Process Capability - Changing Limits to Improve Cpk or find Root Causes?

[BaconLikeItsHot]

The engineering group is performing its annual review of manufacturing process capability, and the output from each process analyzed where Cpk has changed since the previous study has been to change the limits on the data in question. In most cases this means widening the specs to bring Cpk up to 1.33, though occasionally it means tightening the spec to reduce Cpk to 1.33. In no cases are the processes examined for causes of the capability change.

I'm new to these things and working with veteran engineers, so I'm hesitant to be the Lone Ranger on this one, but to me it would make more sense if the capability study went more like this:
1. Is the process in statistical control? (we don't ask this question)
a. If yes, go to 2.
b. If no, determine causes, correct them, collect data, and reanalyze
2. How does the Cpk compare to last year's? (we don't ask this, we just care if it's greater/less than 1.33)
a. Decreased? determine causes, correct them, collect data, and reanalyze
b. Increased? determine causes and look if they can be reproduce in other processes
c. No change? Good, and consider if improvements are required


Am I barking up the wrong tree on this? What else should I be considering?

Thanks!

-the rookie-

 

[Bev D]

OMG. why even bother taking measurements - just keep reporting Cpk=1.33?

there are so many things wrong with this I'm not sure where to start. so here's a random start:

Cpk above 1.33 is GOOD. why on earth would one want to revise it down?
Cpk is intended to be a measure of REAL capability with the assumption that the specifications are engineered and based on function. out of spec is bad for the product and in spec is good. changing the spec limits to get a Cpk value is simply wrong and may be unethical depending on what you are doing with the Cpk values and how robust your spec limits are.

then there is the whole issue of variation in the Cpk value. small sample sizes can have large variation in the Cpk simply due to natural variation. at a minimum, the two values should be compared using confidence intervals.

then there is the whole issue of how the samples are spread especially in the face of a non homogenous process

 

[BaconLikeItsHot]

I'm fairly discouraged over this.

Further, we need customer approval for one of the spec changes (widening the spec). In the email to the customer, we explain that "in an effort to improve quality... we need to adjust the limits to more accurately represent what the process is actually doing" while reassuring them that we are "only adjusting the paperwork and not changing the process."

 

[TWA - not the airline]

From what I've read so far I totally agree with Bev that I do not like this at all. And I also agree with you regarding the approach to regularly reassessing process capability (assuming that the inital qualification actually makes sense).
But let's not be hasty. First thing is to find out where the specs come from. Typically they are derived from the initial qualification and set in a way to assure that there is ample warning before the process actually yields faulty product that has to be scrapped. What I find so hard to believe is that veteran engineers simply widen the limits w/o having good reasons for it (or at least the gut feel that it does not matter much). If after such a change the process yield goes down then there will be the urgent need to find a culprit on which to lay the blame. Veteran engineers typically are just to experienced to take such risks...

Some other things to explore. Sometimes in companies you find the craziest procedures (and workarounds):
- If customers who have no clue of SPC always ask for cpk>1.33 then maybe management becomes obsessed with such notions (they typically do not understand SPC either). So a whole system is build around the holy grail...
- If you do not determine process limits during initial qualification but just set the nominal and the process window then maybe you take the first results that will give you cpk>1.33 that you are (at that time) confident you can continuously keep. If then something changes in the process which is still deep inside the process limits then the easiest way is to just change the specs. And maybe, as Bev pointed out, there may actually be no real change, just a new result that perfectly fits into the initial confidence interval...

So what's that mean: Ask more questions, analyze how processes are qualified and what information about the process stability you actually have and do not have. If you encounter such crazy procedures then you can safely assume that the starting point is equally crazy...Then do a stakeholder analysis and work on the procedure for the initial process qualification, so that it makes sense. After that change the reevaluation procedure accordingly and help people to understand and get used to the new processes. Use the new processes to point out improvements that actually have a quantifiable benefit. If you need to have a more structured plan, use the DMAIC cycle.

 

[bobdoering]

I am always suspect of customers that "demand" capability - Ppk or Cpk - of 1.33. First, they assume that they know your process so well that they can be assured capability indices are applicable - although I certainly doubt it. These indices are only applicable to distributions that fall within two-sided tolerances with a target in the center. They are also only applicable to distributions where the variation from the process (not the sampling) is random and independent. If you have any variation that is dependent - such as tool wear - they do not yield accurate decisions. I smell rubber stamping going on.... Capability is a guide, not a goal. Just because the sample you took was capable, there is no guarantee the process will remain capable forever. And, it may appear capable due to sampling error, and may not be capable at all!!!

 

[Steve Prevette]

I'd say the six step process you laid out in the initial posting is not bad. However, the quote "in an effort to improve quality... we need to adjust the limits to more accurately represent what the process is actually doing" is really really scary.

It may be worth having a discussion as best you can with folks using the "voice of the process" vs "voice of the customer" model that Dr. Wheeler uses.

I am not a fan of Cpk largely due to the behaviors that are being exhibited in your process so far - manipulate the specificaitons to get what you want for Cpk. I'd suggest striving to step back one for a while and plot the ACTUAL results data and understand what it is telling you (itself and in relationship to the customer spcs). Then when it is understood how to deal with the data itself, then shift to Cpk calculations.

 

[bpritts]

I think that all of those who have already posted have given useful suggestions.

I am reminded of the analogy of the person who holds a match under a thermostat to make a room warmer!

One additional comment that I haven't seen made yet-- since *real* process improvement takes work, try to learn which characteristics deserve the highest priority for improvement/ greatest risk of problems. Any safety critical? Any where the customer's performance will be severely affected? Start with those, and ignore the paper chase on the remaining 90%.

Secondly, consider how the tolerances were set in the first place. If they were arbitrary (e.g. taken from drawing title block defaults) they don't reflect voice of the customer at all! Don't worry about them unless they fall into the group identified as priorities.

best of luck!

Brad

 

[ChrisPcbaTestEng]

The engineering group is performing its annual review of manufacturing process capability, and the output from each process analyzed where Cpk has changed since the previous study has been to change the limits on the data in question. In most cases this means widening the specs to bring Cpk up to 1.33, though occasionally it means tightening the spec to reduce Cpk to 1.33. In no cases are the processes examined for causes of the capability change.

I'm new to these things and working with veteran engineers, so I'm hesitant to be the Lone Ranger on this one, but to me it would make more sense if the capability study went more like this:
1. Is the process in statistical control? (we don't ask this question)
a. If yes, go to 2.
b. If no, determine causes, correct them, collect data, and reanalyze
2. How does the Cpk compare to last year's? (we don't ask this, we just care if it's greater/less than 1.33)
a. Decreased? determine causes, correct them, collect data, and reanalyze
b. Increased? determine causes and look if they can be reproduce in other processes
c. No change? Good, and consider if improvements are required


Am I barking up the wrong tree on this? What else should I be considering?

Thanks!

-the rookie-

Hi The Rookie, what industry was this post referring to? It sounds very much like PCBA test at ICT to me; why? "because that is the way it must be!"
Regards
Chris

 

[Steve Prevette]

Chris, please keep in mind this is an eight year old discussion. . .