2.78.2 Algorithm for Acceptance Sampling Analysis
Contents
- 1 Attributes Acceptance Sampling
- 2 Variables Acceptance Sampling (Create/Compare)
- 2.1 Sample size and critical distance
- 2.2 Probability of acceptance
- 2.3 Probability of rejecting
- 2.4 Average outgoing quality (AOQ)
- 2.5 Average total inspection (ATI)
- 2.6 Acceptance region (AR)
- 3 Variables Acceptance Sampling (Accept/Reject Lot)
- 4 Reference
Attributes Acceptance Sampling
Sample size and acceptance number
- Sample size (\(n\)) and acceptance number (\(c\)) are the minimum values that satisfy the following inequality
- \[ \text{CDF}(c, n, p_1) < \alpha \]
- \[ \text{CDF}(c, n, p_2) < \beta \]
- \[ 0 < p_1 < p_2 < 1 \]
- \[ 0 < \beta < 1 - \alpha < 1 \]
- where
- \(p_1\): acceptable quality level (AQL)
- \(p_2\): rejectable quality level (RQL or LTPD)
- \(\alpha\): producer's risk
- \(\beta\): consumer's risk
- \( \text{CDF}(c, n, p) \): the cdf function of
- Binomial distribution (\( binocdf(c, n, p) \))
- Poisson distribution (\( poisscdf(c, np) \))
- hypergeometric distribution (\( hygecdf(c, Np, N, n) \)) if the lot size \(N\) is provided.
Probability of acceptance
- The probability that a lot will be accepted, given a specific sampling plan (\(c\) and \(n\)) and the incoming proportion defective or number of defects (\(p\)).
- \[ P_a = \text{CDF}(c, n, p) \]
Probability of rejecting
- The probability that a lot will be rejected, given a specific sampling plan and the incoming proportion defective or number of defects.
- \[P_r = 1 - P_a\]
Average outgoing quality (AOQ)
- The quality level of the product based on the incoming proportion defective or number of defects.
- \[AOQ = \frac{P_ap(N-n)}{N}\]
Average total inspection (ATI)
- The average number of units to be inspected for a certain sampling plan and probability of acceptance.
- \[ATI = n + (1 - P_a)(N-n)\]
Variables Acceptance Sampling (Create/Compare)
Sample size and critical distance
The sample size (\(n\)) and critical distance (\(k\)) depend on the provided specification and whether the standard devidation is known.
Single specification limit and known standard deviation
- \[n=(\frac{Z_\beta+Z_\alpha}{Z_1-Z_2})^2\]
- \[k = \frac{K_1+K_2}{2}\]
- where
- \[K_1 = Z_1 - \frac{Z_\alpha}{\sqrt{n}}\]
- \[K_2 = Z_2 + \frac{Z_\beta}{\sqrt{n}}\]
- \(Z_1\): the \((1-p_1)*100\) percentile of the standard normal distribution
- \(Z_2\): the \((1-p_2)*100\) percentile of the standard normal distribution
- \(Z_\alpha\): the \((1-\alpha)*100\) percentile of the standard normal distribution
- \(Z_\beta\): the \((1-\beta)*100\) percentile of the standard normal distribution
- \(p_1\): acceptable quality level (AQL)
- \(p_2\): rejectable quality level (RQL or LTPD)
- \(\alpha\): producer's risk
- \(\beta\): consumer's risk
Single specification limit and unknown standard deviation
- \[n = (1 + \frac{k^2}{2})(\frac{Z_\alpha + Z_\beta}{Z_1 - Z_2})^2\]
- \[k = \frac{Z_\alpha Z_2 + Z_\beta Z_1}{Z_\alpha + Z_\beta}\]
Double specification limits and known standard deviation
- \[n=(\frac{Z_\beta+Z_\alpha}{Z_{p_L}-Z_2})^2\]
- \[k = \frac{K_1+K_2}{2}\]
- where
- \[K_1 = Z_{p_L} - \frac{Z_\alpha}{\sqrt{n}}\]
- \[K_2 = Z_2 + \frac{Z_\beta}{\sqrt{n}}\]
- \(Z_{p_L}\): the \((1 – p_L) * 100\) percentile of the standard normal distribution
- \(p_L\): determined using the following procedure:
- \(p_L = \text{max}\left\{ \Phi(\frac{L-\mu}{\sigma}), 1 - \Phi(\frac{U-\mu}{\sigma})\right\}\), where \(\mu\) is chosen such that \(\Phi(\frac{L-\mu}{\sigma}) + 1 - \Phi(\frac{U-\mu}{\sigma}) = p_1\)
- \( p_L = p_1 \), if \(\Phi(\frac{L-\mu}{\sigma}) + 1 - \Phi(\frac{U-\mu}{\sigma})\) deviates substantially from \(p_1\)
- \(\Phi\): the cdf of the standard normal distribution
- \(L\): the lower spec
- \(U\): the upper spec
- \(\sigma\): the historical standard deviation
Double specification limits and unknown standard deviation
- \[n = (1 + \frac{k^2}{2})(\frac{Z_\alpha + Z_\beta}{Z_1 - Z_2})^2\]
- \[k = \frac{Z_\alpha Z_2 + Z_\beta Z_1}{Z_\alpha + Z_\beta}\]
Calculation of Maximum Standard Deviation (MSD)
- Default Method[1]
- \[v_1 = \max\left\{0, \frac{1}{2} - \frac{1}{2}k\frac{\sqrt{n}}{n-1}\right\}\]
- \(p = Beta(v_1, a, b)\), where \(a = b = \frac{n-2}{2}\) and \(Beta\) is the CDF of beta distribution.
- \[v_2 = Beta^{-1}(p/2,a,b)\]
- \[k^* = \frac{(n-1)(1-2v_2)}{\sqrt{n}}\]
- \[MSD = \frac{U-L}{2*k^*}\]
- Wallis Method[2]
- \[p = 1 - \Phi(k)\]
- \[Z_p = \Phi^{-1}(1-p/2)\]
- \[MSD = \frac{U-L}{2*Z_p}\]
Probability of acceptance
- The probability that a lot will be accepted, given a specific sampling plan (\(n\) and \(k\)) and the incoming proportion defective (\(p\)).
Single specification limit and known standard deviation
- \[ P_a = \Phi(\sqrt{n}(Z_p-k)) \]
- where
- \(Z_p\): the \((1-p)*100\) percentile of the standard normal distribution
- \(p\): probability of defective
Single specification limit and unknown standard deviation
- \( P_a = 1 - nctcdf(\sqrt{n}k, n-1, \sqrt{n}Z_p) \), where \(nctcdf\) is the CDF of non-central t distribution.
Double specification limits and known standard deviation
- Use the following procedure
- \(P_a = \Phi(\sqrt{n}(\frac{U-\mu}{\sigma}-k)) - \Phi(\sqrt{n}(\frac{L-\mu}{\sigma}+k))\), where \(\mu\) is chosen such that \(\Phi(\frac{L-\mu}{\sigma}) + 1 - \Phi(\frac{U-\mu}{\sigma}) = p\)
- \(P_a = \Phi(\sqrt{n}(Z_p-k))\), if \(\Phi(\frac{L-\mu}{\sigma}) + 1 - \Phi(\frac{U-\mu}{\sigma})\) deviates substantially from \(p\)
Double specification limits and unknown standard deviation
- \[ P_a = 1 - nctcdf(\sqrt{n}k, n-1, \sqrt{n}Z_p) \]
Probability of rejecting
- The probability that a lot will be rejected, given a specific sampling plan and the incoming proportion defective.
- \[P_r = 1 - P_a\]
Average outgoing quality (AOQ)
- The quality level of the product based on the incoming proportion defective or number of defects.
- \[AOQ = \frac{P_ap(N-n)}{N}\]
Average total inspection (ATI)
- The average number of units to be inspected for a certain sampling plan and probability of acceptance.
- \[ATI = n + (1 - P_a)(N-n)\]
Acceptance region (AR)
- Default Method
- \[v_1 = \max\left\{0, \frac{1}{2} - \frac{1}{2}k\frac{\sqrt{n}}{n-1}\right\}\]
- \(p = Beta(v_1, a, b)\), where \(a = b = \frac{n-2}{2}\) and \(Beta\) is the CDF of beta distribution.
- for each pair of \((v_{01}, v_{02})\) so that \(Beta(v_{01},a,b) + Beta(v_{02},a,b) = p\)
- \[\bar{x} = \frac{U(1-2v_{01})-L(2v_{02}-1)}{2(1-v_{01}-v_{02})}\]
- \[s = \frac{U-L}{2(1-v_{01}-v_{02})}(\frac{\sqrt{n}}{n-1})\]
- Wallis Method
- \[p = 1 - \Phi(k)\]
- for each pair of \((p_{01}, p_{02})\) so that \(p_{01} + p_{02} = p\)
- \[\bar{x} = \frac{UZ_{p_{01}}+LZ_{p_{02}}}{Z_{p_{01}}+Z_{p_{02}}}\]
- \[s = \frac{U-L}{Z_{p_{01}}+Z_{p_{02}}}\]
Variables Acceptance Sampling (Accept/Reject Lot)
- Mean and standard deviation are first calculated from the data.
- \[\bar{X} = \frac{\sum X_i}{n}\]
- \[s = \sqrt{\frac{\sum (X_i-\bar{X})^2}{n-1}}\]
- Then calculate the acceptance criteria
- \[Z_L = \frac{\bar{X}-L}{s}\]
- \[Z_U = \frac{U-\bar{X}}{s}\]
- Note that if historical standard deviation is provided, then \(\sigma\) is used for the \(Z\) calculation.
Single specification limit
- If only lower spec is provided, then accept the lot if \(Z_L \ge k \)
- If only upper spec is provided, then accept the lot if \(Z_U \ge k \)
Double specification limits and known standard deviation
- Accept the lot if \(Z_L \ge k \) and \(Z_U \ge k \)
Double specification limits and unknown standard deviation
- Accept the lot if \(s \le MSD \), \(Z_L \ge k \) and \(Z_U \ge k \)
Reference
- Duncan, A. J. (1986). "Quality Control and Industrial Statistics (5th ed.)". Homewood, Ill: Irwin
- Schilling and Neubauer (2009). "Acceptance Sampling in Quailty Control (2nd ed.)"