Sampling a continuous process - How to use AQL with a continuous process

[MikeW]

Does anyone know how to use AQL with a continuous process?

In my application, I am welding a girth seam on a tank. It is a 3 pass double butt weld, with each pass being 377 inches long.

The current inspection criteria is to allow no more than 2 defects per seam. Defect types include slag and porosity (defect when they exceed 1/4") and nonfusion and cracks (any size is a defect).

I inspect for defects in the weld seam with x-ray film. It takes 27 shots to wrap the circumference.

If I weld 42 seams in a shift, is my batch size 42?
Or is the batch size 42 seams x 27 shots = 1134, the number of discrete shots I can review?

Or is the batch size 42 seams x 377" long seam x 4 = 63,336?
(Since a defect is 1/4" or larger, there are 4 opportunities per inch)

Thanks

[Mike S.]

Some of the stats/SPC experts like Steve Prevette can probably answer this better. And I know mostly nothing about welding. So I guess this makes me qualified to comment.

I'd say your lot size is 42 which is the # of seams. Consider a seam 1 item of your "product".

But I don't know that AQL sampling is appropriate here. Is this a critical weld? What inspection do you do now -- 100% or something else? What does the customer say?

[Marc]

I think one the seminal 'link' thread here on continuous process sampling is:

http://Elsmar.com/Forums/showthread.php?t=1983

Hopefully some other folks will be better able to more directly answer your questions.

[MikeW]

MikeS

The seam defects are probably classified as minor, since the code allows for a sampling plan to be used.

The code calls out that the first 3 seams should be 100% inspected. If each seam has less than 2 defects, then the process can be sampled on a 1 in 20 seam basis. If a sampled seam has more than 2 defects, then the inspection goes back to three good ones in a row before sampling can resume.

Mike W

[Tim Folkerts]

Mike,

Ulitimately, the answer comes down to satisfying the customer (and satisfying any legal requirements for your industry). You could calculate the defect rate any way you want, but of course that doesn't change the actual quality.

If you want to use MIL STD 105E = ASQ Z1.4 sampling plans and AQL, you would simply pick a different AQL depending on which number you use in the calculations.

You say 2 defects per seam is acceptable. Assuming you use a Normal, Level II plan,

• If you use each seam as a unit, AQL = 150, batch size = 42
  
  
  
  (2 defects per unit = 200% defect rate, rounded down to the next lower number on the table)
  
  
  
  So you would inspect 8 seams and reject with 22 defects

• If you use each x-ray as a unit, AQL = 6.5, Batch size = 1134
  
  
  
  (2 defects per 27 units = 7.4%, rounded to 6.5)
  
  
  
  So you would inspect 80 x-ray films and reject if there were 11 defects.

The first option would entail fully inspecting 8 tanks. The second would entail the equivalent of only about 3 seams, but drawn at random from parts of all 42 tanks.

WARNING: AQL is the defect rate that will most likely be accepted if the lot is that quality. If you accept the batch, you are not saying the batch is good - you are saying you couldn't prove it was bad. Batches with a 4 defects per seam (rather than the desired 2) would still pass inspection most of the time.

Tim Folkerts


P.S. The length of your seams for the 42 tanks is just a few inches less than 1/4 mile! Just a bit of trivia for you boss or the welders.

[Mike S.]

Originally Posted by MikeW

MikeS

The seam defects are probably classified as minor, since the code allows for a sampling plan to be used.

The code calls out that the first 3 seams should be 100% inspected. If each seam has less than 2 defects, then the process can be sampled on a 1 in 20 seam basis. If a sampled seam has more than 2 defects, then the inspection goes back to three good ones in a row before sampling can resume.

Mike W

It sounds like you already have a sampling plan allowed by "the code" -- which I assume is the applicable national welding code. Why change? What do you want to accomplish? Is your customer happy with how you do things now?

BTW, what "the code" allows sounds kinda similar to "PRE control" -- a technique similar to formal SPC but much easier. There are threads on this subject in these forums -- do a search if you are interested in knowing more.

[MikeW]

Mike S

Yes, the change is driven by the customer who believes that the national code is not a statistically valid procedure. Instead, the customer would like a procedure with an AQL not greater than 2.5.

From what I have picked up on AQL, it is easy to determine the number of samples and whether to accept the lot when you can define a product. So, I'm trying to determine what AQL comes from using the code by defining a logical product in a continuous multi-pass weld.

Mike W