mil-std_414
TRANSCRIPT
• Variables Sampling Plans
• MIL-STD-414 (ISO 3951)
Variables Sampling Plans
• Variables sampling plans are based onthe numerical measurement of qualitycharacteristics
• Advantage: smaller samples arerequired than for attributes samplingplans.
• Disadvantages:
1. Precise measurements are costly.
2. If more than one characteristic ismeasured, a separate plan isneeded for each characteristic.
3. Measurements are assumed tofollow a normal distribution.
4. More computational effort isrequired.
• Example.
* Suppose a 1250 MHz Pentium chipis considered defective if its actualclockspeed is less than 1200 MHz.
* ⇒ LSL = 1200 MHz
* A lot is acceptable if no more than2 % of the chips in the lot aredefective.
* Reject such lots no more than 5 %of the time.
* ⇒ AQL = .02.
* If clockspeed is a normallydistributed random variable X withmean µ and variance σ2 (say 400),we can calculate the µ = µAQL thatcorresponds to the AQL:
.02 = P (X < LSL)
= P (Z <LSL− µAQL
σ)
* ⇒LSL− µAQL
σ= −2.05
so that
LSL− µAQL = −2.05σ.
Therefore,
µAQL = LSL + 2.05σ = 1241
* If the true mean clockspeed for alot is 1241 MHz or higher, the lot isconsidered acceptable.
* Problem: µ is unknown if we don’tinspect every item in the lot.
* Solution: Take a random sample ofchips from the lot and test the nullhypothesis that µ = 1241 versus thealternative hypothesis thatµ < 1241.
Ho : µ = 1241 acceptable lot
HA : µ < 1241 unacceptable lot
* Test statistic:
Z =X̄ − 1241
σ/√
n
where X̄ is the average of a randomsample of n clockspeedmeasurements. Reject Ho (i.e.reject the lot) if
Z < −zα = −1.645.
* ⇒ reject the lot if
X̄ − 124.1 < −1.645σ/√
n
or
X̄ < 124.1− 3.29/√
n.
* In general, reject the lot if
X̄ < (zAQL − z1−Pa(AQL)/√
n)σ + LSL
where zAQL is defined so that
P (Z > zAQL) = AQL.
* To determine the sample size n,use the LTPD.
* Suppose β = .1 and LTPD = 20 %
* ⇒ accept lots with 20 % defectiveno more than 10 % of the time.
* Let µLTPD be the mean clockspeedfor lots at the LTPD.
.2 = P (X < LSL) = P (Z <LSL− µLTPD
σ)
* ⇒LSL− µLTPD
σ= −.84
so
µLTPD = LSL + .84σ = 1216.8.
* ⇒ the mean clockspeed is 1216.8MHz in lots which are at the LTPD.
* We wish to reject 90 % of all suchlots.
* That is, we require
P (X̄ < 1241− 3.29/√
n) = .9.
* Standardize:
P (Z <1241− 3.29/
√n− 1216.8
2/√
n) = .9.
⇒
1.21√
n− 1.645 = 1.28.
We can solve for n:
n = 5.8
• In general,
n =
(
zα + zβ
zAQL − zLTPD
)2
• For one-sided tolerances of USL type,this is the formula for the sample size,but now reject the lot if
X̄ > USL− (zAQL − zα/√
n)σ.
• In both cases (USL or LSL), we canwrite the decision rule as reject thelot if
QL < k
or
QU < k
where
k = zAQL − zα/√
n
and
QL =X̄ − LSL
σand
QU =USL− X̄
σ.
• In other words, we reject a lot if thesample average is ’too close’ to theLSL (or USL).
• Example Find the sample size requiredfor a variables sampling plan with AQL= .1, LTPD = .2, α = .05 and β = .1.
*
n =
(
z.05 + z.1z.1 − z.2
)2
n =(1.645 + 1.28
1.28− .84
)2
n = 44.1
* ⇒ n = 45 should be used.
• If σ = 1, LSL = 10 and X̄ = 11, what isthe lot sentence under the aboveplan?
* k = 1.28− 1.645/√
45 = 1.03.
* QL = 11−101 = 1 < k. Therefore,
reject the lot.
MIL-STD-414
• aka ISO 3951.
• Another acceptance sampling system.It is actually comprised of severaltypes of plans. We will consider thecase in which σ is unknown, and isestimated using s.
• Example. Suppose LSL = 1200 MHz,AQL = 2 %, α = .05, β = .1 and lotsize is 1000. Find an appropriatesampling plan.
* From Table 11.1 (p. 351, Farnum),we convert AQL of 2 % to 2.5 %.
* From Table 11.2 (p. 352), with alot size of 1000, we use sample size
code letter K (under inspectionlevel IV).
* From Table 11.3 (p. 354), underAQL = 2.5 %, and across fromcode letter K, we find that thesample size n = 35, and M = 5.57,under normal inspection. Thismeans that if the estimatedproportion defective based on the35 measurements is greater than5.57 %, we reject the lot.
• Suppose 35 measurements are takengiving x̄ = 1230 and s = 20. What isthe lot sentence?
* Then QL = x̄−LSLs = 1230−1200
20 = 1.5
* Table 11.5 (p. 356) is used toconvert this number to an estimateof the fraction nonconforming:pL = 6.50% using QL = 1.5, n = 35).
* Therefore, we would reject the lot.
* If x̄ = 1235 and s = 20, we getQL = 1.75 so that pL = 3.77%. Inthis case, we would not reject thelot.