A Cone-Penetration-Test Inversion Model Developed via Machine Learning of Physical and Virtual Calibration-Chamber Experiments

Volume-averaging effects significantly influence the accuracy of cone-penetration-test (CPT) measurements in layered soil profiles. This study presents a machine learning (ML)–based inversion model designed to correct these effects, leveraging a comprehensive set of physical and virtual calibration-chamber experiments. This dataset includes newly generated simulations using discrete element methods, combined with existing data from the literature. The proposed ML model demonstrates statistically significant improvement over competing models, achieving a mean absolute percentage error of 13% on unseen test data. Model limitations include (1) reduced accuracy for very thin layers and for conditions outside the training set’s parameter space; (2) challenges in interpretability inherent to ML-driven approaches; and (3) uncertainties inherent to virtual methods’ abilities to represent physical reality. Notwithstanding these limitations, this study demonstrates the feasibility of a novel approach to developing CPT inversion models and results in an easy-to-use model that warrants adoption and further scrutiny.