Application of Machine Learning to Predict Pile Driveability Performance in Offshore Southeast Asia Soils

1. Home
2. Publications
3. Online Library
4. Application of Machine Learning to Predict Pile Driveability Performance in Offshore Southeast Asia Soils

Machine learning (ML) based modelling methods enable geotechnical engineers to leverage state-of-the-art tools to create predictive models to be applied in various complex geotechnical engineering problems. This paper investigates the ML techniques and algorithms that can be adopted to develop suitable ML models with actual pile installation and cone penetration test (CPT) data to predict blowcounts and hammer energy. For this study, ML models from scikit-learn (sklearn) libraries in Python such as Random Forest (RF), Support Vector Machine (SVM), Decision Tree (DT), Linear Regression (LR), and Polynomial Regression (PR) have been considered based on their ability to tackle regression problems. It is necessary to calculate the soil resistance during driving (SRD) using the CPT dataset to have a better ML model since offshore piles comes in many sizes. Then, SRD and actual pile installation datasets from sites around offshore Southeast Asia were resampled to a regular grid of 0.25m intervals to facilitate data handling prior establishing a ML model. Outcome from the ML models were interpreted in form of R-Square and Root Mean Square Error (RMSE). Two ML models were generated, a model to predict the blowcounts and a model to predict the hammer energy used based on the actual pile installation and CPTU datasets provided. Based on the five ML algorithms, RF gave the highest R-Square value for predicted blowcounts ML model with a value of 0.801 followed by DT, PR, LR and SVM. As for the hammer energy ML model, RF again showed better results with R-Square value of 0.911 followed by PR, DT, LR and SVM. Soil variability at each location around Offshore Southeast Asia dictates the result of obtaining a good ML model. Based on this assessment, Random Forest ML model has consistently appeared to be the best ML model to predict the blowcounts and energy required to install offshore piles to its design depth thus minimizing potential issues such as pile refusal.

5th International Symposium on Frontiers in Offshore Geotechnics (ISFOG2025)

5 - Data Analytics and Machine Learning