Identification of flow regimes using back-propagation networks trained on simulated data based on a capacitance tomography sensor

Abstract

Non-invasive techniques such as electrical capacitance tomography (ECT) are beginning to make promising contributions to control systems and are well fitted for flow-regime identification in opaque pipes or conduits. A new method of two-component flow-regime identification based on a neural network and an eight-electrode ECT sensor is proposed in this paper. Time-consuming image reconstruction and analysis are avoided. Ten feature parameters are extracted straight from the capacitance measurements and translated into regime information via a back-propagation (BP) network. The extraction of feature parameters, the architecture and the training of the BP network are given. Simulation results show that the new identification method has good precision and high speed. The use of feature parameters and the BP network for flow-regime identification is promising.