Can physics-informed deep learning enhance flow-field data reconstruction? This study explores the application of physics-informed neural networks (PINNs) for super-resolution of flow-field data from noisy measurements. The goal is to obtain a physically-consistent prediction without high-resolution reference data. PINNs are applied to Burgers’ equation, vortex shedding behind a cylinder, and turbulent channel flow, demonstrating their ability to improve resolution and reduce noise. The results show the potential of PINNs in data augmentation for experiments in fluid mechanics, presenting a promising approach for researchers dealing with incomplete or noisy data.
Published in Measurement Science and Technology, which focuses on advancements in measurement techniques and instrumentation, this research aligns with the journal's emphasis on innovative methods for data analysis and interpretation. By showcasing the capabilities of physics-informed deep learning for flow-field data reconstruction, the paper offers a valuable tool for experimental fluid mechanics and contributes to the journal's mission.