Bayesian Estimation of Nonlinear Soil Model Parameters Using Centrifuge Experimental Data

Calibration of nonlinear soil models from experimental data is an essential capability in research and engineering practice alike; however, this task is typically conducted by trial and error. In this study, we describe a Bayesian filtering technique, which is based on an unscented Kalman filter, to systematically assimilate data and estimate the parameters of a veritable soil plasticity model. We first verify the framework using a numerical example. Then, we use the technique to estimate the statistics of the parameters for a multiaxial plasticity model using data from a series of centrifuge tests and infer the maximum shear modulus G_(max), small strain damping, and shear modulus reduction curve G/G_(max). We show that the calibrated soil model, which has a vanished elastic range and a bounding surface, is successful in predicting the soil response for the range of input excitations used in the centrifuge tests. The technique is applicable to other soil models and can be implemented and utilized easily, provided that a standard interface to the soil material model is available.