Predictive Modelling of Drilling Fluid Rheology: Numerical, Analytical, Experimental and Statistical Studies of Marsh Funnel Flow
Conference Paper
Overview
Identity
Other
View All
Overview
abstract
AbstractThis paper presents a two-parameter approach, using only density and Marsh funnel time for providing a complete rheological analysis of fluids, encompassing all six standard dial readings as established in the American Petroleum Institute (API) Recommended Practice 13D. The scope of the study includes experimental, analytical, numerical and statistical analyses of flow in the Marsh funnel. A mathematical analysis is developed based on the Marsh funnels geometry and a wide suite of machine learning algorithms are applied to model and predict rheological profiles across the funnel.commercially available numerical simulation software is applied to build a model using computational fluid dynamics. Results from the numerical analysis are corroborated by analytical calculations that are then used to develop a statistical framework for predicting dial readings under various shear rates (3-600 RPM). Experimental results from more than 1500 mud tests are utilized to build ten machine learning algorithms modelling the rheological properties of the fluids. Their performances are evaluated to determine the best models based on three metrics: R-squared values, root mean square error and mean absolute error.Predictions are performed on new mud data and comparisons are made among the ten predictive models broadly categorized into generalized regression models, decision tree-based techniques and miscellaneous approaches. The models show high predictive accuracies on new drilling fluid samples with the performances generally improving with increasing shear rates. A mathematical analysis of the geometry of the Marsh funnel has been utilized to establish a methodology to quickly and accurately perform rheological studies on fluids. Results from experimental, analytical, numerical and statistical studies all closely agree with each other.The outcome of this study can readily be employed at the wellsite to obtain much value from the routine hourly Marsh funnel readings. This can serve as a quick substitute to infrequent conventional rheometer outputs, which are typically obtained only one to four times per day at the field. The rheological results and other derivates such as plastic viscosity (PV) and yield point (YP) are immediately produced after each Marsh funnel test. Informed decisions such as updating hydraulics modelling to improve ROP and hole cleaning can be thus be achieved. This work engages the disciplines of rheology and machine learning to produce a result that enhances the workflow during drilling operations.Added value is derived from an already existing tool to better monitor fluid properties frequently, which is beneficial to the overall surveillance of the drilling fluid system.
