Data-driven model for soil-object interaction with large plastic deformation

Soil undergoes large deformation in many soil-object interactions, such as indentation and ploughing. Modeling these processes presents challenges due to the large and predominantly plastic deformation and the associated free surface evolution. This paper proposes a data-driven method to model these processes efficiently. In this method, neural network models are employed to predict force-displacement relationships throughout the deformation process, accounting for the evolution of the free surface. Proper orthogonal decomposition (POD) facilitates the low-dimensional representation of surface geometries. The accuracy and efficiency of the method are demonstrated by applying the approach to three examples: wedge indentation, cylinder indentation, and plate ploughing. By considering various object geometries, soil parameters, and motions, the study also showcases the method's potential for application across a wide range of problems characterized by soil-object interaction.