Optimization of a Centrifugal Blower with Full-3D Backward-Curved Blades using HEEDS

3.5 Automatized Aerodynamic Design via Optimization

In the present study, a centrifugal blower with full-3D backward-curved blades was designed and optimized by combining CAD and CFD programs with Optimization tool. At first, about fifty variables such as diameter, height and incidence angle required to design the blower were defined and initialized, and then, NX UG generated the blower and its domains appropriate for numerical simulation according to them. In order to obtain the full-3D blades, four horizontal surfaces that contain airfoil-shaped curves which are different from each others were configured in span-wise direction. ANSYS CFX known as the best commercial CFD program for rotating machinery simulated the steady-state flow structure around the blower under the operating condition and estimated its static efficiency based on the increase of the static pressure and torque acting on the blower.
These results were transferred to HEEDS, the optimization tool, and were treated as objective variable or constraint. Once the procedure till the numerical simulation was finished, SHERPA, a basic artificial intelligence algorithm integrated in HEEDS, searched a new design variable set corresponding to the optimum by integrating several well-known optimization algorithms such as Genetic Algorithm and Response Surface Method and suggested a design point according to the previous results. Tens or hundreds of iterations described before were repeated in a single stage by controlling the design variables and constraints and the optimization was terminated after seven stages.
According to the results from simulations, the static efficiency increased up to 3%p relative to the initial design even though the outer diameter of the blower was reduced about 5%. The size of separation at the leading edge of the blade was reduced and the strength of separation vortex near the shroud was also decreased. However, when the efficiency was measured in accordance with AMCA standard, it was shown that its performance index was rarely improved. It was thought that the constraints on the computational domains in order to reduce the computation time resulted in this disagreement between the predicted and measured results. Although the efficiency was the same as before, broadband noise from the blower was significantly suppressed. The sound pressure level measured in the semi-anechoic chamber according to KS standard decreased up to 6 dB relative to the initial design.