Gas–liquid intermittent flow rates measurement based on two-phase mass flow multiplier and neural network

Abstract

Online flow rates measurement is encountered in many areas, and efforts are being committed to improving the accuracy of measurement, especially in intermittent flow. This paper concerns gas and liquid flow rates predictions in intermittent flow by fusing the information of the cone meter and conductance ring sensors. A novel correlation of two-phase mass flow multiplier (K) is derived from the separated flow model and modified by experiment data. The feed-forward neural network (FFNN) is used to established the predicted model of Lockhart–Martinelli parameter (X LM). Five FFNNs with different input parameters are compared and results show that FFNNs with dual conductance ring sensor signals have better performance than those with differential pressures of cone meter. Then, the FFNN with conductance ring sensor signals as inputs is used to establish the X LM predicted model. The effect of temperature on the conductance of liquid is also innovatively considered in the neural network. Therefore, no extra reference conductance ring sensor is needed. Finally, combining the X LM model and K correlation, gas and liquid mass flow rates are measured. The relative error of the predicted liquid mass flow rate is less than ±5.0% at the confidence level of 99%. For the gas phase, the relative error is within ±15.0% at the confidence level of 98%. The newly developed metering method provides a high-accuracy as well as a cost-effective measurement method for the intermittent flow.