Geomechanical Rock Properties from Surface Drilling Telemetry

Abstract

We present a novel approach for real-time estimation of the mechanical properties of rock with drilling data. We demonstrate that surface drilling telemetry (also known as mud logging) can be used as an input for a trained machine learning (ML) algorithm to predict the properties of the rock being drilled at the moment. The study involves data from several real wells with horizontal completions. We use mud logging and logging while drilling (LWD) data from one part of the wells to train various ML models. The models are compared by various metrics using the five fold cross-validation technique. We also show the importance of proper feature selection for maximizing models’ performance in operation mode.