Implementing SPC - Control Chart for Individual Measurements

T

tweers

We are working on implementing SPC at our facility for the first time, and I was hoping someone was able to offer some guidence.

Our inspection process currently works by measuring 1 part per shift, 3 shifts, so 3 parts per day. A part may only run for a couple days, or may run for months.

My current understanding leads me to think I am looking at a Control Chart for Individual Measurements using x-bar and MR-bar. Because my subgroup size is really only ever going to be 1, is this the best chart I should use? I'm a little concerned this could lead to type-II errors.

Do the short runs steer me away from measuring part specific dimensions, or should I just look at process variance?

I appreciate anyone's assitance, and I doubt these will be my only questions with a project of this magnitude. I just want to make sure things are done right the first time.

Thanks again.
 

Steve Prevette

Deming Disciple
Leader
Super Moderator
X moving Range will probably work fine. I find that SPC works fine for short runs, and don't subscribe to the specialized "short run SPC" techniques. The key is to learn from what data you have on hand.

With a small amount of data, you might end up setting a bad initial baseline, but if you get a signal early on, try adding in all of the data to date into a new set of average and control limits and see if that eliminates the signal. By the way, leave the original baseline alone unless it is proven guilty.
 
T

tweers

Thank you for your response. One more quick question - what is the recommended number of observations to use for my control limit calculations? I believe the software default is 100 once that many obsevations have been collected.

Thank you
 

Steve Prevette

Deming Disciple
Leader
Super Moderator
The following is my own personal rule, that seems to work through 20 years of experience. Though I will readily admit that there are authors out there that would not agree with me.

As long as I have three changes of direction in the data, such that the data pattern looks like a "M" or a "W", then there is enough for a baseline.
 

bobdoering

Stop X-bar/R Madness!!
Trusted Information Resource
I have to ask - what is the process? It makes a big difference in correctly answering these questions.
 
T

tweers

This is a plastic molding process, in which 1 part is produced every 45 mins - 1 hr. Mainly hand trim and/or CNC secondaries.

Thank you for your follow up.
 

bobdoering

Stop X-bar/R Madness!!
Trusted Information Resource
This is a plastic molding process, in which 1 part is produced every 45 mins - 1 hr. Mainly hand trim and/or CNC secondaries.

For those processes, I would agree with the previous posts on X -MR. As far as how many samples make meaningful control limits - for plastics molding I would use at least 30 parts across 3 set-ups. Your set-up variation may be more significant that your between part variation. If you do not accommodate that, you will have to recalculate control limits each set-up - and that is likely a wasteful exercise (especially if the process is capable.) The exception would be if your tolerance or process robustness of the secondary process totally masks the set-up variation, then 30-100 samples would be good. Be sure to measure the part in the same relative location to capture process variation, and not within-part variation. The process is what you are trying to control.

And, welcome to the Cove!! :bigwave:
 
Last edited:

Bev D

Heretical Statistician
Leader
Super Moderator
If you do have more set-up to set-up variation than part to part, come back as there are approaches that handle this systemic non-homogeneity.
 
T

tweers

As far as how many samples make meaningful control limits - for plastics molding I would use at least 30 parts across 3 set-ups. Your set-up variation may be more significant that your between part variation.

By 3-setups, do you mean 3 measurements on the same characteristic, for each individual part? Or would you mind further explaining that recommendation? I know we experience a lot of variation from part to part, due to the nature of hand trim operations. Other variation we see is natural molding variation that we have less control over, however I'm interested in investigating both.

Thanks again for your help.

:thanx:
 

bobdoering

Stop X-bar/R Madness!!
Trusted Information Resource
By 3-setups, do you mean 3 measurements on the same characteristic, for each individual part? Or would you mind further explaining that recommendation? I know we experience a lot of variation from part to part, due to the nature of hand trim operations. Other variation we see is natural molding variation that we have less control over, however I'm interested in investigating both.

No, I was thinking 3 runs on the injection molding - unless it is dedicated to one part and never stops. I am looking at either mold swaps, or - if not - at least 3 runs after a shutdown (such as after a weekend).

I would expect a lot of variation from a hand process - and no real "adjustment" for those dimensions (or, is there?) - so, more or less "report card" charting for such a dimension.
 
Top Bottom