AQL (Acceptable Quality Level) vs. LTPD (Lot Tolerance Percent Defective)

Yea, I do tend to ramble sometimes. Especially after a full day in meetings with those whose vision, to put politely, is myopic. I have GOT to learn to calm down before moving on to other topics when this happens. But anyway, it was late and I was tired.

I did not mean to imply that AS was an unacceptable method of operation, only that those who use it to "justify" producing a level of nonconforming product were shortsighted. The idea of a renegotiated AQL is, I agree, a good one, but only if applied. The phrase "but, it has always been this way" is another way of saying, "we do not want to change."

The following is a true story from my (shaded?) past.

The company in question (name withheld) produced a semiconductor component from raw material (wafers) supplied from different vendors. The raw material was supplied with chemical composition sheets that provided the basis as to which wafer would be applied to which product line. The fabrication engineer would examine the data sheets and select the wafer(s) he felt was right for the job that was to be processed into the components. After fabrication, each component was tested prior to being processed into the final package. At this in-process stage, there would be 40% to 50% of the components that would not function as expected in the final form. These would be put aside with the standard disclaimer "well someone may be able to use them sometime." When I was first asked to investigate, there were over 7500 of these "leftovers" that had collected (that I was told about). Many more had been discarded over the years. I asked the QM the obvious question, "How can you continue to function with this amount of failures?" The answer was (of course) "It has always been this way. It cannot get any better." After collection and examination of the raw data supplied by the vendor and the results of the in-process testing, I found a direct relationship (correlation by multiple regression with r=0.87) between the vendor's data and the in-process test results. Using the regression techniques, I developed an equation that could be used to purchase the raw material on the basis of the parameters expected of the final product. Using this equation, I selected several available wafers for a trial run. This trial run yielded approximately 85% acceptable (obviously I had not found the only source of variation, just an apparent one). After implementing the equation in the selection process, computer simulations indicated the company in question would save over $550,000 per year. The General Manager and QM were pleased with my results and final report, but obviously never implemented this system or try to find the additional sources of variation (There were a LOT. My preliminary count yielded over 20 key input variables). The company no longer exists and 58 souls became unemployed.

The idea of continuous improvement is just that, continuous. Renegotiated AQL's are a way, just as continued monitoring and improvement of Cpk (or whatever) is. I also become discontent (to put it politely) when someone becomes "satisfied with a certain level of quality." When dealing with these types and the myopic ones mentioned above, I find the following helpful:

"The trouble with people is not that they don't know, but that they know so much that ain't so." Josh Billings

barb, I have the book mentioned above. It is an excellent read. There is also a lot on the web about learned ignorance and Dr. Semmelweis if you need additional info.

Also see: AQL (Acceptable Quality Level) vs. PPM (Parts per Million) - Equivalences?

I note that this thread started about four years ago. I think there is much good and bad philosophy above so I thought I would add my rant to it. :thedeal:

It is important to understand that acceptance sampling and process improvement have different purposes but both are essential business practices. Acceptance sampling ensures that, if the process does manufacture off-grade lots, they will not go to the consumer.

Lyndon Diong asked on 9/29/1998:
I favor LTPD rather than AQL. But some people disagree. Could someone let me know the practical advantage of AQL?

Click to expand...

I think that you should never choose between AQL and LTPD. You choose a sampling plan (or rather you design a sampling plan). The plan that you choose will have both an AQL and an LTPD. Out of habit, I will use the term RQL (rejectable Quality Level) in place of the equivalent term: LTPD.

You asked about the practical advantage of AQL. But first I will address what AQL means statistically. I will ignore the dumbed down definitions in Military and Z1.4 standards.

I am assuming that you refer to attribute sampling. The four statistical parameters are the decimal fractions AQL, Alpha=producer's risk, RQL, and Beta=consumer's risk. A manufactured lot that has fraction defective = AQL, if submitted to the sampling plan, will have a high probability of passing. The probability is (1-alpha). If alpha=0.05 (fraction) then the probability of accepting it is 1.0-0.05=0.95. (95%) This does not mean that any actual lot sampled will have a fraction defective equal to AQL.

Now to your question. The practical advantage, or rather the use of AQL, depends on whether you are the producer or consumer. (using those terms in a general sense) If you are the producer, one of your objectives would be to keep the cost down. By reducing AQL you can reduce the sample size and thus lower inspection cost. (assuming that the other three statistical parameters of the plan do not change.)

The limiting factor in reducing the specified AQL is the fact that you probability have a process that produces lots with p' fraction defective. If your current process quality is saleable, it would be reasonable to set AQL equal to p' or a little higher to assure passing.

Don't fall for the myth that setting AQL lower than process capability will improve quality. It will only cause the sampling plan to reject some lots that are probably not any worse than the lots that pass the plan.

In order to use acceptance sampling competently, you need to understand, not only how to set AQL as described, but how to detect the misconceptions about AQL that have been written by otherwise intelligent quality engineers and statisticians.

As a practical matter, if you are a customer don't let the producer trick you into negotiating AQL levels. The customer should specify the quality level that he/she does not want -- which is RQL. And if you are a producer, your sampling plan's RQL should be consistent with the customer's RQL. This is explained thoroughly here:

www.samplingplans.com/aqlprimer.htm

If you are dealing with other people about sampling criteria, don't automatically assume that they understand what you just read.

Your question was about AQL so I did not say much about LTPD=RQL. Suffice it to say here that a lot that is RQL fraction defective will usually not pass the sampling plan and go to the customer.

www.samplingplans.com/modern3.htm#EVALUATE

Sincerely, Stan Hilliard