Quality Control Problems Using Probability Theory
Quality Control Problems Using Probability Theory
A quality-control calculation begins with the inspection rule: decide which observed samples pass and which detect a defect. A probability describes repeated sampling under a stated model; it does not reveal the condition of every uninspected item.
Independent sampling
Suppose each inspected item has the same defect probability p, and the outcomes are independent draws. The number K of defects in n inspections is a binomial count. Zero defects has probability P(K=0)=(1−p)n; its complement, detecting at least one defect, has probability 1−(1−p)n.
For an at-most-c rule, the acceptance probability is P(K≤c)=∑k=0c(kn)pk(1−p)n−k. Add the disjoint allowed counts; do not silently replace “at most one” with “none.” If the rule accepts only a zero-defect sample, the first term alone gives the pass probability.
Finite-lot sampling
A fixed lot with N items and exactly D defects changes composition when sampled without replacement. Uniformly chosen samples of size n therefore have a hypergeometric count: P(K=k)=(kD)(n−kN−D)/(nN). The denominator counts all equally likely samples, while the numerator chooses defective and good items for the specified count.
Use only feasible values of k: at least zero and n−(N−D), and at most n and D. As with independent sampling, add the counts allowed by the rule or subtract the disallowed event from one. The fixed-lot calculation requires a known defect count; a defect rate alone does not specify the composition of a particular finite lot.
Worked example
Suppose four inspections are independent, each with defect probability 0.25, and a sample passes only with zero defects. The good-item probability is 0.75, so P(pass)=0.754=0.31640625. The complementary chance of detecting at least one defect is 1−0.31640625=0.68359375.
These values are operating probabilities for the stated sampling model. Repeating the inspection many times under the same conditions would produce varying samples; one passing sample does not prove the whole lot is defect-free.
A lot contains 10 items, exactly 2 defective. Inspect 2 distinct items uniformly without replacement. What is the probability of detecting at least one defect?
A lot of 20 contains exactly 3 defective items. Inspect 4 uniformly chosen items without replacement; accept the lot if the sample has at most one defect. What is the acceptance probability?