Abstract
Centrifugal compressors are suited for high pressure applications. Generally, many centrifugal compressor stages are stacked together in order to guarantee the overall pressure ratio implying larger machines. With the aim of more compact machines, it is possible to design high pressure ratio stages, thus reducing the total number of components. In this case, the impeller rotates at high speed to obtain an adequate work input. That in turn implies that the flow exits the impeller and enters the following diffuser with a large tangential velocity.
Vaned diffusers are thus more effective in this context to obtain a high-pressure recovery keeping the widest operating range. Ideally, fixed diffuser vanes should not have any gap between the tip of the blade and the casing however, due to assembly tolerances, the diffuser has a clearance impacting performance.
On these premises, this paper presents the optimization of a vaned diffuser exploiting artificial neural networks as a surrogate model. The maximum clearance value allowed by the manufacturing tolerances is taken into account during the optimization. The optimized geometry shows a large improvement in the diffuser performance ensuring an extended operating range. Moreover, the optimized diffuser is more gap insensitive, since the performance lay in a narrow range of variability demonstrating an improved robustness with respect to the clearance values. The performance of the optimized geometry is assessed within a preliminary experimental campaign carried out on the compressor stage. Finally, the developed design rules are validated by means of an extensive experimental campaign on a full-scale model test.
