Les incertitudes de mesure dans le cadre de l'assurance qualité

Measurement uncertainties in quality assurance

CRECEP, Ville de Paris, 144 av. Paul Vaillant Couturier, 75014 Paris

Résumé

La pratique courante des laboratoires pour estimer l'incertitude est d'évaluer la confiance en un résultat et sa comparabilité à travers des paramètres de caractérisation de la méthode analytique globale : linéarité, limite de détection, taux de recouvrement, répétabilité, reproductibilité, justesse... Dans un domaine dominé par une approche expérimentale, il est important de montrer comment les essais effectués selon la norme T90-210, utilisée comme support de caractérisation, peuvent former la base d'une estimation globale de l'incertitude et d'informations utilisables en contrôle qualité interne en accord avec les demandes de la nouvelle norme NF EN ISO 17025 appliquée pour l'accréditation des laboratoires d'analyse.

Abstract

Uncertainties of measurements play a key role in the data that are achevied under quality assurance insofar as they associate ideas of need adequation, quality assurance, and internal quality control. The application field of these three concepts is often difficult to charaterize in the laboratories. Actually, the method to get various information is not always clearly defined, and the obtained data can be redundant. The measurement uncertainty correspond to the interval on the scale of extent in which the true value is with a given probability and when all the error sources have been taken into account. This very theoretical definition is more easily expressed as a parameter correlated with a result of measurement, which defines the dispersion of the values what can rationaly be allotted to the value. The validity of the uncertainties of measurements depends on two postulates: on the one hand the determination must be performed with a validated method, and on the other hand, the analysed materials must be inside the valid scope. If these criteria are not filled, a significant uncertainty should not be associated with a measurement. The measurements uncertainties include several components. Some of these components can be evaluated from the statistical distributions of the results of a series of measurements and can be characterized by experimental standard deviations. Others components, that can also be characterized by standard deviations, are evaluated by probability distributions based on experiments or other information sources. One of the main problems regarding of the laboratories is related to the absence of control of factors that may interact with tthe complete process sample, and especially on its sampling and partition into aliquots. The error related to these stages is, in most cases, much higher than uncertainties due to the analysis, and is not taken into account in calculations.

After the aspects of the problem of uncertainties of measuresas been set in the context of quality assurance and in the context of the new standard NF EN ISO 17025, this work brings up the various approaches of the calculation of uncertainties in an environmental laboratory and their application with a concret example. According to whether we consider a standard uncertainties associated with a given test, or a total uncertainty based on a unit of tests, time spent on the different phases of the analysis, and oncalculation, will be greatly variable. The choice of determination of a total uncertainty presents some disadvantages, especially when a modification is made on one of the stage of the analytical technique, but it allows the use of the statistical data already available through the files of validation of the methods, and the interlaboratory tests. Indeed, when for a given sample, a true value can be used, the error on the result of measurement can be easily calculated from the variances of repeatability and reproducibility, associated with the bias of the method. This uncertainty can then be expressed as a simple form of which means a parameter (standard uncertainty or wide uncertainty), defining the maximum tolerable limit on the result. In ultimate stage, the internal quality control allows to be sure that fixed uncertainties are respected on each particular analysis.