Data-Driven and Probabilistic Approaches for Seabed Characterization using Free Fall Penetrometers

Free-fall penetrometers (FFPs) enable rapid seabed characterization, but interpreting their data and relating it to soil properties remains challenging. This study employs Bayesian inference to quantify uncertainties in converting FFP measurements, such as acceleration and velocity, into quasi-static tip resistance. In parallel, a multilayer perceptron (MLP) model is trained using laboratory FFP tests in kaolin clay. Results demonstrate that Bayesian inference effectively optimizes semi-empirical equations for FFP data interpretation, even with limited datasets. Furthermore, the MLP achieves substantially higher predictive accuracy in estimating quasi-static tip resistance compared to conventional approaches.