ISO 9001
section 8.2.4 says that the organization must monitor and measure the
characteristics of the product in order to verify that the product meets
requirements, but is silent on the methods to be applied to monitor and measure
product characteristics. The choice of
inspection methods is left to the organization.
Acceptance sampling is one such method. Since it is not practical to inspect every
item in a large batch, acceptance sampling allows you to infer batch quality by
examining a random sample from the batch.
Statistically designed acceptance sampling plans, as a method of
measuring product conformity, have been around since at least World War II. Sampling plans like MIL-STD-105, authored by
Harold F Dodge and others, have been in use for over 60 years.
Because not all items in the batch are examined, there are
risks associated with statistical acceptance sampling. Two risks are typically calculated:
·
The Acceptable Quality Level (AQL), also known
as producer’s risk, is the percent defective that is likely to be accepted 95 %
of the time. There is a 5% chance that
product of higher quality than the AQL will be rejected by the sampling plan.
·
The Lot Tolerance Percent Defective (LTPD), also
known as consumer’s risk, is the percent defective that is likely to be
accepted 10% of the time. There is a 10%
chance product as defective as the LTPD will be accepted by the sampling plan.
Each sampling plan has an associated operating characteristic curve (OC) which describes the probability of lot acceptance as a function of the lot’s percent defective. The AQL and LTPD represent two points on the OC curve. As the lot percent defective increases, the probability of accepting the lot based on the sampling plan that the OC curve represents decreases.
In addition to the risks associated with acceptance sampling plans, there are some practical disadvantages:
·
While acceptance sampling greatly reduces the
number of items inspected, other sampling methods such as statistical process
control reduce inspection even further and provide process control feedback.
·
When a lot is rejected, we will know why the lot
is rejected, but we will not know the root cause of the defect. We will only know that the product is
defective. There is no control mechanism
that will help us control the process the product comes from.
·
Acceptance sampling assumes random sampling, but
in most cases the sample is stratified because the product is normally stored
in boxes. As such, it is possible that
some product will never have a chance of being sampled and the sample will not
be random.
·
Sampling plans based on AQL, LTPD, or AOQL
(Average Outgoing Quality Limit) assume that a certain amount of defective
material is acceptable. This sends a
message to employees and suppliers that some level of defectiveness is
acceptable. This is not the best message
to send if the organization is trying to be a best in class producer.
Sampling plans can be designed by users, or selected from
standards such as MIL-STD-105, now obsolete, or ANSI
Z1.4 which implements the same plans. A common pitfall suffered by users not
familiar with sampling plan design is to design constant percentage sampling
plans. Avoid constant percentage
sampling plans (a fixed percentage of the lot is sampled regardless of lot
size). For small lots, constant
percentage sampling plans may not afford enough protection. For large lots, an excessive amount of inspection
will usually result, and the sampling plan will be over critical.
Consider a sampling plan that samples 10% of a lot. For a 50 piece lot, a 5 piece sample will
result in approximately a 1% AQL, but a 46% LTPD (a very weak sampling plan). For a 5000 piece lot, a 500 piece sample will
result in an AQL of .001% and the LTPD will be .46% - a plan unlikely to accept
any lot.
While there are better ways to control production processes
than acceptance sampling, acceptance sampling can be an effective method for a
customer to protect itself from accepting defective purchased product. Since the customer has no control over the
manufacturing process, it is not important for it to understand what process
variable caused defective material. It
need only know that the product is defective.
The customer is susceptible to stratified sampling, but accepts this
disadvantage in favor of inspecting a small percentage of the entire batch.
No comments:
Post a Comment