Experimental Investigation of Two-Phase Flow Performance of Electrical Submersible Pump Stages

Pessoa, Rui (PDVSA-Intevep) | Prado, Mauricio (The University of Tulsa)

OnePetro 

Abstract

Two-phase flow behavior prediction of centrifugal pumps is a hard task due to the complexity involved in modeling multiphase flow inside turbo machines. No models are currently available for this purpose. Some empirical correlations are available in the literature, but they are valid only for the tested pumps in the experimental range used to develop them. An experimental study has been conducted at The University of Tulsa Artificial Lift Projects - TUALP with a 22-stages GC6100 pump to gather data for pump performance under two-phase flow conditions. Air and water were used as working fluids. This study differs from other experimental works because the pressure changes were recorded stage-by-stage. The results of previous works have been reported as an average of the intake and discharge conditions, and depend on the number of stages used.

Phenomena like surging and gas locking were observed during these tests and their boundaries have been mapped. It will provide some insight regarding when they appear, and the way they are revealed.

The pressure increment and total hydraulic horsepower for the average pump and per stage as a function of the liquid flow rate, and each gas flow rate considered are presented. The average brake horsepower and efficiency for the pump are also plotted for the variables mentioned.

The results indicate that the average behavior for the pump is significantly different from that observed per stage.

Introduction

Centrifugal pumps are dynamic devices which use kinetic energy to increase liquid pressure. They are successful with handling water and other incompressible fluids ranging from low to medium viscosities but are severely impacted by free gas or highly compressible fluids.

Significant amounts of free gas may be found during hydrocarbons production. This motivated important research from the petroleum industry focusing on improving the successful application of ESP as an artificial lift method.

The consequences of entrained gas on centrifugal pumps depend on the relative amount of gas and liquid present, and vary from a slight deterioration on performance up to a complete blockage known as "gas locking". Before gas locking occurs, another phenomenon known as surging takes place.

Each pump is characterized by performance curves, which include the head developed, brake horsepower consumption and efficiency as function of the flow rate through the pump for a certain rotational speed (see Fig 1). Traditionally these curves are determined experimentally using water.

The head characteristic curve is used to size the pump, while the brake horsepower information is useful to size the motor required to drive the pump. The sizing of a multi-stage ESP for water wells is fairly simple, and good accuracy of the predicted performance is achieved using the water performance information supplied by the manufacturer.

The design of an ESP system using the water information for oil wells with high free gas fraction at pump intake conditions is a harder task, and is based on the prediction of performance curves by modification of the water curves. The leading parameter is the mixture density at the flow conditions of each stage. Applying this procedure, the ESP system often shows some degree of under or over sizing when operating.

An accurate prediction of the performance for any pump handling free gas is challenging. Some empirical and mechanistic approaches have been attempted in the past. The main problem of the experimental approach is that the developed correlations are based on the average performance of the pump. These correlations become specific for the type and number of stages tested. On the other hand, theoretical models are difficult to develop since the geometry of the channels inside the pump is complex. The phenomena that take place in such channels are not well understood, and thus the use of empirical parameters to close the model is required.