  are inside  of  the  specifications.

If  we  set  them  far  enough  inside  these  spec  limits,

we  reduce  and  essentially we  can  eliminate  shipping  bad  parts.

Doesn't  that  fix  our  problem?

Using  inspection  limits   that  are  inside  the  spec  limits

is  a  good  way  to  reduce the  risk  of  passing  bad  parts

when the  gage  measures  apart  near the  specification  limits.

This  is  a  common  strategy, something  that  companies  call

a guard banding  or  specifying manufacturing  instructions.

How  do  you  choose  how  much to  narrow  your  inspection  limits?

Well,  frankly,  I'm  not  sure  we put  that  much  rigor  into  it.

We  just  move  our  inspection  limits in  board  of  the  specs  by some amount

that  our  subject  matter  experts have found  work  well  over  the  years.

Well,  you're  in  good  company.

Many  others  like  you employ  the  same  approach.

There  is  a  downside,  though.

Oh,  what's  that?

By  arbitrarily  choosing  how  much  to  narrow

inspection  limits,  you  may  be unnecessarily  throwing  out  good  parts.

That's  just  assuming  they're  too  narrow.

Or  you  could  still  have  bad  parts

that  escape  inspection if  the  limits  aren't  narrow  enough.

Have  you  considered  taking a  more  statistical  approach?

We  have  not.

I've  always  thought  we  should  have a  better  way  to  justify  our

manufacturing  instructions, but  we  don't  know  how  to  do  that.

Well,  Dr. Wheeler proposes  a  method  of  setting

manufacturing  instructions based  on  the  statistics  we  get  from  doing

a  measurement  systems analysis  in  his  book,  EMP  3.

I  honestly  haven't  heard of  Wheeler's  EMP  method  before.

Is  it  new?

It's  been  around  for  a couple  of  decades  now.

More  importantly  to  my  knowledge, Dr.

Wheeler  was  the  first  person  to  publish

a  solution  to  the  problem  of  objectively setting  these  inspection  limits.

His  approach  uses  outputs  from  MSA, like  we  were  mentioning.

Some  of  these  are  probable  error,

along  with  expectations,  your  expectations were  conformance  to  specifications.

Do  any  of  the  modern  statistical  analysis

package  that  are  on  the  market today  support  this  EMP  method?

I'm  glad  you  asked.

Jmp,  which  is  my  preferred  data  analytics

tool,  JMP  is  a  general  purpose, easy  to  use  data  analytics  package

that  has  many  quality and  process  control  features.

Jmp  makes  quick  work  of  the  analytics  part of  process  improvements,

so  more  time  can  be  dedicated to  improving  the  process  itself.

You  know  that  that's  not  a  trivial amount  of  work,  typically.

Jmp's  EMP  personality  and  the  measurement

systems  analysis  platform is  based  on  Wheeler's  book.

All  that  stuff  is  in  there.

The  calculations  for  manufacturing instructions  are  not  available  in  JMP  yet.

We  built  an  add-in  that  does  them  for  you

and  provides  some  tools for  cost  trade-off  analysis.

Okay,  good.

I  was  afraid  I  might  have  to  come up  with  my  own  equations.

I'll  order  a  copy of  that  book  for  my  team.

If  you've  already  got  an  add-in

that  does  that  for  us, I'm  definitely  interested.

What  does  it  cost  and  how  does  it  work?

Well,  good  news,  it's  free.

You  can  find  it  in  the  file exchange  under online  or  JMP  User  Community.

Let  me  show  you  how  it  works. Great.

All  right. Here  on  the  left  side,

we  have  to  enter  a  little  bit of  information  about  our  product.

First  is  the  specification  limits.

You  know,  this  is  what  our  customer  wants.

This  is  voice  of  the  customer.

An  upper  and  lower  specification  limit.

We  need  to  enter  what  we  know about  our  measurement  error.

We're  going  to  use  the  standard deviation  of  our  measurement  error.

This  is  something  that  we  can  get from  a  measurement  systems  analysis.

We  also  need  to  supply  some  information

about  what  we  expect  the  true  part distribution  to  look  like.

The  parameters  mean  and  standard deviation,  give  us  a  description  of  what

the  true  part  distribution is  for  our  process.

Wait  a  minute.

I  don't  know  my  true  part  distribution.

That's  the  problem  here.

All  I've  got  is  my  measured part  distribution.

Okay, let's  for  the  moment,

assume  that  the  true  part  distribution  is normally  distributed,

and  the  gage  are  also normally  distributed.

We  can  use  a  simple  relationship to  get  the  true  part  variance.

If  I  did  a  measurement  systems  analysis,

I  have  the  variance associated  with  my  gage.

If  I  did  a  bunch  of  measurements

of  my  process,  I  have  a  measured part  distribution  of  my  process.

I  just  have  to  subtract  those,  and  then I  can  get  the  true  part  variance.

From  that,  I  can  get  the  standard deviation  for  my  true  part  distribution.

Mean's  simple  also.

We're  just  going  to  assume  that  that  mean

is  the  measured  part  distribution  less any  bias  that  we  have  with  the  gage.

Okay. I  think  I'm  with  you  so  far.

We  are  still  moving  towards  how  much my  gage  is  costing  my  business,  right?

My  line  still  has  that  problem that  I  need  to  address.

Got  you. We  are.

If  you'll  indulge  me just  a  little  bit  longer.

We  have  more  work  to  do.

First,  let's  look  at  the  statistics summary.

Probably  layer,  interclass  correlation and

precision  to  tolerance are  ways  of  describing  gage  variation

that  we  get  by  conducting  Wheeler's measurement  systems  analysis.

These  values  are  computed  using

specification  limits and  gage  sigma  inputs.

I'll  let  you  read  about  these calculations  when  you  get  Wheeler's  book.


The  capability  indices are  used  to  describe  how  much  of  the  spec

limit  range  is  consumed by  process  variation.

This  is  the  variation of  the  true-part  values.

Capability  is  calculated  using  spec limits  and  our  true-part  mean  in  sigma.

Yes,  at  Acne,  we  regularly  use  CP  and  CPK.


Collectively,  you're  probably  getting an  idea  that  these  values  provide  us

a  nice  snapshot  of  gage and  process  variation.

It's  going  to  help  us.

For  this  example, an  ICC  of  0.86  is  telling  us  that  14  %

of  our  measurement  variation is  a  result  of  gage  error.

But  our  process  is  not  performing  so well  at  a  capability  of  only  about  67%.

The  last  thing  that  we  need  is  expected conformance  at  manufacturing  instructions.

That's  a  mouthful.

What  do  you  mean  by  that?

It  is. Let  me  see  if  I  can  unravel  that.

This  is  the  probability that  a  measured  part  will  truly  conform  to

specifications if  the  gage  measures  exactly

at  specification  limits that  we  discussed  earlier.

Exactly  at  55  or  75.

For  example,  let's  say  that  we  want  99% conformance.

To  achieve  this,

we  need  to  set  our manufacturing  instructions.

Let  me  make  this  a  little bit  bigger  for  us  here.

We  need  to  make  our  manufacturing

instructions 57.3  and  72.7.

That's  actually  narrower  than  the  55  and 75 that  are  our  spec  limits.

Then  if  we  do  that,

we  can  expect  a  99%  chance that  will  reject  or  accurately  accept

a  part  that's  near  those specification  limits.

Okay,  let  me  echo  that  back  to  you to  see  if  I've  got  that  right.

If  I  set  my  inspection  limits  for  this example  to  57.3  and  72.7,

and  then  if  I  measure  a  part  with  my  gage that  measures  exactly  57.3,

there's  a  99%  chance  that  it  meets my  product  specifications  of  55  and  75.

That's  right. You  said  it  even  better  than  I  could.

It's  really,  in  my  mind,  it's  useful  to think  about  measurement  risk  in  this  way.

As  measurements  that  we  make  near  spec limits  are  the  riskiest  that  we  encounter.

It  also  exposes  trade-offs  between  gage effectiveness  and  process  capability.

Think  about  this,  Jerry.

If  the  process  is  highly  capable,

you  can  get  away  with  the  marginal  gage even  when  measuring  near  the  spec  limits.

However,  if  you  have  a  poor  CPK, it  may  be  smart  to  adjust

the  manufacturing  instructions  to  mitigate risk,  even  if  the  gage  is  effective.

That  sounds  like  what  my  subject  matter

experts  have  been  trying  to  do, but  you're  saying  that  we  can  put  a  lot

more  statistical  rigor  behind  setting those  manufacturing  instructions.

Your  example  sets  manufacturing instructions  to  give  99%  conformance.

I  assume  if  I  want  99.9%

conformance,  those  manufacturing instructions  will  narrow  further?

They  will.

In  fact,  let's  just  take a  look  at  how  much.

Wow,  that's  quite  a  bit.


58,  8,  and  71.1.  Okay, I  think  I'm  getting  it.

Narrowing  those  manufacturing  instructions is  great  for  guaranteeing  that  I'm

shipping  good  parts, but  I  don't  necessarily  like  that.

Why  not? In  fact,  I'll  switch  it  back.

Why  don't  you  like  that,  Jerry?

Well,  the  narrower  we  set  those manufacturing  instructions,

the  more  good  product  I'm going  to  throw  away,  aren't  I?

Yes,  that's  right.

I  think  you're  beginning  to  see

the  trade-offs  associated  with  setting manufacturing  instructions.

It's  pretty  cool. Yeah.

Can  you  tell  me  how  much  it's  going to  cost  me  when

I  throw  that good product

away  given  the  selected manufacturing  instructions?


That's  given  in  the  profit  loss explorer  outline  of  the  add-in.

Before  I  can  answer  that,  though, we  have  to  enter  some  values.

Hypothetically, when  we're  talking  about  one  of  your

parts,  how  much  revenue can  you  expect  per  part?

Okay,  well,  that  number  looks  good.

For  the  sake  of  this  demo  or  this  example, let's  just  say  $100  per  part.


How  many  parts  do  you  expect  to  make?

100,000  sounds  good.

That's  typical,  at  least for  maybe  a  year's  production.

Nice. Good  part  run.

How  much  does  it  cost  you  to  make  a  part?

Pretty  easy  to  figure  that  out.

Out  of  the  $100  that  we  sell  a  part  for,

it  costs  us  about  $70  to  make  it, so  we  make  a  $30  profit.

Last  but  not  least,  what  is the  cost  of  shipping  a  bad  part?

What's  the  penalty  if a  part  escapes  your  inspection  process?

I  see  what  you're  saying.

That's  a  little  tougher.

There's,  of  course,

the  potential  cost  of  return and  the  associated  cost  of  repairs.

Those  are  fairly  easy  to  calculate.

There's  also  damage to  Acme's  reputation.

Our  customers  demand  quality,

and  if  we  start  putting  bad  product  out the  door,

it  can  quickly  get  out  of  hand  in  a  hurry and  result  in  lost  future  sales.

That's  a  lot  more  difficult  to  calculate.

Really,  it  would  take  us  a  while to  scratch  our  heads  and  figure  that  out.

For  the  sake  of  argument, let's  say  that  amounts  to  about  $200  per

bad  part  that  we  let  out of  the  plant  and  sell.

All  right,  great.

Now,  I  see  one  more  entry  box.

Another  word,  salad.

Number  of  PEs,  manufacturing instructions  lie  from  WSLs.

What  in  the  world  does  that  mean?

All  right,  you  definitely  have  exposed

another  worthy  of  explanation set  of  terminology,  Jerry.

Let  me  take  a  stab  at  this.

What  we're  referring  to  are  integer multiples  of  probable  error.

Remember,  we  talked  about  that  as  being one  of  the  outputs  from  an  EMP  measurement

systems  analysis, talking  about  the  integer  multiples

of  probable  error  away from  watershed  limits.

This  is  how  we  will  describe  the  location of  manufacturing  instructions  relative

to  watershed  limits  for  the simulation  that  we're  running.

Turns  out  Wheeler  uses the  same  approach  in  EMP 3.

When  you  get  the  book, take  a  look  at  that.

It  should  be  pretty  consistent.

He  uses  watershed  limits  as  opposed

to  spec  limits  to  account for  granularity  of  measurements.

If  we  performed  a  measurement  systems analysis  using  the  EMP

method  on  this  gage,  we'd  be given  a  measurement  increment.

Watershed  limits  are  essentially  half  a measurement  increment  outside  the  specs.

From  a  practical  point  of  view, bottom  line,  manufacturing  instructions

should  be  set  from  watershed  limits as  opposed  to  specification  limits.

That  just  helps  us  get  around that  granularity  problem.

Now, it's  also  worth  noting  that  manufacturing

instructions  can  be  set  inside  or  outside the  watershed  limits.

Now,  wait  a  minute.

That  sounds  crazy.

Are  you  saying  that  there  are  times  when  I

might  want  to  set  my  manufacturing instructions  outside  of  my  product  specs?

Why  would  I  ever  want  to  do  that?

Yeah,  it  seems  counter  to  it, and  I'd  argue  that  it's  relatively  rare.

There  are  some  conditions  where

for  strictly  economic  reasons, you  might  be  better  off  choosing  an  even

wider  set  of  manufacturing  instructions than  your  product  specs.

We  can  talk  through an  example  in  just  a  second.

Remember  when  we  were

exploring  what  the  statistics  meant, and  we  talked  about  this  idea  that  if  I

have  a  process  that  is  very  capable, but  we  have  an  imperfect  gage.

It  may  be  smart  for  us  economically to  actually  push  our  spec  limits  wide

and  adjust  the  inspection process  accordingly.

There  are  circumstances  where  that  might

happen,  but  yeah,  I'd  argue that  they're  pretty  rare.

I  see. I  don't  know.

For  now,  for  the  sake  of  the  simulation,

let's  set  the  number  of  probable errors  to  minus  one.

What  this  means  is  that  we're  narrowing our  manufacturing  instructions.

Let  me  make  this  a  little  bit  wider  here.

We're  talking  about  narrowing

the  manufacturing  instructions inside  the  specification  limits.

When  we  do  that,  we  can

focus  your  attention  down  to  the  expected net  profit  at  manufacturing  instructions.

Here  we  can  see  that  the  selected

expected  net  is  the  net  profit, and  we  can  stand  to  make  about  $1.9

million  for  100,000  parts for  the  parameters  that  we've  entered.

Okay,  well,  at  least  that's a  profit.

I  see  a  maximum  expected  net that  shows  an  even  higher  profit.

Is  that  something  I  should consider  or  can  I  achieve  it?

Yes,  definitely.

The  maximum  profit  that  you  can  achieve for  this  measurement  process  and

this  economic  set  of  circumstances is  about  $2.1  million.

Now,  what  this  is  telling  us  is  that  we can  achieve  it  by  making  our  manufacturing

instructions  equal to  the  watershed  limits.

If  I  change  this  to  zero,

we can  see  that  we've  hit  that  maximum expected  net.

We've  just  done  that  by

moving  the  manufacturing  instructions to  the  watershed  limits.

This  is  optimal  for  the  conditions that  we've  entered.

That's  really  interesting.

With  any  combination  like  this, potentially,  I  could  find  an  optimal

where  I  need  to  set  these manufacturing  instructions.

to  get  me  the  maximum  profit.

Yeah,  you're  nailing  it.

In  fact,  for  the  add-in, there's  an  asterisk  on  the  graph

that  identifies  what  the  optimal profit  you  can  achieve  is.


When  you're  doing  this  optimum, where  are  you  getting  these  numbers?

I  don't  understand  where they  might  come  from.

Sure. They  result  from  trade  offs  between

revenue  per  part,  cost  per  part, and  damage  to  your  reputation.

Those  are  the  things  that  we  enter.

When  we  sell  a  bad  part, that's  the  reputation  component.

Given  the  process  and  gage  parameters.

That's  the  true  part  and  measurement error  components  that  we  entered  earlier.

For  now,  like  I  said,  let's  keep  the

example  to  where  we're  narrowing  our

manufacturing  inspection  limits. It's  so  funny.

We  get  all  these  different  limits.

We're  talking  about  manufacturing

instructions,  which  are inside  the  watershed  limits.

Now  that  we've  got  this  and  we're  agreed

that  hey,  this  is  the  way  we want  to  run  this  simulation.

Let's  take  a  look  at  the  profit simulation  panel.

The  matrix  down  below  is  where  we're  going

to  start,  breaks  the  profit  loss into  nine  different  categories.

Okay,  there's  a  lot  of  information  there.

I'm  going  to  need  to  study this  for  a  minute.

Let's  start  in  the  center, the  center  of  that  three  by  three  grid.

S tarting  there,  of  the  100,000  parts that  we'll  make,

about  91,000  are  truly  good  parts  that  our gage  reads  as  good  and  that  will  ship.

That  produces  2.7  million  in  profits, and  I  assume  that  works  out  to  91,000

parts  times  we  said  the  revenue  per  part was  $100  minus  $70  to  make  the  part.

Is  that  right?

That's  right.


Moving  straight  to  the  left,

you're  predicting  we'll  have  370  parts that  we  will  ship  because  the  gage  says

they're  good  parts,  but  these  parts are  truly  below  the  lower  spec.

That's  a  relatively  low  number, but  it's  not  really  good.

You  say  it's  costing  me  $63,000.

Am  I  reading  that  right?

Yes,  indeed,  you  are.

Okay,  and  I'm  sure  that  has  to  do

with  that  $200  penalty  I'm  paying for  letting  bad  parts  get  out.

In  this  case,  it  looks  like  I  may  lose

about  as  much  by  shipping parts  that  are  bad  on  the  high  side.

That's  right.

You  made  2.7  million  selling  good  parts, but  now  the  imperfect  gage  has  cost  you

a  total  of  140,000  because  of  the  letting bad  parts  out  into  the  field.


Okay,  well,  that's  not  very  good.

Now  you've  got  several  more  cells

in  that  three  by  three grid  that  show  losses.

What  are  those  about?

The  upper  left  cell  shows  parts  that  are truly  bad  because  they  are  below  spec.

And  your  gage  is  catching  them.

That's  good.

It's  catching  them, it's  throwing  them  out.

The  gage  is  doing  what  you  expected  to  do.

Downside  here  is  that  there  are  nearly 1,900  parts  that  you're  throwing  out,

and  each  of  those  are costing  you  $70  each.

Same  thing  for  the  lower  rights, similar  math  there.

That's  another,  what,  $260, $270,000  coming  out  of  my  pocket?

I  can't  blame  that  on  the  gage.

Those  can  only  be  fixed  if  I improve  my  production  process.

What  about  those  other  cells?

Let's  look  at  the  upper  center  cell.

It's  showing  that  we  made  a  little  over

2,200  good  parts that  were  thrown  out  because  the  gage

measured  below the  manufacturing  instructions.

How  do  you  feel  about  that?

Oh,  not  so  great.

We  have  a  separate  team  of  inspectors

that  reinspect  rejected  parts  in  hopes of  reclaiming  some  of  our  losses.

This  is  making  me  think  that  even

with  that  reinspection, we're  still  throwing  out  good  parts.

I  assume  the  bottom  center  cell  is

the  same,  good  parts  that  my  poor gage  says  are  higher  than  spec.

Yeah,  that's  right.

Yet  there's  another  150-somewhat thousand  dollars  you're  losing.

All  of  those  profits  and  losses are  summarized  in  the  bar  chart.

We  get  a  graphic  picture  of  those  as  well.

I  see.


but  now  that  assumes  that  we're  using manufacturing  instructions  that  are  one

probable  error  inside the  watershed  limits.

What  if  we  change  the  manufacturing

instructions  to  the  optimum you  were  talking  about?

How  does  that  change  the  profit  picture?

Let's  see  what  happens.

Just  watch  the  bars  change  off

to  the  right,  and  we'll  talk about  those  a  little  bit.

We'll  bump  that  up  to  the  optimum, and  what  did  you  see  happen?

Yeah,  so  it  looked  like  the  shorter  bars

got  shorter  yet, and  the  long  bar  got  a  little  longer.

That's  good,  and  my  net  profit  went  up. Good.

We're  making  more  money.

It's  probably  worth  thinking  about  that

this middle  bar  because  of  that  reputation cost  that  you  talked  about.

Yeah,  sure,  we  may  be  making  more  as

a  business,  but  you  ought  to  at  least  be talking  with  stakeholders  in  the  company

and  asking  how  we're saving  a  little  bit  of  money  here  to  be

more  profitable,  but  is that  the  right  thing  to  do?

Yeah,  I  get  it.

That's  interesting.

Yeah,  I  might  choose  to  use  the  narrower manufacturing  instructions  because

that  cost  to  our  reputation, you  really  got  to  think  about  that  $200  we

put  in  earlier, that  could  be  a  much  higher  number.

We'd  have  to  sharpen  our  pencils  on  that.

Then  again,  I  guess  I  could  experiment with  that,  with  this  add-in,  right?

I  could  try  different  values.

Yeah,  absolutely.

I  hope  you  will.

I  feel  like  that  would  be  valuable.


I  think  I'm  starting  to  see  now  how

changes  in  process  capability  and  gage effectiveness  and  how  I  set  the  inspection

limits  may  have  a  big impact  on  my  profitability.

That's  what  we  hoped  for  when we  developed  this  add-in.

Considering  profitability  when

answering  questions  about  gage  and  process capabilities

could  be  really  helpful  when  you're  trying to  justify  improvement  projects,

or  it  could  also  give  you  a  lot of  confidence  that  the  process  and  gage

are  delivering  the  results that  you  desire.

I  must  say,  Jason,  I  am  impressed.

However,  I  feel  like  I  need to  muddy  the  waters  a  little  bit.

This  is  all  great  for  normal

distributions,  simple gage  errors,  et  cetera.

Those  calculations, as  you've  shown  before,  are  easy.

What  if  I  have  linearity or  bias  problems  with  my  gage?

Or  if  I  have  a  skewed  part  distribution,

getting  a  true  part  distribution  out of  the  measured  part  distribution  becomes

a  lot  more  difficult  than  just  using  that simple  formula  you  showed  earlier,  right?

Can  you  even  do  that?

Today,  no.

Really,  we're  limited  to  normal distributions,  but  we're  almost  there.

It's  in  the  plan.

In  fact,  it's  the  next  step  in  our development  plans  for  the  add-in.

I do have a colleague  who's  presenting

at  JMP's  online  Discovery  Conference who  has  done  an  on-demand  presentation

for  estimating  the  true  part  distribution based  on  measured  part  distributions

that  may  be  non-normal  based on  gage  characteristics.

It  handles  even  arbitrary  distributions,

has  some  common  ones and  arbitrary  distributions.

We  will  be  incorporating  these

capabilities  into  the  add-in, like  I  mentioned.

It's  going  to  be  sometime  soon.

You  should  definitely  right now  go  check  out  the  talk.

Yeah,  it  sounds  like  an  interesting talk  by  an  interesting  presenter.

This  all  sounds  interesting. I  will,  Jason.

Now,  we've  covered  a  lot of  material  today.

Would  you  mind  quickly  taking  me  back

through  the  add-in  again to  make  sure  I've  got  it?

Absolutely. Where  we  started  our  conversation  was

supplying  a  little  bit  of  information about  our  measurement  process.

We  entered  specification  limits.

Again,  those  are  the  voice of  the  customer  that  we  talked  about.

We  talked  about  getting  an  idea  of  what

our  measurement  error  is from  a  measurement  systems  analysis,

and  then  using  the  standard  deviation from  that  study  for  a  gage  error.

We  also  talked  about  supplying  some

information  about  the  true part  distribution.

The  parameters  mean  and  sigma,

we  talked  about  ways  that  we  can  infer that  from  the  measured  part  distribution.

We talked about  the  statistics  in  the  add-in

being  a  snapshot  into  our  gage  capability as  well  as  our  process  capability.

I  think  that's  important  as  we  begin

to  have  these  conversations  about  how much  the  gage  is  going  to  cost  us.

We  walked  through  ideas  around  moving  our specification  limits  in  these

manufacturing  limits, narrowing  them  or  even  expanding  them

to  reduce  risk  when  we're  measuring at  the  specification  limits.

There  are  other  bells  and  whistles in  the  add-in  that  we  didn't  have  time

to  show,  but  these  are  just  ways  of  really illustrating

these  risks  by  now  overlaying  true-part distributions,  assuming  we're  measuring

the  true-part  near  the  manufacturing instructions.

Profit  Loss  Explorer  was  our  way

to  simulate  the  costs  associated with  our  measurement  errors.

It  gives  us  the  opportunity to  explore  what  ifs.

If  we  are  making  a  certain  product  and

product  has  certain  costs  associated with  it,  how  much  profit  can  we  expect  if

we're  running  manufacturing  instructions at  different  multiples  of  probable  error.

We  wrapped  up  our  conversation

with  an  in-depth  conversation  about  those profits  and  losses

and  how  the  goal  of  our  business and  the  reputation  of  our  business  should

be  considered  when  we're  making those  trade-offs  as  well.


Thank  you.

I  think  that  summarized  it  really  well.

Now,  besides  using  estimated  true-part

distributions,  is  there  anything  else that  will  be  developed  for  the  add-in?

Well,  we  also  have  already  developed this  misclassification  Explorer.

It's  just  another  way  of  looking  at  gage errors  and  our  process  variation.

We  didn't  have  enough  time  to  go into  this  component  of  the  add-in.

We  plan  on  explaining  this  a  little  bit

more  in  the  near  future, very  likely  as  we  start  incorporating  some

of  those  non-normal distributions  that  I  mentioned.


That's  fantastic.

What  about  other  training  opportunities?

What  we've  shown  here  today  is  what  our

add-in  currently  does,  short  of  the misclassification  explore  that  we  skip.

We'd  like  to  get more  into  the  nuts  and  bolts

of  the  equations  behind the  add-in  and  the  approaches.

We  plan  a  series  of  online  talks  coming in  the  near  future  where  we  can  explore

some  of  those  differences  between  the  AIAG classification  methods  and  EMP  and  how  EMP

can  give  us  more  realistic  and  useful information  about  our  gage.

We'll  also  spend  some  more  time talking  about  gage  performance  curves.

We  talked  about  this

classification  graphs, but  we'll  introduce  you  to  some

traditional  gage  performance  curves as  well  as  conformance  curves.

Actually,  it's  conformance  we  talked about  classification  we'll  get  to.

Look  for  those in  the  near  future.


Who  else  should  we  thank for  this  effort  so  far?

Well,  Brady,  Brady.

We  should  have  a  picture of  Brady right  here.

Yeah,  that  would  have  been  good.

Brady,  Brady  is  the  brains behind  the  add-in.

Unfortunately,  he  couldn't  join  us,  but

definitely  is  heavily  involved,  and  we have  the  add-in  and  the  code  to  thank.


Thanks  to  Brady,  and  thanks  to  all  of  you for  attending  and  those  who  will  be

watching  this  video and  using  the  add-in  later.

For  the  live  demo,  we'd  be happy  to  answer  questions.

Otherwise,  if  you  put  your  questions in  the  chat  area  below  the  presentation,

we'd  be  glad  to  get  back with  you  as  soon  as  we  can.

Thank  you. Thank  you  very  much  for  attending.