Machine learning based prediction of piezoelectric energy harvesting from wake galloping

Wake galloping is a phenomenon of aerodynamic instability and has vast potential in energy harvesting. This paper investigates the vibration response of wake galloping piezoelectric energy harvesters (WGPEHs) in different configurations. In the proposed system, a stationary obstacle is placed upstream, and a cuboid bluff body mounted on a cantilever beam with piezoelectric sheets attached to it, is placed downstream. Three different types of WGPEHs were tested with different cross-section S^∗ of the upstream obstacles, namely square, triangular, and circular. At the same time, the tests were conducted by changing the equivalent diameter ratio η=1~2.5 of the upstream and downstream objects, the dimensionless distance between two objects’ centers L^∗=L/D=2~8, and the velocity span U^∗=2.93~14.54. The results reveal that S^∗, η, L^∗ and U^∗ have significant effect on the vibration response of WGPEHs. Then, considering these four parameters as input features, this study has trained machine learning (ML) models to predict the root mean square values of the voltage (V_rms) and the maximum displacement (y_max), respectively. The performance of three different ML algorithms including decision tree regressor (DTR), random forest (RF), and gradient boosting regression trees (GBRT) on predicting V_rms and y_max were compared. Among them, the GBRT model performed optimally in predicting the V_rms and y_max. The GBRT model provides accurate predictions to V_rms and y_max within the test range of S^∗, η, L^∗ and U^∗.