In the last 5 years, there is not much talk about “Acceptance Sampling”. Very few articles in this subject are published in journals. On the other hand, SPC techniques are discussed in every Quality Magazine, many published papers and in conferences. Is there any Correlation to these two trends? :confused:

Has the development and rapid recognition of Statistical Process Control techniques have made the "Acceptance Sampling" almost obsolete?
Govind.

In the last 5 years, there is not much talk about “Acceptance Sampling”. Very few articles in this subject are published in journals. On the other hand, SPC techniques are discussed in every Quality Magazine, many published papers and in conferences. Is there any Correlation to these two trends? :confused:

Has the development and rapid recognition of Statistical Process Control techniques have made the "Acceptance Sampling" almost obsolete?
Govind.
I don't know if this is the definitive answer, but here is my theory at least.

Acceptance Sampling is primarily a go-no go test. Let me use an arbitrary, simplistic example for sake of discussion. If the bolt is longer than 3.5 mm it is bad, if it is shorter than 3.5 mm it is good. Granted, with these sort of tests I can make operator aids (a gage) or even automate the test. But acceptance sampling doesn't provide much information, which has costs.

The first cost is since the data are binary, I must sample a lot more items. If I were actually measuring the length of the bolt, I may be able to generate equal confidence levels with a lot less data.

The second cost is the go-no go nature of the specification. A bolt that is 3.49999 mm long is "good", 3.500001 is "bad". Yet both bolts are likely indistinguishable in use. The "true losses" of a bolt that is not the proper length is not a step function, but likely a parabola. This opens up the topic of the Taguchi loss function, which I will not go into detail on this post (but will if requested).

The third cost is that in attribute sampling I don't keep the time sequence of production. I simply sample 59 bolts from the bin. I could keep track of time sequence, but short of using logistic regression on this binary data, it is very difficult to trend versus time.

The use of SPC allows me to overcome much of the three problems stated above. I can determine if my production is "capable" of meeting the spec with a smaller sample. I can work to reduce variation, which would allow parts to be more of a "snap fit" and gain other savings over parts which are right at the specification limit. And I can trend versus time.

Now I have no personal knowledge if indeed "acceptance sampling" is "out" and SPC is "in". But hopefully this contributes some ideas about the differences between the two.

Steve Prevette
ASQ CQE
Fluor Hanford
City University

Steve,
Thanks for your views. I actually posted the same question in the inside Quality discussion forum. Here are the other responses:
http://www.insidequality.com/index.v3page?d_v=rm&d_mid=68919&p=9310

My theory for this trend is:
Many Quality practitioners see SPC as a real time proactive approach to process quality management and acceptance sampling as a reactive approach to product quality management. Iam not supporting this theory. There are advanced Sampling techniques that are proactive. They are published but not necessarily enter industry.

Did your reply take into consideration the "variable" Sampling as well?
Thanks,
Govind.

I see acceptance sampling as "insurance" - paying a little to mitigate major problems.

If the situation is "healthy" - the supplier's processes are in control and producing items to spec - then the shipments coming in will automatically be fit for use. Acceptance ampling is actually counterproductive here! The money spent on testing will never find a bad lot (because there are none!) but it will occasionally reject good lots, adding costs for both consumer and producer. It is a waste of money. Just like health insurance is a waste of money as long as you stay healthy!

If the situation is "ailing" - the processes aren't in control and the shipments have varying quality - then you will get good shipments and bad shipments. Now the "insurance" is vital. Receiving an "unhealthy" shipment will cost you money and time, but it will cost a lot less than paying the price down the road for the bad situation.

I suppose you could equate the deductable to the style of sampling you do. A high deductable is like a relatively weak sampling plan (like "reduced inspection" or "Level I" in Z1.4 language). With a high deductable (weak sampling plan), the premium (testing expense) is low, but each accident (bad shipment) results in higher out-of-pocket expenses.

Just like with insurance, the amount & type of insurance depends on your lifestyle and willingness to take risks. If your business is "young and carefree", you might live a little dangerously and skip the insurance. Similarly, if you are in a "trusting, monogamous relationship" with a supplier, then you may now need as much insurance. If you take up with a new, unknown salesman every month, you might just want a little more "health" insurance. http://elsmar.com/Forums/images/smilies/wink.gif

I suppose we couls say that SPC is like getting yearly (or in this case daily or hourly) physicals. If you check your "vital signs" on a regular basis, then your know how your health is doing and you can head off problems before they start.


OK, OK. I probably pushed the analogy a bit far, but I think it works pretty well. Preventive maintainence keeps you healthy, but a little insurance may still be a wise investment.

Tim F

Did your reply take into consideration the "variable" Sampling as well?
Thanks,
Govind.
Yes. The basic statistical principle is that there is more information within a datum if it is actually recorded as a number, its "actual" value. You lose a lot of information (and statistical power) when you only record a result as go - no go. And yes, I can do SPC on go - no go data, but you will find the chart will tell you a lot less than when you record the data values. And for those that follow six sigma and/or want to know process capability, you have to have the data values for that.

Just a thought, but I have seen the deployment of the automotive APQP process (PPAPs, MSA, etc.) virtually eliminate acceptance sampling in my company. In fact, we have eliminated incoming inspection.

If there is an issue on the floor, the supplier must pay a penalty for lost production and pay for a 3rd party to sort 100%.

Just a thought, but I have seen the deployment of the automotive APQP process (PPAPs, MSA, etc.) virtually eliminate acceptance sampling in my company. In fact, we have eliminated incoming inspection.

If there is an issue on the floor, the supplier must pay a penalty for lost production and pay for a 3rd party to sort 100%.

Ben,
Any specific reason why the acceptance sampling was eliminated?

To cut down inspection Cost and or The process is highly Capable?

If you had an issue on the floor, is it not too late to discover and correct the problem? Supplier covering cost is a different subject though.
You would still miss your customer delivery.

I would follow the Sampling Switching rules rather than eliminating the Sampling. How is your experience so far after elimination?

Regards,
Govind.

OK, OK. I probably pushed the analogy a bit far, but I think it works pretty well. Preventive maintainence keeps you healthy, but a little insurance may still be a wise investment.

Tim F
Tim,
I think the analogy is appropriate and explained the benefits of Sampling and SPC in real life example. :thanx: Thanks for that insight.

I found this interesting article that talks about combining the strength of SPC and Sampling (SQC). Let me share with the forum:

Integrating SPC and SQC to Overcome Weakness in Either
http://www.isixsigma.com/library/content/c040202a.asp

Regards,
Govind.

Another article about SPC and Aceptance Sampling.

http://www.variation.com/techlib/as-9.html

Here is a non statistical based answer - sorry.



Acceptance sampling starts from a desire to avoid the cost and pain of 100% inspection. It is a risk and cost based decision to find "most" of the bad ones.



The base assumption is old school (think QC, MIL STD, inspect the quality in).



SPC (as Steve has said so well starts) with a recognition that all processes have variation. If I choose to, I can roll up my sleeves and use SPC to improve my process to any capability I choose.



Both Acceptance sampling and SPC still leave bad parts as a possibility. Poka Yoke offers the impossibility of zero defects. It is really hard to do well. It is harder to convince people (especially accountants) that it can work. I have seen only 2 perfect poka yokes. Both tested in excess of 1 million parts with zero defects for the feature in question. I have seen dozens of ineffective poka yokes.



We use currently all three tools here.

I will watch this topic with interest since I now have to come up with an improved sampling scheme for x-ray inspection.

I sent this discussion link to a reputed academic specialized in Acceptance Sampling to hear his views.
He provided me a reply the very next day and has graciously allowed me to share his view with the “Covers”. :thanks:
Regards,
Govind.

1. Integration is the key. In the broader field of statistics of quality,
experimentation, SPC, and sampling inspection have their own individual and
joint roles to play. See recent ISO TC69/SC5 activities and US related
activities at http://www.variation.com/index.html .

2. The scope of sampling inspection is much broader, not limited to parts
manufacturing and inspection. This includes inspection of (i) any end
items, (ii) components, (iii) raw materials, (iv) operations, (v) materials
in process, (v) supplies in storage, (vi) maintenance operations, (vii) data
or records and (viii) administrative procedures. There are also mandatory
pre-shipment and receiving inspections, particularly with regard to imported
and exported goods.

3. There are historical reasons why it has become less important to resort
to acceptance sampling when process non-conformances are in the ppm levels
in some manufacturing industries. There are also various non-industrial
products, particularly agricultural and fisheries products, where variation
is very high. A 10% non-conformance is not uncommon in agricultural
products.

4. Acceptance sampling area originally evolved to provide "consumer"
protection. This is still important at the societal level, or when
contractual shipments are made. There will always be safety and compliance
issues in various products and services. The originator of acceptance
sampling Prof. HF Dodge himself made the remark "you cannot inspect quality
into a product". Acceptance sampling was never intended to replace the role
of SPC for quality improvement. The fault was with its wrong use at the
shop-floor rather than with the field of acceptance sampling.