Correlated noise and prior models in electromagnetic flow tomography

Abstract

Electromagnetic flow meters are a gold standard in measuring the mean flow velocity of conductive liquids and slurries in process industry. A drawback of this approach is that the velocity field cannot be determined. Information about velocity fields is important for characterizing multiphase flows in the process industry. Recently, electromagnetic flow tomography (EMFT) has been proposed for measuring velocity fields using several coils and a set of electrodes attached to the inner surface of the pipe. The velocity field reconstruction method utilizes a finite element based computational forward model and a Bayesian framework for inverse problem. In the approach, a priori probability and noise models are written describing the flow and measurement error characteristics, respectively. In this work, the effect of additive, possibly correlated, measurement noise and different prior models on the velocity field reconstructions in EMFT are tested using numerical simulations. The results show that the velocity field reconstruction method produces feasible estimates even with relatively high level of correlated measurement noise if the covariance structure of the noise is taken into account. In practice, the noise covariance can be estimated from measurements using sample based methods. Moreover, it is shown that a smoothness prior using a squared exponential covariance function is in general a good choice for the prior model and more advanced prior models for specific flow types such as stratified or turbulent flows can be used.