Or does everyone here now accept that 3.4 dpmo is historical rubbish ... and we can move on ?
I will buy that a process of no more than 3.4 defect Parts Per Million (PPM) opportunities is a poor
long term target. I like 0 PPM. I have some customers that dig that, too. Handling process capability in the short term utilizing the concept may be acceptable for a normal process, or distributions that are similarly acceptable.
Of course, with the continuous uniform distribution found in properly controlled
precision machining operations, 0 ppm from common causes is rather easy to achieve - short term or long term. I can live with that, too. And, true, one does not achieve that with +/-3 standard deviations from the mean. The mean is not even considered in controlling that process. So, it clearly does not apply to all processes, if that makes you feel better.
I will also buy that Taguchi's loss function is
useless in the short term, because it
needs historical data to determine both the process distribution and the determination of the applicable variation to apply. It can not be derived from thin air. It may better direct long term variation control -
IF one bothers to properly develop the
total variation (that is, the sum of the individual cost variation contributors) that contributes to the loss function. If not, it is just exploitation of academic jargon.