Research on flow regime identification of gas-liquid two-phase flow based on EMD-AR models and CHMM

In view of the non-stationary and nonlinear characteristics of conductance fluctuating signals from gas-liquid two-phase flow, while considering that the neural network has slow convergence in training process and is easy to fall into local minimum, a novel method applied to identify flow regimes was presented in this paper. Firstly, conductance fluctuating signals measured by conductance probes were processed through empirical mode decomposition (EMD), and then a few of stable intrinsic mode functions (IMF) could be obtained. Further several IMF components which contain main information of flow patterns were selected and normalized, and with regard to these IMF components AR models were constructed respectively. Thus, several main auto-regressive (AR) parameters from AR models were input into the continuous hidden Markov models (CHMMs) with different states as feature vectors, and the trained CHMMs were used to identify flow regimes. The results showed that this method has higher discrimination and is simpler and more effective when compared with RBF neural network.