Development of New Models to Determine the Rheological Parameters of Water-based Drilling Fluid Using Artificial Intelligence

Abstract

It is well known that drilling fluid is a key parameter for optimizing drilling operations, cleaning the hole, and managing the rig hydraulics and margins of surge and swab pressures. Although the experimental works present valid and reliable results, they are expensive and time consuming. On the other hand, continuous and regular determination of the rheological mud properties can perform its essential functions during well construction. More uncertainties in planning the drilling fluid properties meant that more challenges may be exposed during drilling operations. This study presents two predictive techniques, multiple regression analysis (MRA) and artificial neural networks (ANNs), to determine the rheological properties of water-based drilling fluid based on other simple measurable properties. While mud density (MW), marsh funnel (MF), and solid% are key input parameters in this study, the output functions or models are plastic viscosity (PV), yield point (YP), apparent viscosity (AV), and gel strength. The prediction methods were demonstrated by means of a field case in eastern Iraq, using datasets from daily drilling reports of two wells in addition to the laboratory measurements. To test the performance ability of the developed models, two error-based metrics (determination coefficient R2 and root mean square error RMSE) have been used in this study. The current results of this study support the evidence that MW, MF, and solid% are consistent indexes for the prediction of rheological properties. Both mud density and solid content have a relative-significant effect on increasing PV, YP, AV, and gel strength. However, a scattering around each fit curve is observed which proved that one rheological property alone is not sufficient to estimate other properties. The results also reveal that both MRA and ANN are conservative in estimating the fluid rheological properties, but ANN is more precise than MRA. Eight empirical mathematical models with high performance capacity have been developed in this study to determine the rheological fluid properties based on simple and quick equipment as mud balance and marsh funnel. This study presents cost-effective models to determine the rheological fluid properties for future well planning in Iraqi oil fields.