Comparison of different approaches for detection and treatment of outliers in meter proving factors determination

A meter proving factor can be considered as a calibration parameter, by expressing the ratio the reference volume and the gross volume of liquid passed through a meter. The international guideline recommends Dixon's test for outliers to a meter proving factor set. However, the literature says that this statistic test is restricted only to data with Gaussian behavior, besides of not to be able to detect and treat two outliers at the same time. Here, Gaussian behavior of the meter proving factor set is evaluated, then different parametric and nonparametric approaches for detection and treating outliers applied to turbine meter proving factors for custody transfer of liquefied petroleum gas are compared. Afterwards, this effect is evaluated in relation to the number of outliers and how this handling affects the variable range criteria for expanded uncertainty in average meter proving factor. The results show that different average meter factors can be reached for each nonparametric and parametric test; anyway, no statistically significant effect between them is noticed. (C) 2016 Elsevier Ltd. All rights reserved.