Experimental and computational modeling of Electric Submersible pumps (ESP) in two-phase gas-liquid flow systems enhanced with machine learning

Electric submersible pumps (ESPs) handle high flow rates but suffer performance losses under gas–liquid multiphase flow, which introduces instabilities and limits accurate prediction. This study combines experimental data, CFD modeling, and machine learning to improve performance analysis, using Simcenter STAR-CCM+ with a multistage P100 pump and a k-ω SST model with refined mesh (y+ ≈ 1.8).

Results show that transient simulations in a Rigid Body Motion reference frame are essential to capture dynamic interactions and pressure fluctuations, unlike steady-state approaches. Integrating these results with machine learning enhances prediction accuracy and provides a robust framework for optimizing ESP performance in two-phase conditions.