The study explores different system identification techniques that utilize machine learning, including the Restoring Force Surface (RFS) method and Sparse Identification of Nonlinear Dynamics (PySINDy).
These methods have potential to help uncover physical models for soil-pile interaction.
To test these methods , the research first applies them to well-known benchmark systems—simple mechanical models with known nonlinear behaviours. This ensures that the identification techniques work correctly before applying them to real pile-driving experiments.
The experimental data comes from lab-scale vibratory pile-driving tests using strain gauges and accelerometers. The study analyses how forces acting on the pile change over time, focusing on both the tip and shaft resistance. Various mathematical models are tested to see which best captures the nonlinear behaviour.