AQL sampling plan

AQL sampling plan LTPD sampling plan AOQL sampling plan... 

A detailed discussion of all three classes of sampling plans is beyond the scope of this book. Our discussion will be confined to the AQL sampling plan. 

AQL sampling plans are the most widely accepted and used plans throughout the industry and government. The best-known published source of such AQL sampling plans has been the government publication titled Sampling Procedures and Tables for Inspection by Attributes MIL-STD-105 E. Since 1995 ANSI/ASQC Zl-4 has replaced MIL-STD-105E with minor modifications. 

When using the defect-detection approach to quality control certain problems develop. Inspection does nothing to improve the process and is not very good at sorting good-from-bad. Also, sampling plans developed to support an acceptable quality level (AQL) of 5%, for example, say that a company is content to deliver or reject 5% defects. 

Table 3.2 shows a summary of single sampling plans for normal inspection (note not all sample size code letters or AQLs are shown - please refer to ISO 2859-1 for a complete set of sampling plans). The sample code tells you the... 

Table 3.3 shows the operating characteristics for a normal inspection level sampling plan with a sample size (i.e. number of items examined) of n = 200 at three different AQLs. This table shows that if a product has 3% nonconforming items (p) then, using an AQL of 2.5%, approximately 95% of lots would be expected to be accepted. Operating characteristics curves are given in ISO 2859-1. 

Sampling Procedures for Inspection by Attributes - Part 5 System of Sequential Sampling Plans Indexed by Acceptance Quality Limit (AQL) for Lot-by-Lot Inspection , ISO 2859-5 2005, International Organization for Standardization (ISO), Geneva, Switzerland, 2005. 

Seal strength Below 5% defective at worst-case conditions for critical inputs Below 1% defective under normal conditions Perform hourly inspections using variables sampling plan with AQL = 1.0%... 

For example, a 95% probability that a batch is as good as the AQL does not exclude other probabilities that the batch is worse (for example the 10% probability level, sometimes called the lot tolerance per cent defective (LTPD) or unacceptable quality level, can be several percentage points worse than the AQL). It is the relationship of AQL to LTPD that more closely identifies the quality risk in any statistical sampling plan, and illustrates why the use of national plans as simple accept and reject figures fails to tap their full potential. 

If we now combine these points, in a comparison of some simple sampling plans, where the batch size is ignored and the AQL is considered to be constant, then it is clear that sampling risk is directly dependent on sample size (Table 4.5). (The theoretical plans used show that, depending on the acceptance of special risks , a batch of 20,000 items could be inspected against a 4% AQL with a single sample of from 13 items to 315 items.)... 

One further point in dealing with the size of sample is the choice between single, double and multiple samples. Plans have been produced which allow comparison between these options at the same AQL, and using the same example as above (i.e. a batch of 20,000, AQL 4%) (Table 4.6). 

It is worth stressing that the AQL is simply the defect percentage level which defines the 95% probability of acceptance. From the example it is obvious that there are other probabilities which can be calculated and, if graphed, provide the operating characteristic (OC) of that sampling plan, i.e. 

All sampling plans have an OC in which no matter how low the defect level is at the theoretical AQL, there will be a risk of accepting a batch with a considerably higher level of defectives. 

The sampling plan cannot be adequately defined by reference to just one point on the curve (i.e. the AQL), but two points will define the plan. 

All sampling plans have OC curves, and any consideration of a scheme for sampling inspection should begin with an examination of the curve (see MIL-STD-105, or BS 6001). The ambition in examining the curves is to obtain the smallest sample size that provides the greatest confidence of batch compliance with the required quality, i.e. reducing the degree to which the batch may be different from the chosen AQL. 

ISO 2859-1. 1989. Specification for sampling plans indexed b acceptable quality level (AQL) for lot-by-lot inspection. 

Let s look at right quality the level the customer wants or better if possible. Most customers want zero defects. The traditional method was to use a random sample and develop an acceptable quality level (AQL). A sampling plan was developed and then the sampling standard tables were looked at to determine if the lot should be accepted or rejected. Using the acceptable quality level indicated that the customer was willing to accept some defects. In most cases today this method is not used. With the advent of Six Sigma the levels of acceptability have come way down, approaching zero defects. 

The Acceptable Quality Level (AQL) is the maximum percentage of defectives or the maximum deviation from label claim that is acceptable as a long-term average. It is the poorest quality level for the supplier s process that a consumer would consider to be acceptable as a process average. AQL is a property of the supplier s manufacturing process, not a property of the sampling plan. 

A sampling plan for attributes is a method to overcome this problem. An example of such a plan is the Accepted Quality Level system (AQL) (See Sect. 24.5.4. for a more complete description of AQL and Sect. 20.4.5 for a statistic background). In order for the AQL system to be successful an extensive and statistically planned random sample has to be selected. The defects that are found are classified into levels, for example critical, major, minor. Within each level the system defines an acceptable quality level . When a quality level is exceeded, then a batch should be rejected. This method of testing requires time and expertise. Sampling has to be performed from a large number of containers from the same batch. Within the pharmaceutical industry it is often necessary for such tests to be carried out by the container manufacturer. For smaller enterprises such as pharmacies quality control can be undertaken by the wholesaler or an independent laboratory. Within a (hospital) pharmacy the quality control is often limited to a visual comparison to reference samples and a check of the presence of the supplier s statement that the containers comply with the agreed specifications [46]. 

From the information regarding the inspection level, sample size code letter, sampling plan, and specified AQL value, determine the acceptance and rejection number. If the AQL value is unspecified, start with a 2.5 percent defective AQL value. 

Using Table 16-3 (single sampling plan for normal inspection), we find that for code letter J and AQL of 1.0, the sample size is 80 with acceptance number AC = 2 and rejection number RE = 3. Thus, the course of action should be accept—if the number of defectives is 2 or less—or reject—if the number of defectives is 3 or more. 

The statistical relevance of the action plan should be included for cross-referencing between organisations, i.e. that the plan is based on MIL-STD-105E, Inspection Level II, Normal Inspection, Table IIA AQL 1% (single sample). 

Most countries now have national sampling and inspection tables which relate batch size to sample size and, by using precalculated accept/reject figures for a range of preferred AQLs, allow acceptance of a batch at 90/95% probability levels. In the USA these plans are identified as MIL-STD-105, in the UK as BS 6001 and internationally as ISO 2859. 

The system of statistical end control, as mentioned in the previous section is called Acceptance Sampling. Elements are diverse parameters such as AQL, Producer s Risk, LQL, Consumer s Risk, the method of inspection or analytical procedure and the OC curve derived from them. Typically a plan contains not only the maximum and minimum limits of the content of product or batches but also the relative frequencies by which the outcome on content may be passed (or not) and what should be done accept or reject. 

An AQL of 1 % defects in the population has been chosen in this example. The producer s risk is then extremely high, namely about 25 % for a sample size n = 30 and will increase even more for larger sample sizes. Apparently it is not possible to obtain the desired acceptance plan by manipulating the sample size. The following example shows how to improve the characteristic and get a better result by changing the acceptance criterion as well. 

---