Visualization technique for the deformation distributions in transparent synthetic soil with object detection

Damage caused by land subsidence due to excess pumping has become a serious problem in coastal major cities and landscapes around the world. To prevent this damage, it is necessary to predict local deformation in the ground and design the appropriate pumping amount. To verify the predicted deformation from land subsidence modeling, experimental methods with visualizing deformation distribution is of importance.
A visualization method using transparent synthetic soil (TSS) as a physical model of soil behavior has been developed in the field of soil mechanics. This experimental method simulates the geotechnical properties of natural soil using a transparent surrogate containing a transparent porous medium and pore fluid. In this study, the authors performed a tank experiment using a TSS made of polymers which is inexpensive and easy to control.
In the previous experimental study by the authors, a pumping test was carried out in an acrylic tank measuring 300 mm wide x 250 mm long x 249 mm high, filled with a transparent hydrated polymer to represent an aquitard (clay layer) above an aquifer (saturated silica sand). Using the target racking method, 100 particles with a diameter of 3 mm were submerged in the synthetic clay layer, and the subsidence in the synthetic clay layer caused by the pumping of pore water in the silica sand was constantly monitored. 
In this study, an AI-based object detection method was used to more quantitatively visualize the spatiotemporal distribution of deformation inside the TSS caused by the propagation of pore water pressure change in the TSS after pumping was stopped. It successfully revealed the three dimensional elastoplastic deformation distribution. The developed methods and the obtained results are expected to contribute to a better understanding of land subsidence mechanisms and verify the numerical land subsidence modeling.