AI-based real-time multi-model framework for groundwater-responsive pile behaviour

Abstract

Piles react nonlinearly and time-dependently to fast groundwater level fluctuations due to pore pressure and effective stress changes. Traditional design models use static stress assumptions and cannot update soil–structure interactions in real time, limiting their hydrologically active prediction power. This research presents an integrated multi-model framework with sensor networks, adaptive soil–mechanics formulations, machine-learning-based modulus development, probabilistic failure evaluation, and actuator-driven stiffness control. The framework includes a real-time groundwater–Terzaghi interaction model for dynamic effective stress evaluation, a transformer-network architecture for evolving soil modulus prediction, a Bayesian load-redistribution model for failure probability estimation, and a self-actuated response optimizer for localized stiffness corrections. Hydro-mechanical conditions are updated continuously via closed-loop elements. The integrated system outperforms earlier methods in settlement prediction, directional stress estimation, modulus evolution tracking, and risk-aware load control using full-scale pile testing datasets & samples. Pile foundations subjected to different groundwater regimes improved in accuracy and reliability, enabling intelligent, self-regulating geotechnical infrastructure sets.