Site Response in a Layered Liquefiable Deposit: Evaluation of Different Numerical Tools and Methodologies with Centrifuge Experimental Results

Abstract

Results of a centrifuge experiment simulating seismic site response in a layered level liquefiable soil profile are used to evaluate and systematically compare the predictive capabilities of two common numerical platforms and two classes of advanced soil constitutive models (multiyield and bounding surface) used by three different teams. The pressure-dependent multiyield (PDMY2) and simple anisotropic SAND (SANISAND) constitutive models, implemented in open source and commercially available software, were independently calibrated by three teams using the same set of monotonic and cyclic triaxial test results. Class C predictions of the elastoplastic soil response in centrifuge following verification and initial calibration showed excessive dilative tendencies in all constitutive models to different degrees. These tendencies led to a notable overestimation of acceleration spikes at higher frequencies and an underestimation of net excess pore pressures in dense sand. The second calibration phase focused primarily on reducing soil’s dilative tendencies to match centrifuge tests, even at the cost of slightly sacrificing aspects of the response at an element level or abandoning the number of cycles to liquefaction. Despite differences in calibration methodologies and priorities among three modelers, the results show that small element tests and centrifuge experiments do not always lead to the same calibrated soil parameters. Further, although current numerical platforms and advanced constitutive models were capable of reproducing many aspects of seismic site response observed in the centrifuge, they still need fundamental improvements to capture volumetric settlements. This is an old problem that needs attention in future numerical and physical model studies.