Physics-Informed Neural Networks (PINNs) Approach for Interpretation of Piezocone Tests Based on Cavity Expansion and Consolidation Theory

This award addresses a critical gap in the modeling and interpretation of the fundamental piezocone test for geotechnical analysis and design. The cone penetrometer test is an efficient method for (inversely) determining key soil properties in situ. However, the current achievements made on the cone penetration interpretation are largely based on empirical correlations, rather than on a rigorous theoretical framework that properly models the penetration procedure. This research will leverage the emerging Physics-Informed Neural Networks technique (a theory-guided and data-driven approach grounded in Artificial Intelligence) to establish a new paradigm for tackling the fundamental yet challenging inverse problem of piezocone interpretation. The novel interpretation model will enable the in situ geotechnical parameters to be predicted in a more comprehensive, physically meaningful, and accurate manner. The developed computational codes will be publicly available through GitHub. Improved interpretation of piezocone data plays a vital role in mitigating the risk of geotechnical failure and reducing uncertainties in soil behavior prediction, hence enhancing the resilience and sustainability of civil infrastructure. Outcomes of this research also have cross-disciplinary impacts in underground, petroleum, and mining engineering, as well as applied mathematics. The project will further support workforce development through educational and outreach activities involving both college and middle/high school students via established programs at Louisiana State University. This award will develop a Physics-Informed Neural Networks framework for solving the piezocone penetration problem, in both its forward and inverse forms. The soil behavior will be described by using the modified Cam Clay constitutive model. The research first aims to develop two sets of rigorous yet tractable differential equations in Lagrangian form governing the large-strain cavity expansion and consolidation processes. These governing equations will then be embedded within the Physics-Informed Neural Networks framework, serving as the constraints for the surrogate model to solve both the forward and inverse cavity expansion and consolidation boundary value problems that effectively simulate the piezocone penetration/dissipation process. The forward Physics-Informed Neural Networks solutions will be verified with analytical and finite element results. The inverse framework, combined with field-measured data as additional training constraints, will enable the piezocone interpretation models to produce more realistic estimates of essential soil parameters. The research will advance the current state of knowledge in cavity expansion/consolidation theory and modelling and foster the application of scientific machine learning in geotechnical in situ characterization. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.