Sparse Sensing and Surface Pressure Field Reconstruction for Flow Over a Hemisphere

Reconstructing detailed flow fields from sparse sensor data is a critical challenge in aerodynamic studies. This work evaluates the performance of several reconstruction methods, including Proper Orthogonal Decomposition (POD), Linear Stochastic Estimation (LSE), a combined LSE-POD method, and a method based on Shallow Decoder Networks (SDN). These methods were tested to reconstruct surface pressure fields around a hemispherical turret. The study incorporates data-driven sensor placement using QR factorization to optimize sensor locations and compares results to randomly placed sensors. The reconstruction methods were validated using high-resolution experimental pressure fluctuation data obtained from a hemispherical model in subsonic flow. Results demonstrate that the LSE method achieves the highest accuracy among the methods tested, while the SDN-based method proved to be the most robust to noise and to varying flow speed conditions. QR-based sensor placement was shown to improve reconstruction performance for all methods, particularly in regions with sparse dynamic activity.