I am going to jump in with a side discussion on in-process inspection. The production manager want to pull samples in process based on the scheduled lot size. Example 1-2 piece lot size- inspect 1- reject on one failure; 3-8 pieces- inspect 2, reject on 1st failure; 9-15 pieces-inspect 3, reject on 1st failure and reinspect all since the last accepted sample, etc up to a lot size > 1200 pieces- inspect 50, reject on 1st failure and reinspect all back to last approved sample. The sample pieces would be drawn on every 10th piece. My position is this is like playing Russian roulette with the sampling plan.
If the process has a fixed, single assignable cause that once introduced would result in all pieces beyond that point to be defective, I might agree with this plan. However, if the assignable cause or causes occurred randomly, this plan is faulty. My position is to sample between 3 and 5 pieces relatively close together or pull them consecutively and spread the sample frequency out so as to sample in a level plan through out the run and move toward using an Attribute SPC chart (any defect in the part = a reject).
Additionally, from my viewpoint, following the production manager's method means the probability of pulling a defective part drops for a simple probability of 50% to 33% @10 pieces where the lot size no longer matters because every tenth piece then checked and the probability level out at 10%. I have not run a true probability of selecting a defective part a the historical defect rate is unknown, but I plan to test with several theoretical defect rates. I am just looking for some practical feedback.
The process under question is capable of single stream output and also multiple piece output. Think extruding a single thread or a bundle of threads cut to length. The manager's plan becomes less desirable as the bundle size increases.
The argument is simplicity vs. effectiveness of the approaches.
If the process has a fixed, single assignable cause that once introduced would result in all pieces beyond that point to be defective, I might agree with this plan. However, if the assignable cause or causes occurred randomly, this plan is faulty. My position is to sample between 3 and 5 pieces relatively close together or pull them consecutively and spread the sample frequency out so as to sample in a level plan through out the run and move toward using an Attribute SPC chart (any defect in the part = a reject).
Additionally, from my viewpoint, following the production manager's method means the probability of pulling a defective part drops for a simple probability of 50% to 33% @10 pieces where the lot size no longer matters because every tenth piece then checked and the probability level out at 10%. I have not run a true probability of selecting a defective part a the historical defect rate is unknown, but I plan to test with several theoretical defect rates. I am just looking for some practical feedback.
The process under question is capable of single stream output and also multiple piece output. Think extruding a single thread or a bundle of threads cut to length. The manager's plan becomes less desirable as the bundle size increases.
The argument is simplicity vs. effectiveness of the approaches.
Last edited: