Neural Network Based Performance Model of a Compact Axial Flow Cyclonic Separator

By the nature of their design and because of the lack of residence time, separation efficiency of compact separators is inferior to the bulky gravity separators. Irrespective of this, compact separators have found many applications where complete separation is not necessary. This paper reports on an experimental study of a novel axial flow cyclonic separator (I-SEP) in air water two phase flows. The experimental matrix covered slug and annular flow regimes, with inlet gas volume fractions between 35% to 98%, and the mixture velocity at inlet to the separator ranging from 5m/s to 60m/s. As direct measurements of gas and liquid flow rates at the I-SEP outlets were not possible, in these experiments, the outlets of the I-SEP were connected to downstream gravity separators. The gas and liquid flows at the I-SEP outlets were inferred by measuring the ‘pure’ gas and liquid flow rates at the outlets of the gravity separators. As the two-phase flow is highly pulsating and changing instantaneously, Kalman filter technique was used and was found to be more accurate than using simple mass balance method. The performance parameters, i.e. the Gas Carry-Under (GCU) and the Liquid Carry-Over (LCO) were found to be highly non-linear. An artificial neural network was constructed to predict the separation efficiency for given flow conditions. It was found that by manipulating the pressure difference between the two outlets and the inlet, the GCU and LCO could be controlled. The inverse function of the neural network was used to predict the required pressure differences for the given inlet conditions and the required separation efficiency. This approach will assist the operator to manually or automatically control of the I-SEP performance.