Data-driven machine learning for accurate prediction and statistical quantification of two phase flow regimes

Two different two-phase flow regimes including slug and dispersed flows are examined through the implementation of system identification methods to attain reduced-order models. The obtained models accurately capture the flow dynamics of the studied regimes. The models also provide state-space frequency by defining the transfer functions. The system identification results are compared with those of the bidirectional neural network to predict the phase fraction of the considered two-phase flows. The result of long short-term memory shows correlations of 91% between the real and predicted phase fractions.

Automated summary

Two different two-phase flow regimes were examined using system identification methods to obtain reduced-order models. These models accurately capture flow dynamics and provide state-space frequency through transfer functions. Comparison with bidirectional neural network results showed 91% correlation in predicting phase fractions of the two-phase flows.