An online Data Assimilation (DA) strategy based on the Ensemble Kalman Filter (EnKF) is used to obtain an augmented prediction based on Large Eddy Simulation (LES) for the analysis of the oscillating compressible flow in a flow rig type geometry (OFR). The test case investigated, which is based on Dellenback flow expansion geometry, includes an open valve in its intake to mimic the internal combustion engine configurations. This kind of flow, which is not permanent owing to the trigonometric behavior of the mass flow rate at the inlet, exhibits difficulties for analyses relying on numerical simulation. In fact, capturing the emergence of instantaneous extreme events is an additional challenge to the accurate prediction of statistical moments of the velocity field. In this work the numerical LES solvers are augmented via an online DA procedure, which relies on the platform CONES to perform on-the-fly physical state updates and parametric optimization of the LES model. The EnKF is also improved via the use of hyper-localization, allowing faster and more stable calculations. The algorithm sequentially combines a coarse LES (LES$_c$) prediction with high-fidelity, sparse instantaneous data obtained from a fine LES (LES$_f$). It is shown that the procedure provides an augmented state that exhibits higher accuracy than the LES$_c$ model and it synchronizes with the time evolution of the high-fidelity LES$_f$ data if the hyperparameters governing the EnKF are properly chosen.