Prediction of Strength of Geomaterials with Machine Learning Integrated Measurement While Drilling-Paper

Paper

Pdf Summary

The study evaluates whether Measurement While Drilling (MWD), combined with machine learning, can improve geotechnical site characterization. Instrumented rotary drilling at five sites across Illinois continuously recorded drilling rate, thrust, rotational speed, and torque. The resulting dataset covered varied soil and rock conditions, though some torque readings were missing.<br /><br />Researchers used machine-learning imputation to estimate missing torque values, then trained XGBoost models to predict soil type, Standard Penetration Test (SPT-N) values, and undrained shear strength (Qu). The models used MWD measurements and related drilling information as inputs. On held-out test data, the soil-classification model achieved 98% accuracy, while the SPT-N and Qu regression models achieved R² values of 0.93 and 0.96, respectively.<br /><br />The results indicate that MWD data can support continuous subsurface profiling and help fill gaps between conventional, discrete strength measurements. In practice, the approach could complement standard site investigations and potentially reduce the need for extensive in-situ testing. However, the authors note that the current dataset limits the models’ use as a standalone alternative. Broader data covering more soil types and ground conditions will be needed to establish wider generalizability.

Keywords

Measurement While Drilling

MWD

machine learning

geotechnical site characterization

XGBoost

soil classification

SPT-N

undrained shear strength

torque imputation

subsurface profiling