My appraoch is to study all methods – each has something to offer. For example, Shainin taught me to test the repeatability within a DOE cell as compared to cell to cell variation before continuing with large and expensive DOEs. This has stood me in good stead; especially with ‘component search DOE’ where levels are determined from test results. From Taguchi I learnt that my levels can sometimes vary, depending on the resolution of the controller, or as a percentage of FSD. From classical statistics I learnt that random independence is a fundmental assumption of most statistical methods, which is why we need to consider carefully what visually inspected DPMO really tells us; before our entire manufacturing industry is tranported over to the Far East. (No offence to anyone over there.) I remember well a subsidiary of a USA company complaining to me that while he reported all defects on his PCBs, and they were in the thousands, his mirror facility in the States did not. As someone with some process engineering experience I reminded him about the assumption of random independence, but his response was that he was not going to cheat. His facility is now shutdown, presumably on the basis of the visual inspections; since few PCB manufacturers in the UK rarely electrically test boards! (Nor do they perform design rule checks!)
.
