AI-enhanced soil classification with incomplete CPT data for offshore wind farm

Area of Science:

• Geotechnical Engineering
• Artificial Intelligence in Geosciences
• Offshore Renewable Energy

Background:

• Accurate soil classification is critical for offshore wind farm foundation design.
• Conventional cone penetration test (CPT) methods struggle with incomplete datasets in offshore environments.
• Developing robust AI solutions for geotechnical data analysis is essential.

Purpose of the Study:

• To present an artificial intelligence (AI) enhanced framework for soil classification using the Robertson Classification.
• To emphasize the framework's robustness with incomplete cone penetration test (CPT) data.
• To provide a practical solution for offshore wind farm geotechnical design.

Main Methods:

• Generated a synthetic CPT database of 229,808 samples using diverse sampling strategies.
• Evaluated four machine learning models, including random forest.
• Simulated missing CPT input parameters and employed Monte Carlo simulations for uncertainty analysis.

Main Results:

• The random forest model achieved the highest performance with R² of 0.99 and 92.53% classification accuracy.
• Reliable soil classification predictions were maintained despite incomplete CPT data.
• Cone tip resistance (q_c), sleeve friction (f_s), and effective stress (σ'_v) were identified as key features.

Conclusions:

• The AI-enhanced framework offers a robust solution for CPT-based soil classification with incomplete datasets.
• This approach is practical for offshore wind farm geotechnical design.
• The study highlights the potential of AI in overcoming data limitations in geotechnical investigations.