Fan 2022 conference - Senlis (France) 6-8 April 2022 - International Conference on Fan Noise, Aerodynamics, Applications and Systems

Technical Program

< Go back to the papers' list


Aerodynamic Optimization of a Low-Pressure Axial Fan using Adjoint Computational Fluid Dynamics


A3 Analytical, Numerical & Experimental Methods I


University of Siegen

Siegen - Germany

Steinbeis-Transferzentrum Strömungstechnik und Strömungsmaschinen

Netphen - Germany


The presentation discusses the aerodynamic optimization of an axial fan using adjoint Computational Fluid Dynamics (CFD). As an example it is tried to improve the aerodynamic performance of the University-of-Siegen reference low pressure, rotor-only axial fan USI7. The existing OpenFOAM solver "adjointShapeOptimizationFoam" is customized since the original code does not account for rotating frames of reference. Moreover, a new boundary condition is implemented since in adjoint simulations, the case-specific objective function has an impact on the boundary condition. Together, the solutions of the primal and adjoint simulations yield a sensitivity map which indicates weather material should be added or removed in each cell of the numerical grid. This information is interpreted to alter the fan geometry. The new geometry is then again simulated by the adjoint code and further improved. At the fourth iteration, no further improvement can be observed. The overall CFD-predicted improvements are an increase of the total-to-static efficiency at the fan design point of + 0.5 percentage points and a delay of the stall point towards smaller volume flow rates.
The obtained new fan rotor (USI8) is manufactured via CNC-milling with high accuracy. Its performance characteristics are determined experimentally on a chamber test rig. The experiments confirm the CFD-predicted improvement at the design point, the delay of the stall point, however, cannot be validated.
The improvement of efficiency as such proves that the adjoint method is working. The fact that only four relatively inexpensive RANS simulations were necessary to obtain an improvement confirms the extremely high performance of the adjoint method as compared to other optimization methods. However, it must also be acknowledged that an improvement of 0.5 percentage points of total-to-static efficiency is very small. Reasons such as a non-optimal interpretation of the adjoint results, uncertainties of the CFD model (especially in the near stall region) or simply the too good reference fan USI7 without significant potential for improvement are addressed.