Introduction
Among quality management terms, AQL stands for Acceptance Quality Limit, but also referred to as Acceptable
Quality Level. AQL means the poorest level of quality (maximum quantity of defective parts) that is considered acceptable in a particular population
or in a pre-defined sample size. That means AQL is relevant only in case of sampling inspection / test.
AQL tables are quality control tools that were developed based on statistical basis, in order to provide an
easy to use lot-by-lot validation scheme.
ISO’s Technical
Committee ISO/TC 69 (Applications of statistical methods) has created a standard for AQL, under the number of ISO 2859:
ISO 2859-1:1999: Sampling procedures for inspection by attributes -- Part 1: Sampling schemes indexed by
acceptance quality limit (AQL) for lot-by-lot inspection.
Source: qMindset.com
Key Features
The intention of AQL is to give an estimation of a bigger population, by inspecting only a smaller proportion
of the particular population. By other words, the intention of sampling inspection is to find a compromise between quality, safety and
cost-effectiveness.
That’s why statisticians and mathematics experts had developed AQL tables, which provide information
about:
- Sample size: how many pieces should we check from the given lot / population (which represents a safe assumption about the
population).
- Acceptance number: how many pieces are allowed to be defective from the sample, to reach the desired quality limit related to
the whole population.
When starting an AQL-based inspection, you need to select the applicable "inspection level", based on the
features of your inspection. Seven inspection levels exist: three general and four special. Sample sizes increase from level I to III, and
from S1 to S4. Special inspection levels are selected usually in those cases, when the smaller sample sizes are reasonable due to destructive
inspection, or is the small sample size properly represents the bigger population. There are many aspects, that affect the selection of proper
inspection level, e.g.:
- Category of inspection (destructive / non-destructive).
- Complexity of inspected parts.
- Risk of defect pass-through.
- Variation of manufacturing process, that affects inspected characteristic.
- Representativeness: relation of sample size and population.
The following table provides support for choosing the right inspection level:
Aid for choosing proper inspection level. |
Inspection feature |
General inspection levels |
Special inspection levels |
|
I |
II |
III |
S1 |
S2 |
S3 |
S4 |
Category |
Non-destructive inspection |
Destructive inspection |
Complexity of inspected parts |
Low |
High |
Risk tolerated |
High |
Normal |
Low |
High |
Risk of defect slip-through |
High |
Normal |
Low |
High |
Inspected quantity |
~40% of level II (reduced) |
Normal |
~160% of level II (tightened) |
Small |
Variation of process, that affects the tested characteristic |
High (e.g. manual process) |
Low (e.g. automatic) |
Representativeness: relation of sample size and population |
Low |
High |
After you selected the inspection level, table A helps to choose a letter code based on your inspection level
and lot size. This is a master table for normal inspection.
For example: you have chosen General inspection level II (which is the normal basis), and your lot size is 1000. Then, your letter code
will be J, which will give you further inputs later.
Table A: Sample size code letters.
(Source: qMindset.com; ISO 2859-1:1999)
By having the letter code, table B helps to determine the sample size, and the acceptance / rejection rates. Note: more master tables exist for
normal, reduced or tightened inspection.
For example: with letter code J, your sample size will be 80 pieces. Let’s assume you do not intend to
release any lots with more than 0,15% detect ratio. In this case you can only release your lot, if you find zero defective parts among the 80
samples. If only one part fails, you have to reject / revalidate your lot. As we are doing only a sampling test, it might happen by chance that the defective part we found is the overall
defective amount, and the remaining 999 pieces are OK. So AQL can result both type I (rejecting a conforming batch) and type II errors (accepting a non-conforming batch).
Table B: Master table for normal inspection. (Source: qMindset.com; ISO 2859-1:1999; ANSI/ASQ Z1.4)
The allowed defect ratio (in percent) is one of the most important aspects of AQL. It defines how many
defects are allowed in the complete population (see top line of table B). It varies based on the criticality of the inspected characteristic and
the product. The next table shows examples of allowed defect ratios based on criticality:
Relation of allowed defect ratio and criticality. (Important: in case of safety relevant products, 100% testing is necessary, and cannot be exchanged
with AQL-based sampling!) |
Classification |
Low / Mid cost product |
High cost product |
Critical defect |
AQL 0.0 |
AQL 0.0 |
Major defect |
AQL 2.5 |
AQL 1.0 |
Minor defect |
AQL 4.0 |
AQL 2.0 |
There are special procedures to change inspection levels, in other words: switching rules. For example: the
supplier leaves normal (General II) inspection, and implement tightened (General III) inspection, because 2 out of 5 consecutive lots were
rejected. These rules are clearly regulated by the relevant ISO standard.
AQL switching rules (Source: qMindset.com; ISO 2859-1:1999; ANSI/ASQ Z1.4)
Source: qMindset.com; ISO.org; ISO 2859-1:1999; ANSI/ASQ Z1.4
Hints
It is very important to note, that AQL does not mean complete confidence, and its reliability is not 100%.
Although sample based check is cheaper, than 100% inspection, it contains attendant risks. What does this mean mathematically?
If you find 1 defective piece from 20 pieces (sample size), it means your detected failure rate is 5%. It
does not mean, that the 5% failure rate is valid for the whole population. So your population of 2000 pieces may contain much more, than 100 defective elements.
AQL tables were developed on a statistical basis, to give statements about a population with a given confidence level. Unfortunately this confidence level is
definitely not 100%, so you can be only partially sure that your samples represent the whole population. By other words, once out of twenty times, your decision
can be false on the long term:
- You may release a lot, which is not reaching the acceptable quality level (increases your internal or external failure rate, depending on
the intention of the sampling).
- You may reject a lot, which is at or better than the acceptable quality level (increases your scrap cost, or the cost of 100% re-inspection).
Another problem with AQL in the 21st century is that quality expectations are increasing, and the old AQL
is overdue in the times of "zero-defect". The first MIL-STD 105 (military standard) was invented one hundred years ago by the US Department of Defense.
Many inspection plans contain its methodology, and quite many organizations rely on it, however its confidence level is simply not acceptable in various
sectors, which produce with very high quality demands (aerospace, automotive, healthcare, etc.). On the other hand, it is fairly usable in case of
components that bring no risk, and in case of characteristics, that do not affect function.
It is always about the good old contradiction: quality or cost.
Sampling inspection vs 100% check |
|
Reliability |
Direct cost |
Risk of indirect cost |
Sampling inspection (e.g. AQL) |
Low / Mid |
Low |
Can be high |
100% check |
High |
High |
Low |
Source: qMindset.com