• Alsaba, M. T.
• Amadi, K. W.
• Elgaddafi, R. M.
• Prabhu, R.
• Iyalla, I.

Abstract

<jats:title>Abstract</jats:title><jats:p>Wellbore instability is the most significant incident during the drilling of production sections of most wells. Common problems such as wellbore collapse, tight hole, mechanical sticking, cause major delays in drilling time due to extended reaming and sidetracking in worst-case scenario. Geomechanical property of rock such as Unconfined Compressive Strength (UCS) affects wellbore stability, drilling performance and formation in-situ stresses estimation. Conventional methods used to estimate UCS requires either laboratory experiments or derived from sonic logs and the main drawbacks of these methods are the data and samples availability, high costs and time This paper presents an alternative technique of utilizing real-time drilling parameters and machine learning (ML) algorithm in the prediction of UCS thereby enabling timely drilling decisions. ML algorithm enables a system to learn complex pattern from the dataset during the training (learning) phase without any specified mathematical model and afterwards the trained model can predict through a model input. In this work, five ML models were used to predict UCS using offset well data from an already drilled wells. The models include; artificial neural network (ANN), CatBoost (CB), Extra Tree (ET), Random Forest (RF) and Support Vector Machine (SVM). The ML models were first trained with 1150 data points using a 70:30 percentage ratio for training and testing the model respectively. After that, 560 datapoints from a different well were used to validate the developed model. The real-time drilling parameters required included weight on bit, penetration rate, rotary speed, and torque. The analysis result revealed good match between the actual and predicted (UCS) with correlation coefficients for training and testing dataset; 0.970 and 0.70 and 0.85 and 0.77 for CatBoost and ANN respectively. The main added value of this approach is that these drilling parameters are readily available in real-time and timely drilling decisions can be modified to improve the drilling performance.</jats:p>

Topics

• impedance spectroscopy
• phase
• experiment
• strength
• random
• machine learning