Abstract

A novel AI‑driven tool converts deflection test results into key soil parameters required for compaction QA, including density and void ratio. Determining these properties in unsaturated soils is challenging due to the interplay among suction, moisture content, void ratio, and measured deflections. Our approach integrates unsaturated soil mechanics with advanced AI—particularly reinforcement learning—to fuse in‑situ sensing, experimental observations, and physics‑based finite‑element modeling. The system adapts online to changing field conditions, enabling real‑time, accurate inference without directly measuring moisture content or suction.

Animation guide: The RL loop cycles through Measurement → Estimation → Learning. The left grid highlights the active test location; the violin panel updates the inferred void‑ratio distribution for that location. Status chips below indicate the current phase.

Publications & Repositories

Technologies