Wet gas flow metering with pattern recognition techniques

The development of many gas condensate fields and the increasing number of marginal fields whose economics do not support conventional bulky separation and processing facilities means that new wet gas flow metering techniques are becoming of greater importance to the oil and gas industry worldwide. For the purpose of this research wet gas flow is defined as multiphase flow (gas-liquid) having in-situ gas volume fraction greater than 95 % at the point of measurement. This research presents a novel wet gas measurement technique involving the use of a standard Venturi meter together with advanced pattern (PR) recognition methods for the detection of liquid presence in wet gas flow conditions and the simultaneous measurement of gas and liquid flow rates without the need for preconditioning of the flow or prior knowledge of either phase. The technique involves four major steps: 1) collection of experimental data spanning the range of flow regimes likely to be encountered in wet gas flow conditions; 2) extraction of flow dependent variables from the Venturi pressure sensors in the form of features; 3) development of PR model for mapping between input features and corresponding gas and liquid flow rates; 4) generalisation test to new and previously unseen flow conditions to determine the accuracy of the Venturi-PR methods developed in this research work. Data was sampled at 50 Hz using two axial differential pressure sensors and one singleend absolute pressure sensor on a 2-inch horizontally mounted Venturi meter using airwater at normally atmospheric conditions. Extensive features were extracted from the time and frequency domains of the raw data and evaluated for their discriminatory ability between different flow conditions. A Bayesian multi layer perceptron (MLP) neural network was used to construct a non-linear mapping between the different feature vectors and the corresponding gas and liquid flow rates using a correctly labelled training data. When the generalisation performance of different measurement scenarios developed was tested, the cross-sensor data fusion of the amplitude features achieved 100 % of the test data to within ± 5 % error across the whole flow domain of interest. The Venturi-PR results also performed significantly better than published wet gas differential pressure flow correlations over the flow domain of interest.