Pressure drop estimation in horizontal annuli for liquid–gas 2 phase flow: Comparison of mechanistic models and computational intelligence techniques

• We measured pressure drop for different liquid and gas flow rates experimentally.
• Lockhart and Martinelli model was modified using the experimental data.
• Four computational intelligence techniques were developed for pressure drop estimation.
• Mechanistic model and computational intelligence techniques estimated pressure losses successfully.
• We proposed the simplified calculation methodology to estimate pressure drop for two phase flow.

Frictional pressure loss calculations and estimating the performance of cuttings transport during underbalanced drilling operations are more difficult due to the characteristics of multi-phase fluid flow inside the wellbore. In directional or horizontal wellbores, such calculations are becoming more complicated due to the inclined wellbore sections, since gravitational force components are required to be considered properly. Even though there are numerous studies performed on pressure drop estimation for multiphase flow in inclined pipes, not as many studies have been conducted for multiphase flow in annular geometries with eccentricity. In this study, the frictional pressure losses are examined thoroughly for liquid–gas multiphase flow in horizontal eccentric annulus.Pressure drop measurements for different liquid and gas flow rates are recorded. Using the experimental data, a mechanistic model based on the modification of Lockhart and Martinelli [18] is developed. Additionally, 4 different computational intelligence techniques (nearest neighbor, regression trees, multilayer perceptron and Support Vector Machines – SVM) are modeled and developed for pressure drop estimation.The results indicate that both mechanistic model and computational intelligence techniques estimated the frictional pressure losses successfully for the given flow conditions, when compared with the experimental results. It is also noted that the computational intelligence techniques performed slightly better than the mechanistic model.