# What and how do we know/calculate +/- 2 simga

Six Sigma – iSixSigma Forums Old Forums General What and how do we know/calculate +/- 2 simga

Viewing 11 posts - 1 through 11 (of 11 total)
• Author
Posts
• #46173

christina
Participant

I have the following reqeust from customer:
action needed during setting up if average data cannot fall in CL +/- 2 sigma and adjust until achieving “within +/- 1 sigma”
— if the average data fall between +/-1 and +/-2 sigma, re-sampling and adjust until achieving “within +/-1 sigma”
we are talking about a spec of 0.5548 +/- 0.05mm.
Question: How do I calculate if my process will meet +/- 2 or 1 sigma
Thanks

0
#152253

accrington
Participant

What drugs is your customer on?
If it’s in statistical control, 95% of your process will meet +/- 2 SDs either side of the mean, 68% will meet +/- 1 SD of the mean. Process standard deviation is defined by the process not the customer’s spec.
You are in a difficult position here, because it is the customer who is imposing this incomprehensible nonsense upon you. Is your customer a reasonable person who is open to discussion? Is he aware of the chaos his request will cause to your process/ product?
By CL, do you mean control limits, as calculated from the process, or do mean limits that the customer has defined for you.

0
#152255

clb1
Participant

The problem we have here Christina is “CL” can mean both Control Limits – which could be limits imposed externally by the customer and Confidence Limits – which are (typically) the 95% limits that are part of the physical properties of the process.  In order for anyone to offer anything meaningful you will have to tell us what you and your customer mean by “CL”.

0
#152256

accrington
Participant

Dear clb1, I think you may a little confused here.
Control limits (the Voice of the Process) are defined by the natural capability of the process. They cannot be imposed by the customer, who can define specification limits (the Voice of the Customer), but the spec cannot ‘make’ the process behave differently to its natural capability.
Confidence limits are not part of the physical properties of the process. Confidence limits estimate an interval within which the true (population) mean may lie if we were to sample repeatedly from the same population. For a 95% confidence interval, if we collected lots of samples, in the long run, about 95% of these intervals will contain the true mean, (assuming that we are dealing with a single population distribution)
Since we cannot assume that a process which takes place over time and is subject to constant chance cause disturbance is a single population (unless we have some evidence of stability over time), or that we can ever know the form of its distribution, confidence limits are irrelevant in this context (let’s hope that’s not what Christina’s customer means).

0
#152269

christina Lim
Participant

in this particular case, CL refers to control limits

0
#152268

christina Lim
Participant

CL refers to Control Limits
I really do not know what this customer wants and how do I know whether I am following his rule.

0
#152271

Ken Feldman
Participant

If your customer expects that all your process data will perform within +/- 2 s.d.  or +/-1 s.d. based on the control limits then he is a friggin idiot as was pointed out in earlier posts.  The process and the corresponding calculated control limits are what they are.  If that is not acceptable from a spec limit perspective then the process needs to be changed.  Wishing the data within 1 or 2 s.d. for the control limits is stupid.  The only way you can guarantee that his product will not exceed that is to do a 100% inspection and cull out all product not within those limits.  I think there is still some confusion regarding whether the customer is referring to some spec limit versus a control limit.  I would bet that he is concerned with the product meeting the spec limits, not the control limits.  You need to get a bit more clarification.

0
#152272

GDS
Participant

Christina,
According to your first post, your SPEC limints are to produce your widget to 0.5548mm + or – 0.05mm (0.4048 to 0.6048).
Where is your process operating at?
What is your mean & stdev?
What does your current control chart look like?
What are your UCL & LCL?
What does a historgram look like?
How many widgets do you produce?

0
#152295

goodie
Participant

Christina,
Sometimes customers don’t know what they really want (especially japanese) but we as suppliers need to comply with their demand.
You can play with them. Since you need to have the initial set of data from your process to get your CL, then loosen your process which will give you a very loose control limits. After that tighten your process-I’m sure the data you will be collecting is within +/- 2sigma or even +/- 1sigma.
this is not a good advise, just in case you’re trapped, you have a way to save yourself.
goodie

0
#152297

Brown
Participant

What utter rubbish!

0
#152331

BritW
Participant

Th eonly way for you to achieve the request is to follow Darth’s and Accrington’s posts.

Fix the process so it is in control.
Do 100% inspection
Trash the 5% that will inevitably be outside the +/- 2 s.d. range
Charge the customer for the inspection and waste costs or eat the loss
Your process will naturally run between +/- 3 s.d.’s.  You will never meet the customer request without waste.  My gut tells me (like some others) that the customer wants the parts to fall within a certain spec limit.  Control limits are irrelevant to the customer as long as they are receiving parts within spec at the price they agreed to.  The control limits are for process improvement – you can always throw away the bad stuff before it reaches the customer – it will cost you more, but why should the customer care if it costs you more to produce what they need as long as they are paying the same amount?
If your customer really wants the control limit language, then I suggest some education from your end to describe the idiocy of their request.

0
Viewing 11 posts - 1 through 11 (of 11 total)

The forum ‘General’ is closed to new topics and replies.