Perpustakaan Digital ITB

Fluid flows are at the core of many applications and to model them well with minimal complexity, you need to capture nonlinear dynamics effectively, which is important for areas ranging from turbulence modeling, flow control to fluid system optimization. Using traditional techniques to identify such dynamics is expensive and cumbersome, often requiring high-fidelity simulations to capture the salient features of the fluid flow. In this thesis, a hybrid approach is used. Proper Orthogonal Decomposition (POD) and Sparse Identification of Nonlinear Dynamics (SINDy) are combined to identify the dominant nonlinear dynamics of fluid flows effectively. POD is used to reduce dimensionality by extracting a specific set of orthogonal modes from high-dimensional fluid flow data that describe the most dominant spatial and temporal structures. These reduced-order modes provide the basis for a simplified model that captures the essential flow physics. The reduced-order data are then processed through the SINDy method, which relies on sparsity-promoting procedures to compute these minimal nonlinear differential equations, which determine the temporal patterns of the POD mode amplitudes. By targeting a sparse representation of the underlying dynamics, the method reveals the physical processes that allow the flow to retain its original design while still making the model computationally efficient and interpretable. To demonstrate the effectiveness of reproducing the flow dynamics for a fluid problem, The POD-SINDy method applied to the flow over three tandem equilateral triangle cylinders and the classic NACA0012 airfoil to demonstrate that it can accurately reproduce the nonlinearities of the flow. These models not only highlight the essential nonlinear interactions of what is happening in the models, but also provide new insights for further research into fluid dynamics as primitive processes. Combined with POD and SINDy, this approach forms a powerful method for flow prediction, control, and predictive turbulence modeling that can be applied in both fundamental research and engineering systems.