Prioritization of Yield Detractors

[#1]

Besides the question is in spec(value) --- for a single parameter.
If there are several parameters with spec limits there could be sets  with multiple spec violation (fail A, fail A+B , fail A + K ... fail K, pass) ... is there a possibility to define a prioritization of yield detractors in JMP?

device 1 failed due to spec A - subsequent spec's do not have to be tested anymore.
->  In total x% of the devices fail due to (prioritized) spec A.

...

device 2 failed due to spec K - subsequent spec's do not have to be tested anymore. 
-> In total x% of the devices fail due to less prioritized spec K - but no higher-prio spec.

...

device 3 fulfills all specs -> pass.

[#2]

@hogi ,

So is your request to see if a device ( a row in the data table) passes all specs and if not flag it and state which process(es) it failed on?

If so I wrote an add-in that I have not yet submitted to the Marketplace yet and I could use some additional testing by others (if wouldn't mind testing for me).

Cheer,

[#3]

I think hogi is looking to identify the first failure based on some specific order/priority. Below are few reasons that quickly come to my mind for something like this

• you might test only until failure (rest of the results might be missing or something)
• you might still test everything but you only care about the first failure for specific analysis

There are of course also cases where you wish to see all the failed processes.

[#4]

Hi @Chris_Kirchberg , sounds great : )
happy to help testing ...

concerning:
state which process(es) it failed on?, @jthi is right:  the goal is to use process importance to get a unique failure flag for every part.
e.g.

If a a candy is smashed, the color doesn't matter.

If shape and color are OK, how about the taste?

and so on ...

[#5 · Accepted solution]

My add-in is close but does not have pre-determined prioritization. Only a list in order of failure (with column as multiple response). It also allows user to specify k x Sigma (std dev) to see if that row/unit falls outside of that range for every process.

Attached is what I am working on. Maybe this framework can be a start of that? One would have to come up with a scheme to in import prioritization list and type of flag for failure mode. the rest would be relatively easy (new column with failure mode flag).

My add-in purpose was to look at 100s of processes using Spec Limits and then use a k x sigma limit to see if it was better or worse than the set spec limits to see if that unit (or row) conformed. 

So a flag of passed (1) if all spec limits or k x sigma limits (each with their own column) and then a column of the processes (in column order and thus order of checking ) is created for each as well. A column of number of processes out of limits is also created. See screen shot below of output:

 Its a work in progress so your comments are valuable.

[#6]

For the Prioritization, I will use the Process Importance (part of the spec limits).

[#7]

I would consider three ordering options (if we are just keeping with simple order based ones)

• Column property like hogi has
• Order of columns
• and possibly separate list user can load in (you could for example create a table with Ppk, order your table based on that and then use that ordering)

Just based on very quick test, some ideas which come to my mind. Some of these might be totally out of the scope for this addin. These are not in any specific order

• I always like having Keep dialog open option on the launcher
• Make it possible to just create separate report instead of adding new columns to the existing table
  • Maybe linked subset with user selected ID column (and non linked as breaking the link can be annoying)?
• As it is possible to just select specific columns columns, make it obvious which columns were used for the calculation (maybe use Notes column property?)
  • Optionally make it possible to use same launcher to update existing columns with new selections (you could have button to load earlier selections to the add-in launcher window).
• Add button for "add all columns with spec limits" or something similar
• Move Ok and Cancel to the right instead of bottom to be make it similar to JMP Platforms
• Add option to color out-of-spec values
  • With an option to select single color for all failures or two colors (like process screening does it)

[#8]

Thanks @jthi and @hogi for the ideas.

[#9]

Thanks for the AddIn, quite close to what I need.

I will add a Sort(nc) and that's it :)

cool tricks :)
1)

clb << Get Items("Column Reference");

... returns column references instead of strings

2)

Insert Into( list of lists[i], item )

...  deep dive manipulation of the list, wow!

3) histogram with multiple response - I am a big fan!
with 2 mouse clicks you can get a fancy statistics which is neither pssible via Tables/Summary not Tabulate ...

but don't dare to use another column with property multiple response in the same Graph Builder object:
 Re: bug with Graph Builder? multiple response + multiple histograms 

I will play with the visualization, combining the multiple response failure information and the "highest priority" failure information into one picture. Perfect timing - with v19EA4 it's not possible again to use a multiple response column for the color :)

Maybe:
replace sigma with a robust version based on MAD - then extreme outliers don't give other outliers the chance to hide under the "stretched" carpet : )

[#10]

For the plot of the fail devices, instead of the Bar chart:

one might use a histograms chart to get further insights about correlations of multi-fails: