Application of a Multi-Block CFD Code to Investigate the Impact of Geometry Modeling on Centrifugal Compressor Flow Field Predictions

Abstract

CFD codes capable of utilizing multi-block grids provide capability to analyze the complete geometry of centrifugal compressors including, among others, multiple splitter rows, tip clearance, blunt trailing edges, fillets, and slots between moving and stationary surfaces. Attendant with this increased capability is potentially increased grid setup time and more computational overhead — CPU time and memory requirements — with the resultant increase in “wall clock” time to obtain a solution. If the increase in “difficulty” of obtaining a solution significantly improves the solution from that obtained by modeling the features of the tip clearance flow or the typical bluntness of a centrifugal compressor’s trailing edge, then the additional burden is worthwhile. However, if the additional information obtained is of marginal use then modeling of certain features of the geometry may provide reasonable solutions for designers to make comparative choices when pursuing a new design. In this spirit a sequence of grids were generated to study the relative importance of modeling versus detailed gridding of the tip gap and blunt trailing edge regions of the NASA large low speed centrifugal compressor for which there is considerable detailed internal laser anemometry data available for comparison. The results indicate: 1) There is no significant difference in predicted tip clearance mass flow rate whether the tip gap is gridded or modeled. 2) Gridding rather than modeling the trailing edge results in better predictions of some flow details downstream of the impeller, but otherwise appears to offer no great benefits. 3) The pitchwise variation of absolute flow angle decreases rapidly up to 8% impeller radius ratio and much more slowly thereafter. Although some improvements in prediction of flow field details are realized as a result of analyzing the actual geometry there is no clear consensus that any of the grids investigated produced superior results in every case when compared to the measurements. However, if a multi-block code is available it should be used as it has the propensity for enabling better predictions than a single block code which requires modeling of certain geometry features. If a single block code must be used some guidance is offered for modeling those geometry features which can’t be directly gridded.