Answers are grounded in this paper and cite their sources.
Accurate prediction of blow count during pile driving is essential for ensuring the safety and efficiency of offshore pile installation. However, traditional CPT-based pile drivability prediction relies on simplified empirical assumptions and often exhibit limited generalisability across heterogeneous soil profiles. This study proposes a physics-informed feature engineering approach that explicitly incorporates CPT-derived geotechnical information into feature engineering. Using a jacket pile installation dataset from the North Sea, we developed three physics-informed feature strategies: inter-pile distance group, soil behaviour type classification, and soil resistance to driving (SRD) derived from CPT data. Three machine learning models with distinct architectures namely multilayer perceptron (MLP), random forest (RF), and extreme gradient boosting (XGBoost) were selected to demonstrate the efficacy of the proposed physics-informed features. Results show that incorporating SRD substantially enhances prediction accuracy by transforming CPT measurements into physically interpretable resistance indicators. In contrast, inter-pile distance grouping and soil behaviour type classification yield marginal performance enhancements due to weak spatial correlation in offshore settings and the inherent robustness of tree-based models to categorical inputs, respectively. A further SHAP-based model interpretability analysis confirmed that the CPT-informed SRD feature contributes the most to the model’s predictive performance for pile drivability. These findings demonstrate that explicitly incorporating CPT-derived physical resistance through feature engineering can significantly enhance the performance of data-driven pile drivability prediction.
No takes yet. Share an insight, caveat, or question.
Chen et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: