Visual Inspection Galvanized Attribute Study

[Sean Kelley]

We performed an attribute study of our inspectors visual ability to detect a defect. I have attached a copy of our MiniTab chart in Word for those who may not have Minitab. I am trying to understand what this chart is telling me and should I consider doing anything. To me the results look positive but I wanted some feedback from others who may have more experience with these type of studies. We had 3 inspectors look at 50 samples in a random order and they each did it twice. Thanks.

 

[Bev D]

well, the % agreement is pretty poor. for visual inspectors you are really looking for >95% agreement...but % agreement can be influenced by hte mix of parts in the experiment set and so...

...in my experience this type of chart - by itself - is not very informative. The results have little meaning without understanding the structure of the test:

• what was the mix of passing and failing units?
  • was teh true state of the unit (passing failing) known?
  • what % were passing and failing
  • were there marginal units? what proportion were marginal?
  • was the mix of units close to the actual mix in production? in other words if the defect rate is actually 6% were there only 6% failing units in the study set?
• Did the inspection conditions mimic the actual inspection conditions?
  • same lighting?
  • same speed / time to inspect

unless there is 100% agreemetn to truth and within the operator then the best analysis is done by looking at the 'contingency' table to determine what he 'trouble' areas are. Also it is effective to calculate the % of incorrect failures and incorrect passes to understand the effect on the customer and the business - this tells you whether or not you need to fix any disagreement. and its the disagreement that matters, not the agreement...

 

[Sean Kelley]

what was the mix of passing and failing units? 20 good units / 30 bad units

was teh true state of the unit (passing failing) known? Inspectors did not know anything about number of good vs bad. There is some interpretation but we tried to focus on truly bad and truly good vs borderline units.

what % were passing and failing were there marginal units? what proportion were marginal? There were about 10 margnal units out of the 50

was the mix of units close to the actual mix in production? in other words if the defect rate is actually 6% were there only 6% failing units in the study set? No the number of bad units was much higher for the study than what we would expect to find under normal production runs.
Did the inspection conditions mimic the actual inspection conditions?
same lighting? Samples were laid out in the lab for review under similar lighting conditions. Normally the steel would be moving though so not an exact relica of production. They would have had more time to look at the part and determine if it was good or bad.

same speed / time to inspect

Thanks for your help. We have plan to do more products as this was the first study. I want to make sure we understand what we are looking at and doing it correctly.