Study on rapid prediction of flow field in a knudsen compressor based on multi-fidelity reduced-order models

The safe and stable operation of a hydrogen Knudsen compressor is essential for transporting hydrogen in microfluidic systems. This study uses proper orthogonal decomposition to identify the coherent structures within the hydrogen flow field during non-equilibrium evolution. A long short-term memory neural network is then used to create a multi-fidelity reduced-order model, connecting two-dimensional and three-dimensional data to uncover transient flow mechanisms and enable rapid flow field prediction. The results show that the coherent structures of hydrogen flow, representing the most energetic modes, retain 99% of the flow energy and significantly influence the evolution of Poiseuille and thermal transpiration flows during non-equilibrium processes. The multi-fidelity reduced-order model effectively captures hydrogen transient flow and instabilities at various stages, achieving a 99.4% reduction in computational time while maintaining a maximum relative error of 0.53%. This approach facilitates the rapid prediction and control of flow states during hydrogen transport.