Centrifugal Compressor Stability Prediction Using a New Physics Based Approach

Abstract

The accurate prediction of centrifugal compressor stability continues to be an important area of interest in the oil and gas industries. Ensuring stability is critical to the cost-effective installation and operation of these machines in remote environments, where field stability problems are much more expensive to diagnose and correct. Current industry standards and tools for the prediction of impeller destabilizing forces are based on empirical methods that, to date, have served fairly well for systems with reasonable stability margins. However, as stability margins are decreased, use of a modeling method that is more physics based and can better represent the observed trends in machine behavior at low stability margins is required. Furthermore, the development of megaclass liquefied natural gas (LNG) compressors and ultra-high pressure re-injection compressors provides further motivation to improve accuracy. In this paper, a new physics based expression for the prediction of impeller cross-coupling, previously described by (“Rotordynamic Force Prediction of Centrifugal Compressor Impellers Using Computational Fluid Dynamics,” ASME Paper No. GT2007-28181), is further investigated by analyzing several classes and scale factors of impellers ranging from 2D designs used in re-injection to full 3D impellers typically used in LNG. The new expression is based on both computational fluid dynamics simulation and experimental test data from a known instability. These results are then applied to two case studies of marginally stable and unstable compressors in the field that were studied by the authors’ company. For each case study, the system stability is evaluated using both the new physics based expression as well as the more traditional empirical approaches. Comparisons are made for overall stability prediction as well as sensitivity to system changes. Conclusions are made regarding the applicability and limits of this new approach.