Applying Machine Learning for SPT-N Prediction from Raw CPT Data

Create Account

Availability

On-Demand

Cost

Member: $10.00

Non-Member: $20.00

Credit Offered

No Credit Offered

Abstract: Managing geotechnical risk requires robust methods for interpreting field data to inform design and decision-making. Machine Learning (ML) enhances this by providing predictive models that derive actionable insights from raw data. This study applies ML techniques to Cone Penetration Test (CPT) data to predict Standard Penetration Test (SPT) N-values. CPT is increasingly favored for its reliable, swift data collection, often surpassing SPT in applications like field exploration, design evaluations, and quality control. Both tests have been key in correlating soil index and engineering properties, essential for foundation design. Predicting SPT-N values from CPT data aids in managing risk by offering more comprehensive soil profiles and reducing design uncertainty.

This research uses data from a geotechnical investigation of a 16-mile expansion of the I-405 Freeway in Orange County, California, involving lane additions, bridge replacements, and Earth Retaining Systems (ERS). A total of 581 exploratory points, including 207 CPT soundings and 374 SPT borings, were used to assess subsurface conditions. From this, 90 CPT soundings and adjacent SPT boreholes were selected to develop ML models based on raw CPT data, corrected SPT-N values, and Unified Soil Classification System (USCS) groups. The data was prepared for model training and testing, with analyses tailored to different soil groups to reveal correlations between CPT and SPT-N values. Rigorous validation ensures the model's reliability.

By utilizing ML to correlate CPT data with SPT-N values, this research provides tools for more accurate soil characterization, which in turn helps manage the risk associated with infrastructure projects. These models contribute to data-driven decision-making, reduce uncertainty in foundation design, and improve the integration of field testing with predictive analytics, aligning with the broader theme of using testing and data to manage risk in geotechnical engineering.

Dr. Milad Fatehnia; Dr. Sanaz Paran

Annual Conference Proceedings