Real-Time Yield Point Prediction for Water-Based Drilling Mud using Particle Swarm Optimised Neural Network*

Abstract

Yield point (YP) is an essential rheological property of drilling mud that influences the ability of the mud to lift well cuttings from the annulus to the surface, impacting the overall drilling efficiency. Despite its significance, YP is typically measured only once or twice a day using complex rheometers. On the other hand, a simple field equipment such as the Marsh funnel is used to constantly monitor the drilling fluid's behaviour up to about 144 times daily. This only provides an indication of the drilling fluid condition and not detailed rheological characteristics. There have been previous attempts to infer the rheological characteristics from the constantly monitored Marsh funnel parameters.  One of the widely used approach to estimate rheological properties from Marsh funnel parameters has been the implementation of Back Propagation Neural Network (BPNN). BPNN algorithm exhibits some drawbacks such as poor generalisation. Based on that, the present study improved the performance of BPNN in predicting YP using particle swarm optimsation (PSO) based BPNN. It was identified from the study that PSO-BPNN outperformed BPNN in the estimation of YP in terms of correlation coefficient (R) and mean square error (MSE) and variance accounted for (VAF). During testing PSO-BPNN attained 0.929, 1.129 and 92.29 % as R, MSE and VAF score, respectively, while BPNN had 0.868, 1.235 and 83.78 % for R, MSE and VAF score, respectively. These findings suggest that PSO-BPNN offers a more reliable and efficient approach to predicting drilling fluid yield point from Marsh funnel experimentation.