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

PPT Presentation

Pdf Summary

The presentation examines using machine learning with Measurement While Drilling (MWD) data to estimate geomaterial strength and support deep-foundation design. MWD instruments continuously record drilling parameters—including depth, thrust, rotation speed, drilling rate, and torque—while drilling. Field measurements across Illinois sites covered varied soils, including cohesive soils, sands, till, gravel, and shale.<br /><br />Researchers used an XGBoost model trained on 10,606 data points, with an 80/20 training-test split. The model classified soil types with 98% test accuracy and predicted SPT-N values with an R² of 0.93 and unconfined compressive strength (Qu) with an R² of 0.96. These predictions can turn continuous MWD measurements into continuous strength profiles, complementing conventional, discontinuous SPT and strength-test data.<br /><br />The predicted values were incorporated into a spreadsheet tool to estimate pile lengths and bearing resistance. In the example presented, MWD-based estimates produced up to a 10% increase in factored resistance for metal shell piles and H-pile lengths up to seven feet shorter than estimates based on SPT data.<br /><br />The approach could improve the continuity and efficiency of subsurface characterization and inform pile design. However, it is not suitable for all soil types, and the presentation notes limitations associated with discrete measurements and the need to account for the drill rig’s mechanical and hydraulic response.

Keywords

machine learning

MWD data

XGBoost

geomaterial strength

deep-foundation design

soil classification

SPT-N prediction

unconfined compressive strength

pile design

subsurface characterization