Data-driven parametric modelling of split-Hopkinson pressure bar tests on cohesive soils

Abstract

Soil-filled wire and geotextile gabions stand as vital bulwarks in military bases, harnessing soil’s innate capacity to absorb shock and safeguard both personnel and critical assets from blast and fragmentation effects. Yet, the dynamic response of cohesive soils under extreme loads remains largely unexplored, leaving engineers grappling with a significant void in knowledge as they strive to fortify structures against emerging threats. This paper considers the high-strain-rate behaviour of kaolin clay using the split Hopkinson pressure bar in both confined and unconfined configurations, with a range of moisture contents representing dry, partially-saturated and saturated conditions. Analysis of the results indicates distinct phase behaviours in transmitted and radial stress based on strain rate, moisture content and confinement. Leveraging cutting-edge machine learning models such as the Proper Orthogonal Decomposition (POD) and sparse Proper Generalised Decomposition (sPGD), data-driven parametric models were developed based on the experimental data. These models enable the prediction of cohesive soil behaviour at specified strain rate and moisture content, enabling engineers to rapidly predict soil behaviour in response to new threats and ground conditions.