Predicting the performance of L-shaped confined flapping-foil energy harvester: A deep learning approach

Flapping-foil energy harvesters offer significant potential for clean energy generation but suffer from performance limitations within partially confined straight-wall setups. This study introduces a novel design where the flapping foil is partially confined between L-shaped walls to enhance energy extraction. Numerical simulations showed that the proposed configuration can significantly enhance power output and efficiency, by up to 50% over an unconfined foil. These enhancements are attributed to vortices generated at the L-shaped extrusions, which accelerate the incoming fluid flow and increase the pressure differential across the foil's upper and lower surfaces, enhancing lift and power output. Additionally, to account for changes in power generation due to the variations in real-world fluid flows resulting from site-specific designs and off-design operations, we developed a Temporal Dynamics Long Short-Term Memory model tailored to the proposed L-shaped configuration. Trained with data from extensive simulations across a range of Reynolds numbers (50,000-500,000) and wall distances (3c-7c), c -7 c ), this deep learning approach could rapidly predict aerodynamic coefficients and power-generation performance with high accuracy (relative error of +/- 2%), even under unseen conditions within the specified ranges. This capability enables faster and more robust analysis under realistic, time-varying flow conditions. By surpassing traditional numerical simulations in adaptability and computational efficiency, this novel deep learning-based technique marks a significant advancement towards developing more efficient and adaptable energy-harvesting systems.