Accepting a Measurement System for Its Intended Use

Why it matters

A decision has to be made at the end of a measurement system study. But acceptance or rejection does not mean declaring the instrument good or bad for all jobs. The same setup can be useful for monitoring large process changes while falling short for a final product decision with a narrow tolerance. A percentage calculated without knowing the intended use is incomplete information. For this reason the subject of the decision and the size of the important change are defined first. Then resolution, systematic bias, repeat variability, observer effect and stability over time are evaluated together. One being good does not make up for the shortcoming of another. The decision record must clearly state which use is approved and outside which conditions re-examination is required.

The ratio of the measurement standard deviation to the total standard deviation and the variance contribution are different quantities. The standard deviation is in the unit of the measured characteristic; the variance is in the unit squared. When the ratio of the same two standard deviations is written as a percentage it gives one result, and when the ratio of their squares is written it gives another. A variance contribution that looks small is not equally small on the linear scale. Moreover, systematic bias is a concept separate from these ratios. A shift of all readings in the mean does not go away because the repeat scatter is small. Writing only a bias percentage in the report can confuse these three meanings. If the name, numerator, denominator and unit of each indicator are stated, the acceptance decision becomes checkable.

Ratios that use the total variability as the denominator are sensitive to the study group. When parts are selected from a wider range, the contribution percentage may fall even if the measurement setup does not change. This result does not show that a more precise instrument has been obtained. A comparison with the tolerance width asks another question: how much of the decision interval does a given spread of the measurement cover? The definition of spread used here must also be clear. A six-standard-deviation width is a descriptive comparison; it is not the definite limit of all errors or a full measurement uncertainty. The distribution assumption, the bias and other sources of uncertainty are evaluated separately. For a result near the product limit, knowing only this width does not produce a reliable acceptance decision.

Diagram: [Intended use] → [evidence of resolution + bias + variability + stability] → [testing all acceptance conditions together] → [decision and monitoring with a defined scope]. The plus signs show that the evidence is sought together; they are not a numerical sum of indicators in different units. Acceptance criteria must be defined by the consequence of the use and the organization's verified control plan. The numerical thresholds in this lesson are original training decisions; they are not universal rules applied to every industry. A wrong acceptance with a severe consequence and a reversible monitoring warning may not ask for the same level of evidence. If data are missing, conditional or narrow-scope use can be considered, but the shortcoming cannot be hidden inside the approval sentence. Which decision will be postponed, which additional control will be applied and who will evaluate new evidence must be defined. In this way the measurement analysis does not stop at producing a report percentage; it defines the real limit of use and the re-verification requirement. A new product, fixture or environment change can reopen this limit. All numerical scenarios in this lesson's short quiz are also covered by the following notice: This dataset was generated synthetically for training; it does not represent a real business, employee or customer study.

Formula