Artificial Neural Networks for ECT based Interface Detection in Separators

Yan Ru1, Chaminda Pradeep1, Saba Mylvaganam1,2

1Telemark University College (TUC)

Faculty of Technology, Department of Electrical Engineering, Information Technology and

Cybernetics, P. O. Box 203, N-3901 Porsgrunn, Norway

2 Telemark Technological R&D Institute, Kjølnes Ring 30, N-3918 Porsgrunn, Norway Email: saba.mylvaganam@hit.no

ABSTRACT

In the oil and gas industry cascaded separator modules deliver the individual components from the produced mixture of gas, oil and water (brine). These separators are usually very large and cumbersome to deal with. For monitoring purposes, it is much easier to deal with the so called pipe separators, dealing with the two phases of oil and water after a conventional separator does the job of separating the gas from the produced mixture of three phases. The oil separators are used in the second stage of separation, viz. that of oil and water (in fact brine).

Tomographic systems have been tested in separators for estimating the distribution of the three phases (including emulsion and foam) via interface detection in various laboratories. A multimodal tomographic system can possibly give useful information about interface fluctuations in the new generation of pipe separators. Experiments and simulations show that capacitance tomography is capable of detecting the interface accurately in spite of the high conductivity of the water present. Disturbing variations in material properties can hinder the correct estimation of the interface level. Due to the high conductivity of the water/brine in the pipe separator, the sensitivity of the capacitance measurements is to a certain extent immune to variations in material properties. In a series of preliminary tests, capacitance tomography was used to estimate the interface in a pipe separator containing oil and water/brine. Results obtained from laboratory scale models are presented and discussed with some information on the uncertainties involved.

Artificial neural networks (ANN) exhibit enhanced ability to mask variations in unwanted/unimportant parameters in the separation process, thus reducing the complexities involved in the solution of the essentially underdetermined system of equations evolving out of different models developed for the system. Due to ample data being available from tomographic systems, a data driven soft sensor (virtual sensor) approach is also discussed with some considerations on processing times to address the potential of the ECT in real time measurement and control applications.

Keywords ECT / ERT (electrical Capacitance / Resistance Tomography), Separators, Pipe separators, Neural Networks, Interface detection, Artificial Neural Networks (ANN), soft sensors, virtual sensors