Perpustakaan Digital ITB

This study focuses on the application of the ?-variational autoencoder (?- VAE), a deep learning architecture, alongside the classical method of proper orthogonal decomposition (POD) for modal decomposition of flow over an oscillating cylinder induced by vortex-induced vibration (VIV). Data are obtained from numerical simulations using the immersed boundary-lattice Boltzmann method (IB-LBM). The cases are limited to a Reynolds number of 150, a mass ratio of 5.1, and six cases of reduced velocity variations: 3, 4, 5, 6, 7, and 8. Results show that ?-VAE can achieves the same reconstruction accuracy as POD using fewer modes. The temporal evolution of ?-VAE modes exhibits near-orthogonal behavior. Furthermore, in simpler cases, the spatial structures and dominant frequencies of ?-VAE modes closely match those derived from POD. The first mode pair from both methods also shares a dominant frequency that aligns with the vortex shedding frequency.