The statistical properties of the number of observations used for proving meters in dynamic flow measurements

Abstract

Quantitative evidence is presented to show that the present practice in meter proving, of taking proof runs until two consecutive runs check each other within a pre-established variation, may be unreliable. It is shown that, if the number of consecutive runs required is increased from two to five, say, then the precision of the meter factor is increased and the probability of missing an initial transient is significantly decreased. The 'cost' of the alternative procedure, as measured by the average number of observations required, is shown to be small.