Accurate & efficient machine-soil modeling
Modeling granular materials interacting with machines typically involves a tradeoff between fidelity and computational efficiency. Discrete Element Method (DEM) models soil as a collection of millions of individual particles, while classical terramechanics equations offer simple empirically-based approximations. Our research has extended continuum methods, specifically Material Point Method (MPM) with nonlocal granular fluidity, to 3D simulations of wheel and excavator interactions to produce a balance of accuracy and efficiency. Further, we have found that these methods capture the important effects of gravity on granular interactions. We have also trained significantly accelerated models by combining dimensionality-reduction techniques and Graph Neural Networks (GNNs), using experimentally-validating MPM datasets to train these machine learning approaches.
Related Papers
Concordia University's Aerospace Robotics Lab
Research interests include: Space robotics, Planetary rovers, Robot mobility, Vehicle-terrain interactions, Advanced 3D printing techniques, Robotics excavation & construction, Reduced gravity experimentation, Computer vision and machine learning for robotics applications.