Method selection to evaluate measurement uncertainty in microflow applications

Abstract

Abstract Microflow applications are gaining increasing importance in medicine, and are indeed crucial in areas such as neonatology and oncology. However, in terms of uncertainty evaluation the very small levels of flow entail problems characterized by a “close to the physical limit” situation. The same happens in nanovolume applications. In these cases the Law of Propagation of Uncertainty framework (GUM) may not to be adequate, since the interval of coverage probability will, in some circumstances, contain negative values, which is physically difficult to explain. Monte Carlo methods may be a good alternative when there are not many negative readings in the application (limit-of-detection problem), but may also prove inadequate otherwise, since the required functional relationship of the model will produce a peak of values centred at the zero value, which is also not very credible to describe satisfactorily the real situation. In the latter, the application of a Bayesian method will be the better approach to deal with the evaluation of measurement uncertainties. The whole process of the revision of the international metrological standardization lies in this principle, of selecting the best method to tackle the particular problem in hand, and not to apply blind recipes to the evaluation of uncertainty in measurement, and there is a new EMPIR project focused on providing examples to BIPM documents to strengthen this concept.