Exploring bluff body geometries for enhanced energy harvesting from flow-induced vibrations using machine learning
Shohreh Jalali, E. Barati, Amir Sarviha
Abstract
This study investigates energy harvesting from flow-induced vibrations using various bluff body geometries, combining experimental techniques and machine learning for performance analysis. An electromagnetic energy harvester, featuring a permanent magnet in motion within a coil and coupled to a flexible diaphragm, was used to extract energy from vortex-induced vibrations in a flow channel. The study expands prior research by evaluating Circle, Square (at 0, 22.5, and 45 degrees), Rectangle, Trapezoid (small and large cases), and Diamond geometries across Reynolds numbers ( Re = 3000, 4000, and 5000). A key innovation lies in applying six advanced machine learning models—Decision Tree, Random Forest, XGBoost, Gradient Boosting, CatBoost, and LightGBM—for voltage prediction, with a novel Weighted Ensemble method demonstrating exceptional accuracy (MAE: 0.1540, MSE: 0.0459, RMSE: 0.2141, R²: 0.9336). Experimental results revealed that Diamond and Circle geometries achieved superior energy outputs of 3.8 and 2.6 units at Re = 5000, while Trapezoid (large case) and Square at 45 degrees performed optimally at Re = 4000. This work enhances understanding of flow-induced energy harvesting, offering comprehensive insights into optimizing harvester designs through a synergy of experimental validation and machine learning predictions.