Data Driven Models for Estimation of Drilling Fluid Rheological Properties and Flow Rate

Abstract

Pressure control during well drilling operations includes management of the well pressure above the formation pore pressure and below the formation fracture pressure. If these boundaries are not kept, an influx of formation fluids or an uncontrolled loss of drilling fluid may occur. These incidents present serious risk to humans, assets and the environment. To avoid serious risks and accidents, detection of the influx (kick) and loss incidents is of vital importance in any drilling operation. The availability of sensor technologies to discover kick/loss incidents vary greatly for different drilling operations. Common for all is that dedicated measurements of the non-Newtonian drilling fluid can help the early identification of the possible occurrence of measurements of such incidents.

The inflow of drilling fluid is in most drilling operations well measured, either by pump output or flow metering on the flowline, such as Coriolis meters. The return flow is comparatively harsher for flow meters, as the return fluid flow contains formation cuttings and formation fluids. The common industry practice is applying a paddle flow meter, which gives a trend-based measurement that by human interpretation alongside other measurements may indicate anomalies in the return fluid flow. Secondly, a fluid level in the drilling fluid tank may be measured, as a level change can indicate a kick/loss incident. The response of this method is slow and inaccurate. Development of cost effective, fit-for purpose and in-line sensor technology for the return fluid flow will increase the capability for automation and reduce the time delay to detect kick/loss incidents.

This research work studies the applicability of a modified open channel for fluid return with a Venturi constriction. The subsequent level changes in the open channel may be measured and used to model the fluid flow rate. During this work, it was found that precise knowledge of the fluid properties is a requirement to some models, and beneficial to others. It is also vital in determining the volume of the kick/loss incident, and the subsequent correct procedures in handling the situation safely and effectively.

In this research ultrasonic characterizations of drilling fluids serve as inputs to models estimating the fluid rheological properties during drilling operations. The common industry practice of intermittent, offline and manual drilling fluid characterization is not satisfactory for automated drilling operations and continuous measurement systems, and improvements are needed.

Non-Newtonian drilling fluid flow is difficult to model precisely with mechanistic models. Data driven models are selected as suitable methods to handle the non-linear behaviours of these fluids when estimating fluid flow based on the level measurements. The models are trained and verified by using experimental data from a test flow loop. The data driven models are compared to other mechanistic models developed by colleagues that were verified using the same experimental setup.

Ultrasonic wave propagation is affected by the acoustic properties of the medium it propagates. By analysing this propagation in drilling fluids, the effect of the fluid rheological properties on the ultrasonic waves is studied in the present study. The work focuses on density and viscosity, as these are some of the drilling fluid rheological properties essential in the pressure control during the drilling processes. The relationship between ultrasonic properties and rheological properties are not fully described in literature, and data-driven models were identified as potential solutions. Three drilling fluid systems are diluted to yield in total 33 fluid samples that are characterized in ultrasonic transmission experiments, and their rheology is analysed. Data driven models are developed and verified using these data, to estimate the rheological properties based on the ultrasonic measurements.

The data driven models to estimate the fluid flow rate performed to expectation in the experimental setup. Several types of models were developed, but all had accuracies better than the industry standard set by NORSOK, at 5 % accuracy of measured value. Some of the mechanistic models outperformed the data-driven models, and the thesis work discusses the results and the strength and limitation of the models.

Data driven models proved to be an effective approach in estimation of fluid rheological properties. The two selected properties were estimated within the NORSOK suggested accuracy of 2%. The measurement principle with ultrasonic measurements and models has potential to be developed to apply to flowing systems and improve fluid flow models and improve availability of continuous drilling fluid properties measurements.