LAUSR

Abstract Drilling depleted and ultra-deep gas reservoirs required special drilling fluids which should be capable of bridging along the walls of the well and withstands a high differential pressure. High-overbalanced water-based drilling fluid (HOBWBDF) is one of the most common drilling fluid used to drill the ultra-deep gas wells. Drilling fluids properties such as yield point (YP), plastic viscosity (PV), apparent viscosity (AV), flow behavior index (n) and consistency index (k) are very vital inputs in managing rig hydraulics, margins of surge and swab pressure, equivalent circulation density (ECD) and hole cleaning. There is a need to develop a new technique to estimate the rheological properties in real-time using the frequent measurements of mud density and Marsh funnel viscosity, which are carried out more frequently (every 15–20 min). The main goal beyond this work is to build novel empirical models that are capable of predicting the above-mentioned rheological properties of high over-balanced water-based drilling fluid (HOBWBDF) in real-time for the first time. The models are built based on only two inputs which are the mud weight from the mud balance test and Marsh funnel viscosity measured by the Marsh funnel test. More than 1200 field measurements of rheological properties were used to feed an artificial neural network in order to develop new novel empirical models. The developed models using the ANN technique showed high accuracy for predicting the different drilling fluid rheological properties. The obtained results showed a maximum average absolute percentage error (AAPE) less than 8% with a correlation coefficient higher than 93%. The main advantages of these models are their inexpensiveness as there is no need for additional equipment to be added to the rig site. Moreover, these models would greatly help the drilling engineers to manage the ECD, surge and swab pressure and hole cleaning which in return will be reflected positively on the drilling performance.