Sampling is that part of statistical practice concerned with the selection of individual observations intended to yield some knowledge about a population of concern, especially for the purposes of statistical inference (Wikipedia).
Cooperation during Contract Execution
The cooperation usually concentrates on the following activities:
The chose of evaluation method depends on a variety of factors:
- Evaluation of initial samples of product
- Design information and changes
- Surveillance of supplier quality
- Evaluating delivered product
- 100% inspection
- Sampling inspection
- Identifying inspection – examined to ensure that the supplier sent the correct product – no inspection of characteristics made
- No inspection o Using supplier data (supplier certification)
- Prior quality history on the part and supplier
- Criticality of the part on overall system performance
- Criticality on later manufacturing operations
- Warranty or use history
- Supplier process capability information
- The nature of the manufacturing process
- Product homogeneity
- Availability of required inspection skills and equipment
Action on non-conforming product
Communications to the supplier on nonconformance must include a precise description of the symptoms of the defects. Product acceptance involves the disposition of product based on its quality. This disposition involves several important decisions:
- Conformance – judging whether the product conforms to specification
- Fitness for use – deciding whether nonconforming product is fit for use
- Communication – deciding what to communicate to insiders and outsiders
Inspection planning is the activity of (1) designating the “stations” at which inspection should take place and (2) providing those stations with the means for knowing what to do plus the facilities for doing it.
Locating the Inspection Stations
The most usual locations are:
Increasingly, inspection is built into the process rather than being placed at the end of the process.
- At receipt of goods from suppliers, usually called “incoming inspection” or “supplier inspection”
- Following the setup of a production process to provide added assurance against producing a defective batch. In some cases, this “setup approval” also becomes approval of the batch.
- During the running of a critical or costly operation, usually called “process inspection”
- Prior to delivery of goods from one processing department to another, usually called “lot approval” or “tollgate inspection”
- Prior to shipping completed products to storage or to customers, usually called “finished goods” inspection
- Before performing a costly, irreversible operation, e.g., pouring a melt of steel
- At natural “peepholes” in the process
Automated inspection and testing are widely used to reduce inspection cost, reduce error rates, alleviate personnel shortages, shorten inspection time, avoid inspector monotony, and provide other advantages.
A dramatic example of automated inspection is the concept of “machine vision”. A critical requirement for all automated test equipment is precision measurement.
With the emphasis on defect levels in the parts-per-million range, many industries are increasingly accepting on-machine automated 100% inspection and testing.
Inspection accuracy depends on (1) the completeness of the inspection planning, (2) the bias and precision of the instruments, and (3) the level of human error. Human errors in inspection arise from multiple causes, of which four are most important: technique errors, inadvertent errors, conscious errors, and communication errors.
Measure of Inspector Accuracy
Some companies carry out regular evaluations of inspector accuracy as part of the overall evaluation of inspector performance.
Errors of Measurement
The difference between true value and the measured value can be due to one or more of five sources of variation:
Any statement of bias and repeatability (precision) must be preceded by three conditions:
- Bias – the difference between the observed average of measurements and the reference value
- Repeatability – the variation in measurement obtained with one measurement instrument when used several times by an appraiser while measuring the identical characteristic on the same part
- Reproducibility – the variation in the average of the measurements made by different appraisers using the same measuring instrument when measuring the identical characteristic on the same part
- Stability (or drift) – the total variation in the measurements obtained with a measurement system on the same master or parts when measuring a single characteristic over an extended period of time
- Linearity – the difference in the bias values through the expected operating range of the gauge
- Definition of the test method
- Definition of the system of causes of variability
- Existence of a statistically controlled measurement process
Affect of Measurement Error on Acceptance Decisions
The probability of accepting a nonconforming unit is a function of measurement error. Measurement error can be a serious problem.
Observations from an instrument used to measure a series of different units of product can be viewed as a composite of (1) the varation due to the measuring method and (2) the variation in the product itself. This value can be expressed as:
- sO = Sqrt (s2P + s2E)
sO is the variation of the observed data
s2P is the variation of the product
s2E is the variation of the measuring method
- s is Sigma (or variation)
Solving for P yields
sP = Sqrt (s2O - s2E)
The components of measurement error often focus on repeatability and reproducibility (R&R). Repeatability concerns variation due to measurement gauges and equipment; reproducibility concerns variation due to human “appraisers” who use the gauges and equipment. Studies to estimate these components are often called “gauge R&R studies”.
The ANOVA method is preferred to analyze the averages and ranges.
A common practice is to calculate 5.15s (+/- 2.575s) as the total spread of the measurements that will include 99% of the measurements. If 5.15s is equal to or less than 10% of the specification range for the quality characteristic, the measurement process is viewed as acceptable for that characteristic; if the result is greater than 10%, the measurement process is viewed as unacceptable.
How Much Inspection is Necessary?
The amount of inspection to decide the acceptability of a lot can vary from no inspection to a simple of 100% inspection. The decision is governed mainly by the amount of prior knowledge available as to quality, the homogeneity of the lot, and the allowable degree of risk.
Prior knowledge that is helpful in deciding on the amount of inspection includes:
Competition to reduce costs has resulted in pressures to reduce the amount of inspection. The concept of inspection by the producers (self-inspection) has added to the focus of reducing inspection.
- Previous quality history on the product iteam and the supplier
- Criticality of the item on overall system performance
- Criticality on later manufacturing or service operations
- Warranty or use history
- Process capability information (a 6s process will requirement minimum inspection)
- Measurement capability information
- The nature of the manufacturing process
- Inspection of the first few and the last few items in a production run (usually sufficient)
- Product homogeneity
- Data on process variables and process conditions
- Degree of adherence to the three elements of self-control for the personnel operating the process
Economics of Inspection
We have several alternatives for evaluating lots:
The break-even point for sampling sizes is:
- No inspection – this approach is appropriate if the same lot has already been inspected by qualified laboratories. Prior inspections by qualified production workers have the same effect.
- Small samples – small samples can be adequate if the process is inherently uniform and the order of production can be preserved. Small samples can also be used when the product is homogeneous due to its fluidity (gases, liquids) or to prior mixing operations.
- Large samples – in the absence of prior knowledge, the information about lot quality must be derived solely from sampling, which means random sampling and hence relatively large samples. The actual sample sizes depend on two main variables: (1) the tolerable percentage of defects and (2) the risks that can be accepted. Random sampling is clearly needed in cases where there is no ready access to prior knowledge.
- One hundred percent inspection – This technique is used when the results of sampling show that the level of defects present is too high for the product to go on to the users. In critical cases, added provisions may be needed to guard against inspector fallibility.
Pb = Inspection cost per item
damage cost incurred if a defective units slips through
If the lot quality (p) is less than Pb, the total cost will be lowest with sampling inspection or no inspection. If p is greater than Pb, 100% inspection is best. This principle is often called the Deming kp rule.
For example, a microcomputer device costs $0.50 per unit to inspect. A damage cost of $10.00 is incurred if a defective device is installed in the larger system. Therefore,
Pb = 0.50 / 10.00 = 0.05 = 5.0%
If the percentage defective is expected to be greater than 5%, then 100% inspection should be used. Otherwise, use sampling or no inspection.
Acceptance sampling is the process of evaluating a portion of the product in a lot for the purpose of accepting or rejecting the entire lot. The main advantage of sampling is economy. In addition to this main advantage, there are others:
The disadvantages are sampling risks, greater administrative costs, and less information about the product than provided by 100% inspection.
- The smaller inspection staff is less complex and costly to manage
- There is less damage to the product
- The lot is disposed of in shorter (calendar) time so that scheduling and delivery are improved
- The problem of monotony and inspector error induced by 100% inspection is minimized
- Rejection (rather than sorting) of nonconforming lots tends to dramatize the quality deficiencies and to urge the organization to look for preventative measures
- Proper design of the sampling plan commonly requires study of the actual level of quality required by the user
Acceptance sampling is used when (1) the cost of inspection is high in relation to the damage cost resulting from passing a defective product, (2) 100% inspection is monotonous and causes inspection errors, or (3) the inspection is destructive.
The concept of prevention (using SPC or other techniques) is the foundation for meeting product requirements. Acceptance sampling procedures are important in a program of acceptance control.
Quality Indexes for Acceptance Sampling Plans
There are several published quality indexes:
These indexes apply primarily when the production occurs in a continuing series of lots.
- Acceptable quality level (AQL) – the maximum percent nonconforming (or the maximum number of nonconformities per hundred units) that, for purposes of sampling inspection, can be considered satisfactory as a process average.
- Limiting quality level (LQL) – defines unsatisfactory quality. Lot tolerance percentage defective (LTPD). Because an LQL is an unacceptable level, the probability of acceptance for an LQL lot should be low.
- Indifferenace quality level (IQL) – a quality level somewhere between the AQL and LQL. It is fequenced defined as the quality level that has a probability of acceptance of 0.5 for a given sampling plan.
Types of Sampling Plans
There are two types of sampling plans:
- Attribute plans – a random sample is taken form the lot, and each unit is classified as acceptable or defective. The number defective is then compared with the allowable number stated in the plan, and a decision is made to accept or reject the lot.
- Variables plan – a sample is taken and a measurement of a specified quality characteristic is made on each unit. These measurements are then summarized into a sample statistic (e.g., sample average) and the observed value is compared with an allowable value defined in the plan. A decision is then made to accept or reject the lot.
Single, Double and Multiple Sampling
In single-sampling plans, a random sample of n items is drawn from the lot. If the number of defectives is less than or equal to the acceptance number (c), the lot is accepted. Otherwise, the lot is rejected.
In double sampling plans, a smaller initial sample is usually drawn, and a decision to accept or reject is reached on the basis of this smaller first sample if the number of defectives is quite large or quite small. A second sample is taken if the results of the first are not decisive. Because it is necessary to draw and inspect the second sample only in borderline cases, the average number of pieces inspected per lot is generally smaller in double sampling.
In multiple sampling plans, one, two or several still smaller samples are taken, usually continuing as needed until a decision to accept or reject is reached.
ANSI/ASQC Z1.4 (1993) is an attributes sampling system. Its quality index is the acceptable quality level (AQL). The AQL is the maximum percentage nonconforming (or the max number of nonconformities per 100 units) that, for purposes of sampling inspection, can be considered satisfactory as a process average. The probability of accepting material of AQL quality is always high but not exactly the same for all plans.
Disposition of Nonconforming Product
If an inspector finds that a lot of product is nonconforming, the lot is marked “hold” and is often sent to a special holding area to reduce the risk of mixups.
Decision Not to Ship
The investigation may conclude that the lot should not be shipped as is.
Decision to Ship
This decision may come about in several ways:
- Waiver by the designer
- Wavier by the customer
- Waiver by the quality department – may make fitness for use decisions on noncritical matters. For minor categories of seriousness, the delegation may even be made by the quality engineers or by inspection supervisors.
- Waiver by a formal material review board – formal decision and documentation process
- Waiver by upper managers – restricted to cases of a critical nature involving risks to human safety, marketability, or risk of loss of large sums of money.