Novel Machine Learning Approach for Predicting the Gel Strength of the Drilling Fluid While Drilling

Abstract

Abstract Accurately estimating gel strength is paramount for optimizing drilling operations and preventing cuttings from settling at the wellbore's bottom. Traditional methods rely on rotational viscometers, which are time-intensive, equipment-dependent, and lack real-time monitoring capabilities. This study investigates the use of machine learning (ML) techniques to forecast drilling fluid gel strength. A dataset comprising surface drilling parameters and laboratory gel strength measurements was gathered to construct ML models. Selected drilling parameters, such as mud weight and Marsh funnel viscosity, were chosen as model inputs due to their accessibility and cost-effectiveness during drilling. A neural network-based model was trained and assessed using statistical measures like R-squared and average absolute error (AAPE). Results showcased neural networks' ability to predict gel strength accurately, achieving an AAPE below 6.76%. Model validation using an unseen dataset demonstrated close alignment with actual gel strength values, boasting a prediction accuracy surpassing 93% and a low AAPE of 7.21%. Statistical scrutiny affirmed the reliability of the developed neural networks model for real-time gel strength forecasting. This study underscores the feasibility of leveraging machine learning as a practical tool for predicting drilling fluid gel strength, offering real-time monitoring and precise predictions to enhance drilling efficiency, safety, and automation initiatives.