Influence of latent space dimension on flow field reconstruction from limited number of sensors

Unsteady flow field reconstruction from sparse measurements is a critical yet challenging task in many engineering applications. This work demonstrates that the latent space dimension, which is typically not fully explored, has a significant impact on the reconstruction quality. This effect is studied across three well-established reconstruction methods that leverage dimensionality reduction. The reconstruction is performed on CFD datasets of the wake of an oscillating cylinder undergoing varying imposed motions of increasing complexity. This approach sheds light on how the choice of training data influences the performance of model-based reconstruction tools.