Measurement uncertainties and tolerances

Measurement results are used to make decisions.

The result of a measurement can generally be considered as technical information communicated to a user. In the light of this information, the user will be asked to make decisions:

• product acceptance (when measuring characteristics or performance to establish conformity with a specification);

• process validation ;

• adjustment of a parameter as part of the control of a manufacturing process (servo-control) ;

• validation of a hypothesis as part of a development ;

• environmental protection ;

• defining the safety conditions of a product or system ;

• medical diagnosis.

All these decisions contribute to the quality of the product or service.

The validity and relevance of the decisions taken therefore depend directly on the quality of the information communicated, and therefore on the measurement results.

How to qualify the quality of a measurement result: its uncertainty and relevance?

• Uncertainty: a quantitative indication of the quality of a measurement result.

  The uncertainty associated with a measurement result provides a quantitative indication of the quality of that result. This information is essential to enable those who will use the result to estimate its reliability.

  Without uncertainty, measurement results can no longer be compared:

  • or between them;

  • or in relation to reference values specified in a standard or specification.

Based on this information, the company makes decisions: acceptance or rejection of a research and development hypothesis, control of a manufacturing process, acceptance or rejection of a product.

• The relevance of a measurement result: more often than not, there are situations where a measurement result can be completely accurate and completely useless. In this case, the information communicated to the user is correct but useless. To resolve this difficulty, a technical dialogue needs to be established with the metrologist, and the user needs to accurately describe the information actually required, rather than trying to solve it himself. This type of problem is closely related to issues of test representativeness and, in some cases, sampling. We'll look later at the care that needs to be taken in defining the measurand...