Conventional plasticity assumes that a yield surface exists and the direction of plastic strain increment (DPSI) is uniquely dependent on the current stress state. Triaxial stress probing tests of yield and plastic flow of sand have been conducted using discrete-element modelling with polyhedral particles resembling the shapes of Toyoura sand. It is found that a yield surface does not exist, but a memory surface (MS) separating two types of distinct sand behaviour can be established. Within the MS, the DPSI is primarily controlled by the stress increment, and the magnitude of plastic strain increment is insensitive to the stress increment direction. When the stress state is on or outside the MS, a much larger plastic strain increment is observed if the stress increment points outside the MS, and the DPSI is dependent on both the current stress state and stress increment. The shape and size of the MS, which can be modelled by the SANISAND yield function, are dependent on the soil density and evolve with plastic strain.

Modelling the cyclic response of granular materials is key in the design of several geostructures. Over the years, numerous constitutive models have been proposed to predict the cyclic behaviour of granular materials. However, pertaining to the hypoplastic constitutive models, one of the significant limitations is their inability to accurately predict the geomechanical response during the unloading and reloading phases. This study introduces an extension of the MS-IS hypoplastic model designed to enhance the predictions during non-monotonic loading conditions. Addressing the limitations observed in the hypoplastic models during the unloading and reloading phases, the proposed model incorporates an additional stiffness feature. This new stiffness function is integrated into the foundational framework to enhance the model's overall stiffness response. For the unloading phase, the introduction of a stiffness degradation factor aims to modify the volumetric response and account for the realistic stiffness degradation. Additionally, for the reloading phase, stiffness is now a function of the mean effective stress. The novel model's performance is validated against experimental data, encompassing diverse loading and boundary conditions.

The accuracy of the constitutive model affects the precision of the finite element analysis. Four advanced sand constitutive models-the DM04 model, the SANISAND-MS model with memory surface (MS), the SANISAND-Sf model with semi-fluidized state (Sf), and the SANISAND-MSf model with both memory surface and semi-fluidized state-are examined in this article based on a comparison of simulation and experiment results of element tests. First, four constitutive models are implemented in OpenSees, and the constitutive models are calibrated based on the cyclic loading experiments of Karlsruhe fine sand. After that, the advantages and disadvantages of four constitutive models under different test conditions are analyzed. Finally, the finite element model of the LEAP-UCD-2017 (Liquefaction Experiments and Analysis Project, University of California Davis, 2017) centrifuge test is established to evaluate the performance of the four constitutive models for solving boundary value problems. It is found that the SANISAND-MSf model can well reproduce the undrained cyclic properties.

This study assesses the performance of a memory surface constitutive model (SANISAND-MS) in capturing vertical cyclic loading on a suction bucket foundation in sand. The model has been calibrated against drained cyclic triaxial responses and validated against corresponding centrifuge experiments on suction buckets. The model was found to satisfactorily capture the effects of increasing accumulated strain with increasing mean stress level and reducing density. The performance of the model was further investigated through a parametric study on suction buckets at different mean stress levels, densities and loading sequences. The insights gained from investigating the strain and stress responses, along with the movement of the memory surface, reveal that the model can satisfactorily capture the strain accumulation and ratcheting effects under different load histories.