DPPM vs. AQL - What is the difference between DPPM and AQL?

Hello Members,

Pls help to advise whether DPPM is equivalent with AQL.
e.g. AQL 0.65 is equivalent 6500 dppm.

Calculation as below:
0.65/100 x 1000000 = 6500dppm

Pls advise.


Regards,
Kelvin

Your math looks good.

On the other hand, the two measures are usually used differently.

DPPM (defective parts per million) is typically used for an actual measure of defect rate in a lot or a process. Or as a prediction of defect rates based on specs, mean, standard deviation, etc.

AQL (acceptable quality level) is "the maximum percent nonconforming that, for purposes of sampling, and be considered satisfactory as a process average" according to Z1.4-1993. Thus it is typically an overall target, not a measure (or prediction) of a specific lot or process.


Tim F

Thanks Tim.

I'm awared on the difinition for these 2 measurement.The question here is does the calculation of 6500 dppm is able to derive from AQL 0.65.


Kelvin

Mr

There answer is no! As Tim stated, AQL is a process average, it does not give a prediction for DPPM for any GIVEN batch.

Base on the MIL 105E, AQL standard focuses on the producer's point. That means it is unlikely that you will reject a lot that has an acceptable level. Take an instance of AQL 1% General Inspection II. It means that you will accept lots with 1% if defectives 95% of the time.

And if you match with the Operating curve. You will alos find that you will be accepting 14% defectives at 5% of the times.

DPPM - defect part per million is different fro AQL.

As you seen AQL is a sampling plan and it has also lower probability of accepting lots that is several time worse than the selected level.

KELVIN said:

Pls help to advise whether DPPM is equivalent with AQL.

Click to expand...

DPPM, DPMO are meaningless, as has been discussed in other threads. I can explain further if you wish. These terms should not be used.

AQL doesn't fare much better:

"W. Edwards Deming observed that the main purpose of MIL-STD-105 was to beat the vendor over the head. "You cannot improve the quality in the process stream using this approach," cautions Don Wheeler, author of Understanding Statistical Process Control (SPC Press, 1992). "Neither can you successfully filter out the bad stuff. About the only place that this procedure will help is in trying to determine which batches have already been screened and which batches are raw, unscreened, run-of-the-mill bad stuff from your supplier. I taught these techniques for years but have repented of this error in judgment. The only appropriate levels of inspection are all or none. Anything else is just playing roulette with the product."

artichoke said:

DPPM, DPMO are meaningless, as has been discussed in other threads. I can explain further if you wish. These terms should not be used.

Click to expand...

I disagree that DPPM and DPMO are meaningless, any more than percentages are meaningless. If you make 25 parts and one is bad, then you have 4% defective, which is the same mathematically as 40,000 DPPM. 4% doesn't imply you made 100 parts and 4 were bad, anymore than 40,000 DPPM implies that you made 1,000,000 parts.

What is meaningless (and I expect this is what you mean) is the extrapolations made from a small number of parts and assuming a particular distribution (i.e. normal distribution). For example "6 sigma = 3.4 DPPM" is meaningless. The sigma level itself is just an estimate based on an observed mean and st dev; the extrapolation assuming a normal distribution is even more suspect!

"About the only place that this procedure will help is in trying to determine which batches have already been screened and which batches are raw, unscreened, run-of-the-mill bad stuff from your supplier. I taught these techniques for years but have repented of this error in judgment. The only appropriate levels of inspection are all or none. Anything else is just playing roulette with the product."

Click to expand...

This seemed self-contradictory for a moment -- giving a use for sampling, but then saying there was no use. Then I realized what Wheeler was saying - sampling is useful for SCREENING, but useless for CONTROLLING or IMPROVING. I seem to be mirroring Wheeler, only a decade later. I got into the theory of sampling but then realized how limited it actually is for detecting small shifts.

Tim F

artichoke (quoting Don Wheeler) said:

The only appropriate levels of inspection are all or none. Anything else is just playing roulette with the product.

Click to expand...

Wow. This seems a bit extreme. So statistics can't help us?

I have been in places where AQL was the norm for sampling plans.
Unfortunately, their interpretation of AQL was "A Quick Look" ;-0

Tim Folkerts said:

I disagree that DPPM and DPMO are meaningless, any more than percentages are meaningless. If you make 25 parts and one is bad, then you have 4% defective, which is the same mathematically as 40,000 DPPM. 4% doesn't imply you made 100 parts and 4 were bad, anymore than 40,000 DPPM implies that you made 1,000,000 parts.

What is meaningless (and I expect this is what you mean) is the extrapolations made from a small number of parts and assuming a particular distribution (i.e. normal distribution). For example "6 sigma = 3.4 DPPM" is meaningless. The sigma level itself is just an estimate based on an observed mean and st dev; the extrapolation assuming a normal distribution is even more suspect!

Click to expand...

You are quite correct ... the latter is that to which I alluded.

Wow. This seems a bit extreme. So statistics can't help us?

Click to expand...

It can but note Tim's observation about the difference between SCREENING and IMPROVING.