MSA on automated measurement system - Multiple Step Vision System

Ron Rompen

Trusted Information Resource
We have acquired a fully automated inspection system for parts which we manufacture. Parts are loaded onto a belt, and picked up by a robot. Vision systems are then used to identify and orient the part, and each part is then subjected to multiple inspection stations (moved by robot at each stage). There is no human intervention in the measurement process.

What is the most appropriate method for conducting an MSA study on this equipment? The classic 10 parts, 3 operators, 3 times each is cumbersome and time consuming, but I have been unable to determine a more appropriate method which also has some form of justification to it.

Suggestions?
 

Bev D

Heretical Statistician
Leader
Super Moderator
30 parts measured twice is a good rule of thumb. If the output is continuous data. If there is a continuous data measurement that is ‘converted’ to pass/fail, get the continuous data. If the vision system is only looking for visual attributes (presence/absence) adn there is no measurement then get a range of parts that are good and bad focusing on parts that close to the cutoff of good/bad. This will tell you the systems ability to properly catch parts at the cutoff. See my resource on validation of measurement systems
 

Miner

Forum Moderator
Leader
Admin
It sounds like you are only interested in quantifying repeatability. However, even an operator free, automated process can have reproducibility elements such as:
  • Multiple testers/stations/fixtures/nests where the same measurement is taken
  • The measurement is sensitive to how the part is loaded into a fixture whether by different people or different robots
If your measurement system has none of these, a straight repeatability study such as Bev recommended is sufficient.
 

Ron Rompen

Trusted Information Resource
Bev/Miner, thanks for your input. Just to answer the questions you posed, the output is continuous data, and I do have a range of parts which include nonconforming samples. There is only a single station/fixture for each measurement, and the part can only be loaded by a single robot (dedicated systems for each station).
 
Top Bottom