Noniterative Wet Gas Flow Metering in Annular Mist Flow by Combining Vortex Meter and Liquid Film Parameters

Wet gas metering in annular mist flow is important for it is typically used in fiscal metering, and the traditional iteration-based methods are limited due to their sensitivity to initial values and time-consuming in the iteration process. To improve the real-time performance and reliability, this study tries to develop noniterative wet gas metering method based on the vortex meter-conductance dual-modality system. To adapt to different application scenarios, two models are proposed, named nonlinear regression model and neural network model, respectively. The tests are conducted on a DN 15 vertical pipeline with a gas flow rate range of 12–24 m3/h and liquid volume fraction within 1.30‰. To obtain the flow parameters of liquid film thickness, disturbance wave frequency, and velocity simultaneously, the dual-ring liquid film sensor is designed and optimized. The input features are selected by the importance ranking of the correlation heatmap, and the optimized hyperparameters and K-fold cross-validation ensure the generalization of the trained neural network. Finally, the predicted performances are evaluated and compared in detail. It indicates that mean film thickness could be regarded as the unique scale parameter of vortex meter overreading (OR), and $\textrm {OR}=1+2.64{\delta _{m}} / D$ gives a predicted error of ±1.5% for gas flow rate. Liquid flow could be directly predicted by combining different liquid film flow parameters, and $U_{\textrm {sl}}=535.62f_{\textrm {DW}}^{0.214}D^{2.034}\delta _{m}^{0.655}V_{\textrm {DW}}^{2.090}$ gives a predicted full-scale error of ±5.0% for liquid flow rate. The neural network model gives satisfactory prediction both for gas and liquid flow rates, gas flow accuracy within ±1.0% error bands, and liquid flow within ±1.5% full-scale error bands. Finally, comparisons and practical suggestions for the application are given for the two proposed noniterative wet gas metering models.