Author(s): Ahmed Emara; Sameh A. Kantoush; Mohamed Saber; Tetsuya Sumi; Emad Mabrouk
Linked Author(s): Tetsuya Sumi
Keywords: Sediment bypass tunnels; Spatial abrasion of SBTs; Abrasion mapping; XGBoost algorithms; Abrasion pattern predicting
Abstract: Mitigating reservoir sedimentation can be achieved through Sediment Bypass Tunnels (SBTs), which divert sediment-laden flows around dams to downstream river reaches. These tunnels handle high-velocity floods, but such velocities cause significant abrasion damage to the tunnel floors, risking SBT operations for years. Predicting this abrasion is challenging due to the complex interaction between flow hydraulics and sediment transport, and limited high-quality data. This study maps abrasion using XGBoost Machine Learning (ML) algorithms, developing the 2D Abrasion Susceptibility Model (ASM) and 3D Abrasion Susceptibility Depth Model (ASDM). Koshibu SBT in Japan, approximately 4 km long, was the case study. Laser scanning tools measured spatial abrasion topography with a 2 cm resolution. XGBoost algorithms, trained on over half a million data points, effectively predicted 2D ASM and 3D ASDM models with an accuracy of 0.864 and a correlation coefficient (R) of 0.861, respectively, highlighting the potential of ML in predicting tunnel abrasion.
Year: 2024