Neural Network-Based Interface Level Measurement in Pipes Using Peripherally Distributed Set of Electrodes Sensed Symmetrically and Asymmetrically

Pipe separators are currently being assessed as substitutes for conventional separators in the oil and gas industry for the separation of gas, oil, and water. In the process of separation, the interface levels between the different media are important measurands to be monitored to optimize the separation process. Electrical capacitance tomography (ECT) without too much focus on tomograms can be used to detect the interfaces in a separator with enough accuracy for control purposes. With the easing of the CPU time needed for image processing, the possibility of getting enough information from reduced number of electrodes has also to be looked into, in view of reducing the processing time. The performance of the ECT system with reduced number of electrodes is studied in this paper using inferential methods based on artificial neural networks. Performance of a 12-electrode ECT system is assessed by studying its performance with only 6, 5, and 4 electrodes. The detection/estimation of interfaces is done effectively and in much shorter time compared to the processing of data with tomograms using a 12-electrode system. The inferential method can handle nonlinearity, and results from it can be easily integrated into other control algorithms addressing the actuators used in separators.