Predicting Inside-Pipe Displacement Efficiency with a Machine-Learning Model Using Experimental Data

Abstract

The primary cementing process (PCP) is a crucial procedure in oil and gas well construction, providing well barriers and zonal isolation to ensure safe and efficient operations. However, achieving efficient displacement of yield-stress fluids, such as drilling mud and cement slurries, remains challenging due to complex fluid dynamics and varying fluid properties. This study presents a machine-learning model developed to predict the displacement efficiency of nonNewtonian fluids in vertical and inclined pipes, based on experimental data obtained from a dedicated displacement flow apparatus. The model uses XGBoost to correlate experimental parameters (e.g., fluid properties, velocities, and pipe inclination) with displacement efficiency, achieving an R 2 (R-squared) of 0.772 (95% bootstrap confidence interval (CI) [0.475, 0.889]) on the held-out test data set. Shapley (SHAP) value analysis identifies key factors affecting displacement efficiency, including pipe rotation, imposed velocity, and the densities and viscosities of the displaced and displacing fluids. The results highlight the significance of high pumping rates and casing rotation in optimizing cement displacement inside pipes, aligning with established industry practices. This research demonstrates the potential of machine-learning approaches to enhance cementing operations in challenging environments. Future work can aim at expanding the data set and extending the study to annular displacement flows.