Authors: Shivakumar Balachandran – University of South Carolina
Title: Numerical and Machine Learning Models for the Study of Dam Break Waves
Abstract: A 1D or 2D unsteady computational modeling technique is often adopted to study the complex flow conditions in a river such as the flow interactions with hydraulic structure, flood flows, and extreme events such as dam failure. In this study, 1D and 2D explicit finite difference models are developed to simulate the dam break flow. A McCormack explicit finite difference scheme of second-order accuracy is used to discretize the 1D and 2D shallow water equation. The model is validated with the observations obtained from the literature for a small-scale laboratory test. It has been observed that the numerical model predicts the flow pattern of the experimental results with a precise overlapping of wave arrival time and water trough height. Also, the wave-front height predicted by the numerical model is closer to the experimental value but attenuated from the actual peak. The study also identified the constraint in implementing the MacCormack scheme for the numerical modeling of dam break flow with respect to the ratio of water heads upstream and downstream of the reservoir as approximately 0.3. A simple data-driven model is developed for the estimation of dam break wave characteristics within the range of the identified head ratio limit. The model is trained to identify the optimal input requirements for accurate predictions of dam break waves. The flow characteristics i.e., upstream and downstream head, and the channel characteristics i.e., bed slope and flow resistance are considered as the input variables for the model studies. The wave front height, water trough height, and the wave arrival time of the dam break are considered output variables. This machine learning model is trained using data sets obtained from the 1D numerical model studies. The study needs to be extended to large-scale laboratory tests and real-time problems to quantify the computational efficiency.