# Sampling freq

Six Sigma – iSixSigma › Forums › Old Forums › General › Sampling freq

- This topic has 6 replies, 6 voices, and was last updated 11 years, 4 months ago by annon.

- AuthorPosts
- August 29, 2008 at 7:32 am #50841
I wanna change a sampling frequency from 3x/day to 2x/day. What are the appropriate data shd I present to demo that the change is feasible and that doing 3x/day is non-value added?

0August 29, 2008 at 7:57 am #175303

Michael MeadParticipant@Michael-Mead**Include @Michael-Mead in your post and this person will**

be notified via email.Aha…I was hoping somebody would ask this some day!

This should be a purely economic decision. You should balance the cost of the inspection (time to measure each part, time to record the sample) with the cost of defects or out-of-control condition. For example, if you produce 300 pieces per day, and you find you need to sort, you would sort either 100 parts or 150 parts. Plus the cost of the defective products in your sort.

I suggest you read Dr. Montgomery’s, “Introduction to Statistical Quality Control” book–the section on Economic Design of Control Charts.

This calculation can be hard to do because labor is easy to track but the cost of defects is not. Companies have a budget for labor, and no budget for scrap. I have done this type of calculation many times and I have NEVER seen a time when reduced inspection or sampling paid for itself. I have wrtten several papers about this, and I would gladly send you some documentation and an Excel spreadsheet to do the optimization. My e-mail is [email protected]0September 2, 2008 at 10:47 am #175374

J. AnguloParticipant@J.-Angulo**Include @J.-Angulo in your post and this person will**

be notified via email.I have developed risk-benefit analyses of this type at my firm, and I have found that a stochastic simulation works fairly well. First, you can model the RISK by determining the historical failure rate for the process (DPMO) and the typical distribution of your process. Then, you can use Microsoft Excel to develop a set of randomly-generated values that reflect (model) the behavior of the process and set constraints that reflect your pass/fail or specifications criteria. Depending on the complexity of your scenario, the comparison can be as simple as “simulating” the outcome on two different columns, one for 2X a day sampling and another for 3X a day sampling. Alternatively, you can develop a macro in Excel to iterate and summarize you data.

Finally, you determine the costs associated with each sampling scenario.0September 2, 2008 at 2:38 pm #175379Wow, just what we need as a SS advocate – a guy who has never seen the economics in less inspection!

Michael, go read Deming, go read Shingo.

Most of us with experience have completely eliminated inspection as it is known in a traditional sense. It is the only way.0September 3, 2008 at 7:16 am #175404

Michael MeadParticipant@Michael-Mead**Include @Michael-Mead in your post and this person will**

be notified via email.Who said anything about inspection in the classic sense? What I recommended was a method to optimize three sampling parameters for an X-bar chart: sample size, frequency, and width of control limits. Neither Deming nor Shingo would disagree with this. However, rereading some of their contributions is always worthwhile.

I certainly understand the economics involved. This is where my Marginal Inspection Theorem comes from. It states: Additional inspection should be performed up to the point where the marginal cost of a unit of inspection is equal to the failure cost it prevents.

Two corollaries follow from this: 1) If a product failure is expensive or can cause serious downstream damage, inspect it. 2) If inspection is expensive or destructive, perform as little inspection as necessary.

Most (and I mean almost ALL) companies do not have the process knowledge to optimize their sampling plans. For example, they do not know the frequency or magnitude of process shifts or the full cost of producing and replacing a defective product. What they do understand is the labor cost associated with measuring and charting a sample, and this is what they minimize.

The diagram below is from my presentation at the American Statistical Association Fall Technical Conference on this subject.

0September 3, 2008 at 10:31 am #175409

SeverinoParticipant@Jsev607**Include @Jsev607 in your post and this person will**

be notified via email.My guess based on what the OP has posted is that he can probably just change the frequency since there is probably no statistical (or economical) basis for why he is currently doing 3x per day (or anything justifying his current sample size either). If such a rationale existed at his company he would already know what he had to justify to his management to change it to 2x a day.

What he probably should be asking, as Stan points out, is “Why am I sampling at all?”0September 3, 2008 at 2:00 pm #175416Inspection is useful only as a means to assess process capability and provide actionable data back to production for root cause analysis and solution. As a means of controlling the the quality of your process, it is always a bad idea in any form other than as short-term containment strategy.

0 - AuthorPosts

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