Rediscovering the Tesla Valve: Geometric Optimization through Machine Learning

A study into the Tesla Valve’s design was conducted across a large parameter set for optimization of the valve’s use as a pressure drop device. A single Tesla Valve was geometrically parameterized, and an automated design of experiments was created to cycle through a wide selection of geometric parameters. The geometric parameters selected each influenced the amount of flow segregated into the arms of the Tesla Valve, as well as the divergence and convergence angles of the arm flow. Data was collected from completed computational fluid dynamics simulations across all geometric parameter combinations. Tesla valve designs were assessed in the restricted flow direction for overall differential pressure and overall minimum pressure to avoid onset of cavitation. Machine Learning analysis resulted in a robust model across a wide range of Reynolds numbers. The minimum simulation pressure demonstrated the required system pressure to prevent onset of cavitation in the flow field. Qualitative observations of overall behavior of a single Tesla Valve were made, identifying critical parameters to the design such as the symmetry ratio and passage diameter ratio. An optimal design within the parameter set was identified maximizing pressure differential while meeting the required system pressure. The techniques utilized demonstrate a tool in the study of various complex geometries within Fluid Dynamics, as the possibility of more complex designs are now achievable through modern Additive Manufacturing techniques.