Accurate prediction of liquid holdup associated with multiphase flow is a critical element in the design and operation of modern production systems. This prediction is made difficult by the complexity of the distribution of the phases and the wide range of fluid properties encountered in production operations. Consequently, the performance of existing correlations is often inadequate in terms of desired accuracy and range of application. This investigation focuses on the development of a neural network model, a relatively new approach that has been successfully applied to a variety of complex engineering problems. Data from five independent studies were used to develop a neural network for predicting liquid holdup in two-phase horizontal flow. A detailed comparison with existing empirical correlations and mechanistic models reveals that the neural network model shows an improvement in overall accuracy and performs more consistently across the range of liquid holdup and flow patterns.
Introduction
Two-phase flow of gas and liquid in pipes is a near universal occurrence in the petroleum industry. Advancements in subsea completion and multiphase pumping and metering technology have extended multiphase flow over relatively long distances to centralized gathering and separation systems. These developments are becoming increasingly common, especially in remote and hostile locations such as the deepwater Gulf of Mexico. In such cases upfront capital costs are reduced through the consolidation of surface facilities while minimizing flaring and environmental impacts.
The design of multiphase production systems requires an accurate estimation of pressure loss.1,2 Liquid holdup, defined as the in-situ liquid volume fraction, is generally the most important parameter in calculating pressure loss. Liquid holdup is also necessary to predict the occurrence of hydrate formation and wax deposition, and to estimate the liquid volume during pigging operations for sizing slug catchers.
While pressure losses in single-phase flow in pipes have for a long time been accurately modeled with familiar expressions such as the Bernoulli equation and the Navier-Stokes equations, accurate predictions of pressure loss in two-phase flow has remained somewhat elusive because of added complexities. The lower density and viscosity of the gas phase causes it to flow at a higher velocity relative to the liquid phase, a characteristic known as slippage. Consequently, the associated frictional pressure losses result from shear stresses encountered at the gas-liquid interface as well as along the pipe wall. Additionally, the highly compressible gas phase will expand as the pressure changes along the flow path.
Further complicating matters is the variety of physical distributions among the phases. This has led to the common practice of characterizing multiphase flow with flow patterns (Fig. 1). The prevailing flow pattern for a specific set of conditions depends on the relative magnitude of the forces acting on the fluids. Buoyancy, turbulence, inertia, and surface tension forces are greatly affected by the relative flow rates, viscosities, and densities of the fluids as well as the pipe diameter and inclination angle. The complex dynamics of the flow pattern govern effects of slippage and therefore variations liquid holdup and pressure gradient.