Developing ultrasonic soft sensors to measure rheological properties of non-Newtonian drilling fluids

摘要

Surveillance of the rheological properties of drilling fluids is crucial when drilling oil wells. The prevailing standard is lab analysis. The need for automated real-time measurements is, however, clear.Ultrasonic measurements in non-Newtonian fluids have been shown to exhibit a non-linear relationship between the acoustic attenuation and rheological properties of the fluids. In this paper, three different fluid systems are examined. They are diluted to give a total of 33 fluid sets and their ultrasonic and rheological properties are measured. Machine learning models are applied to develop soft sensors that are capable of estimating the rheological properties based on the ultrasonic measurements. This study explores three different machine learning model types and, extensive training and tuning of the models is carried out. The best model types that show good results and the potential to develop a real-time sensor system suitable for use in oil & gas drilling process automation are selected.