Using ANSI/ASQ Z1.4 to Reduce Impact of Field Service Campaign

S

shadowjade

All,

I represent a manufacturer of complex systems (i.e. not nuts, bolts, or fasteners) with an issue I'd like some input on. A situation has been identified with older product wherein the responsibility in the matter is shared between us (the supplier) and our customer. The response will be to perform a field service campaign to inspect over 900 units.

I have been asked to see if statistical sampling can help reduce the amount of units requiring inspection. I've deferred to ANSI/ASQ Z1.4-2003 and am hopeful for clarification from others here as to my understanding of this standard. Our customer has little or no understanding of the standard so I will have to painstakingly describe it in detail.

At a quantity of 80 units for sampling, we conform to code letter J thereby representing general inspection level II for the fielded quantity of over 900 units. We could withstand 5 rejects at an AQL of 2.5 (normal inspection) at the aforementioned quantity inspected. We can withstand 3 rejections with tightened inspection (same AQL).

Anything rejected beyond this necessitates 100% inspection. In my view, it should be further understood that with both normal and tightened inspection we should be looking at C=0 of zero [0] rejections. The AQLs in these cases are 0.15 and 0.25 respectively.

My questions are: should a switching rule (i.e. if we find 0 defectives before reaching 80 units inspected) be considered? How should I explain the AQL of 2.5 versus 0.15 and 0.25 from a risk perspective?

Please respond when you've had a chance to consider this.

Thank you.
 
S

supreecha

Normal to Tightened
When normal inspection is in effect, tightened inspection shall be instituted when 2 out of 5 (or fewer) consecutive lots or batches have been non-acceptable on original inspection (i.e., ignoring resubmitted lots or batches for this procedure).
ANSI/ASQ Z1.4-2003 Standard

In December 2003 the ASQ (American Society for Quality) released the ANSI/ASQ Z1.4-2003 standard. This is a revision to the ANSI/ASQC Z1.4-1993 standard. The ASQ began shipping this new standard revision in middle January 2004.

Changes in ANSI/ASQ Z1.4-2003 revision
  1. The name of the standard has been changed to drop the "C". This reflects the 1997 name change of American Society for Quality Control to the current American Society for Quality.
  2. The definition of AQL
  1. has been changed from Acceptable Quality Level to Acceptance Quality Limit.
  2. The Discontinuation of Inspection rule has been changed from 10 consecutive lots or batches on tightened inspection to 5 consecutive lots or batches not accepted on tightened inspection.
  3. Double Sampling* footnote
  1. [/URL] has been changed. The option to use an alternate double sampling plan has been deleted from this note. Users are directed just to use the corresponding single sampling plan.

    1. There are similar footnote changes for Double Sampling (AQLs 25 and above) and Multiple sampling that do not apply to the AQL Inspector's Rule.
      [*]The notes in this revision specify that the numbers and tables remain the same as the MIL-STD-105E execpt for the changes to the footnotes.
    [/COLOR][/COLOR]The only change to the AQL Inspector's Rule (ANSI version) for the ANSI/ASQ Z1.4-2003 revision will be item 4 above. None of the numbers, sample plans or operation of the AQL Inspector's Rule changes with this revision.
    The AQL Inspection Manual will be updated in 2004 to include the items list above.
    All previous AQL Inspector's Rule's (ANSI version) conform to and may be used for inspection to the ANSI/ASQ Z1.4 2003 and 2008 revisions with the one note listed in item 4 above.

    Acceptance Sampling by Attributes by Minitab

    Measurement type: Go/no go
    Lot quality in percent defective
    Lot size: 900
    Use binomial distribution to calculate probability of acceptance


    Acceptable Quality Level (AQL) 2.5
    Rejectable Quality Level (RQL or LTPD) 10


    Compare User Defined Plan(s)

    Sample Size 80
    Acceptance Number 0

    Accept lot if defective items in 80 sampled <= 0; Otherwise reject.


    Percent Probability Probability
    Defective Accepting Rejecting AOQ ATI
    2.5 0.132 0.868 0.301 791.8
    10.0 0.000 1.000 0.002 899.8

    Average outgoing quality limit (AOQL) = 0.416 at 1.235 percent defective.
 
Last edited by a moderator:
S

sjared

I've been watching this thread with curiosity to see how others would respond because my basic understanding of AQL is that it does not predict the conformity for any one given lot/batch. Rather, it reflects a process average over multiple manufacturing batches. Refer to this thread here:

http://elsmar.com/Forums/showthread.php?t=19186

I am more inclined to think that an exact binomial confidence interval would be more applicable method for this situation as outlined here:

http://elsmar.com/Forums/showthread.php?t=47530

But I could be missing something.
 
S

shadowjade

I appreciate the feedback but am struggling with applying it to our situation. If I stay with an 80-piece sample, do I or do I not conform an AQL of 2.5 if I see less than 5 defectives?

Thank you for any additional insight in this matter.
 
S

sjared

I appreciate the feedback but am struggling with applying it to our situation. If I stay with an 80-piece sample, do I or do I not conform an AQL of 2.5 if I see less than 5 defectives?

Thank you for any additional insight in this matter.

I would say you do not conform with 2.5. I use this online calculator. If 80/80 pass, the exact confidence interval is .9632 - 1, or 3.7% - 0% possible defective for the population.
 
S

shadowjade

supreecha:

How did you determine an RQL of 10? I, too, have Minitab and have been able to re-create the results you provided with your initial response. Could you further define the "Acceptance numbers" field?

My thanks to all who have responded to this thread.
 

Statistical Steven

Statistician
Leader
Super Moderator
Z1.4 is INAPPROPRIATE for this application. If you want to inspect a sample of the 900 units in the field, you need accept zero defects, set a confidence level and set a reliability (percent defective in the field that goes undetected). The formula is

N=ln(1-confidence)/ln(reliability)

For example 95% confidence with 90% reliability would be a sample size of 35. Meaning we are 95% confident that if we inspect 35 units, all are acceptable that the maximum percent defective is 10%. You would need 300 to get the reliability up to 99%.
 
S

shadowjade

Steven,

Is the reasoning for Z1.4 being inappropriate in this matter is in, as another response cited, "it reflects a process average over multiple manufacturing batches"?

The formula you provided may well be what we better present to the customer in our situation (who, as was noted, will be sharing the cost). Please further define the "ln" components of the formula or, if possible, refer me elsewhere for more examples of this in practice.

Thank you for your insight and continued consideration in this matter.
 

Statistical Steven

Statistician
Leader
Super Moderator
Steven,

Is the reasoning for Z1.4 being inappropriate in this matter is in, as another response cited, "it reflects a process average over multiple manufacturing batches"?

The formula you provided may well be what we better present to the customer in our situation (who, as was noted, will be sharing the cost). Please further define the "ln" components of the formula or, if possible, refer me elsewhere for more examples of this in practice.

Thank you for your insight and continued consideration in this matter.

The easiest part first...ln is the natural log (base e)

Z1.4 is for inspection of batches of materials and the AQL is a process average. More importantly, the switching rules deal with batch pass/fail not units within a batch.

Look at this thread http://elsmar.com/Forums/showthread.php?t=31317
 
S

shadowjade

Steven,

Please advise me given the following criteria:
  • Total remaining units to sample (for which we're looking to determine a sample size) = 920
  • Defects found outside of the 920 (which precipated this whole endeavor) = 4
  • Assume customer would prefer 99% reliability.

Are we still looking at 300 units? Is the confidence level at 95%?

Thank you for your continued work in this regard.
 
Top Bottom