Micromechanical Investigation of Sands Using Machine Learning Methods

Description

Artificial intelligence is transforming human society at a phenomenal speed. Its profound impact is attributable to the application of machine learning, the core technology of artificial intelligence, to many areas of scientific and engineering research. However, there have to date been few machine learning-based investigations of the micromechanics of granular materials. Accordingly, the aim of this proposed research is to develop machine-learning methodology for investigating the macro-micro-mechanics of granular soils. The microcomputed tomography data of inter-particle contact and particle morphology within a sheared sand specimen will be combined with discrete element simulation data of inter-particle contact force to form a comprehensive database for the training of two artificial neural networks. The first artificial neural network will use microcomputed tomography data to predict the inter-particle contact forces between sand particles, and the second artificial neural network will use these predicted inter-particle contact forces and microcomputed tomography data to predict the macroscopic mechanical properties of a sheared sand specimen. There is currently no optimal experimental technique to measure inter-particle contact forces within an assembly of real sand particles, and thus this proposed methodology is highly innovative and will yield new tools for understanding and solving the long-standing and complex macro-micro granular mechanics problem. This project will have high impacts on basic and applied research in a variety of fields such as geomechanics, granular materials, powder technology, geophysics, geotechnical engineering, engineering geology and petroleum and mining engineering, concomitant with great economic and societal impacts. The research impacts will primarily derive from the newly established machine-learning tool, which will enable accurate, efficient and robust predictions of the multiscale mechanical properties of granular soils to be made. These predictions will lead to new scientific discoveries and inventions. The economic and societal impacts will derive from the use of the machine learning tool to generate data for geotechnical engineering activities such as site investigation, design, construction and maintenance of infrastructures, risk monitoring and hazard mitigation, or manufacturing, handling and processing of granular materials in relevant industries. This will catalyze the development of optimized engineering activities and processes, subsequently leading to improvements in civil infrastructure, construction management, hazard warning and mitigation systems, and to novel and inexpensive granular materials.