Viscometer Readings Prediction of Flat Rheology Drilling Fluids Using Adaptive Neuro-Fuzzy Inference System

Abstract

Flat rheology drilling fluids are synthetic-based fluids designed to provide better drilling performance with flat rheological properties for deep water and/or cold environments. The detailed mud properties are mainly measured in laboratories and are often measured twice a day in the field. This prevents real-time mud performance optimization and negatively affects the decisions. If the real-time estimation of mud properties, which affects decision-making in time, is absent, the ROP may slow down, and serious drilling problems and severe economic losses may take place. Consequently, it is important to evaluate the mud properties while drilling to capture the dynamics of mudflow. Unlike other mud properties, mud density (MD) and Marsh funnel viscosity (MFV) are frequently measured every 15–20 minutes in the field. The objective of this study is to predict the viscometer readings at 300 and 600 RPM (R600 and R300) of the flat rheology mud in real-time using machine learning (ML) and then calculate the other rheological properties using the existing equations. The developed model using adaptive neuro-fuzzy inference system (ANFIS) predicted the viscometer readings with an acceptable accuracy. The maximum average absolute percentage error (AAPE) was less than 7 % and the correlation coefficient (R) was more than 0.96 for training, testing and validation.