How to define Root Cause when some points are out of control chart

hiepyeuh

Registered
Hi there,

Currently, I am working for an automotive company and trying to control all characteristics by SPC.
But I am facing a problem: After finding some points out of Control (till satisfy USL and LSL) but we don't know where should we start to investigate?

Do you guys have any ideas for investigating when one of eight rules were not been satisfied?

:thanks:
 

Golfman25

Trusted Information Resource
Unfortunately that's the stuff they don't teach. But it will depend on your process. Figure out what parameter controls the spec at issue and look for something changing. Tool wear is a typical example. Good luck.
 

Miner

Forum Moderator
Leader
Admin
You want to look for process control factors, input factors or noise factors that changed at the same time (or shortly preceding) the change in the SPC variable.

You also want to look for a factor that changes in the same manner. For example, tool wear occurs slowly over time, so the control chart would show a gradual trend if tool wear were the cause. A sudden out of control point would be caused by a sudden change (e.g., tool change, raw material batch change, etc.)
 

Ninja

Looking for Reality
Trusted Information Resource
As Golfman and Miner have already mentioned...the "Why" involves knowledge of the process and what affects what through the process.

It is common (in my experience) that looking for the "Why" would include those most familiar with the process, be they engineering, R&D, or simply some of those on the line who work the process daily...or all of the above.

The SPC rules will tell you TO look, and why you SHOULD consider looking...
the looking itself is not related to the SPC chart, but is in the "how things work", what affects what, and "what happens when I do this" areas.
 

Bev D

Heretical Statistician
Leader
Super Moderator
What questions you ask are determined by the type of process and the product. A machining process that creates a 'straightforward' dimension will be quite different from a high level assembly that creates a function that is tested at the end of the line...

While I like the 5-Why framework (starting at the Problem and working backward to the causal mechanism) I've found that 'why' is somewhat weak question that too often gets interpreted as a blame game...

four basic questions to ask*:
What's wrong
What's changed
What's different
What's Happening


remember too, that an out of control point could be the result of an incorrect subgrouping scheme.

* John Allen's "3 Good Questions (and one not so good)" is a handy little reference
 

hiepyeuh

Registered
Hi all,

Many thanks with note.
I think we should focus on process variation to define assignable cause and systematic cause.

Btw, thanks Bev D for the awasome book.
 

Bev D

Heretical Statistician
Leader
Super Moderator
You might try Ellis Ott's Proces Quality Control. It has a lot of practical case studies on solving problems. A more expensive and rarer book is Steiner and Mackay's Statistical Engineering.
 
Top Bottom