Drillbit vibrations enable sonic logs prediction in lateral boreholes using machine learning

The development of unconventional resources requires accurate delineation of formation stiffness along extended horizontal borehole segments. This paper studies the feasibility of prediction of rock stiffness using drilling parameters recorded at the rig, downhole measurements while drilling, and near-bit vibrations. We used a machine learning-assisted workflow to model the wireline logs and assess the importance of various input data. The use of nearbit vibrations reduces the prediction error (of sonic data?) from ∼5.3% achieved relying on to 1.8%, which is sufficient for drilling/completion applications. The machine learning algorithm also quantifies the quality of the wireline logs used for training: shear and density logs have ∼20% of noisy data while the compressional velocity ∼35%. Prediction of wireline shear and compressional sonic data using downhole vibrations measured while drilling has vast potential to reduce logging cost for borehole completion industrywide.