Probabilistic Machine Learning and Uncertainty Quantification

CGG conducts research on uncertainty quantification and machine learning from the probabilistic perspective to characterize uncertainty inherent in each participating source of evidence, and to study how the randomness associated with model parameters influences the model predictions. The broad applicability of this framework has allowed us to conduct several collaborative research on topics of rheological property of sandstone, soil erodibility, etc. 

Micromechanics of Granular Material

Strain localization in sand is a ubiquitous process associated with non-homogeneous deformation occurred to material when subjected to compressive or tensile stress. In CGG, we provide the complete set of first order 3D kinematic operators under curvilinear coordinates that are consistent with specimen geometry and suitable to characterize meso-scale kinematics comprising translational, rotational and volumetric behaviours throughout triaxial compression process. Results provide insight into the overall deformation modes and inherent uncertainties, as well as spatio-temporal correlation patterns of different banding phenomena undergoing three-dimensional stress conditions.

Remote Sensing and Computational Applications

Remote sensing has emerged as a powerful technique for acquiring geotechnical and geological characteristics in a spatial-temporal manner. In CGG, we use satellite images, airborne LiDAR and related data to help solve scientific research problems related to geomorphologic change detection, geohazard assessment, site responses, and climate change topics. Our goal is to combine cutting-off sensing technology, geological/geotechnical information, and computational power to develop, test, and deliver robust solutions to geotechnical/geosciences communities. 

Risk/Reliability Assessment and Management

The streamline of engineering design evolved from Factor of Safety to Reliability Index, and about to the risk-informed design & management in the foreseeable future. In CGG, we focus on using datasets consisting of aerial photography, geology, soils, hydrologic information,  LiDAR (point clouds and DEMS), and infrastructure and residential information, to create machine learning models for geo-hazard analysis and test the probability of failure under different geologic and hydrologic conditions. The results will identify vulnerable areas and the potential impact on the built environment, all of which will be incorporated into mitigation strategies that represent a giant step forward in making community resilient.