This paper presents a methodology to control flange to flange performance prediction of centrifugal compressors using a probabilistic approach. In order to have reliable prediction for the performance of centrifugal compressors, a thorough knowledge of critical parameters contributing to the deviation and an efficient way to control the variation of these parameters becomes necessary. This paper discusses about a robust methodology for identifying and controlling the variation of these parameters and hence the predicted performance. This probabilistic technique involves a Design of Experiments (DoE) study to handle large number of input parameters, sensitivity study to identify critical ones and a Monte-Carlo based approach to identify the uncertainty in flange to flange performance. This approach takes into consideration the compressor stage performance variability driven by impeller manufacturing tolerances, statoric component losses variability and leakages variability in order to compute overall performance variation in a compressor. An in-house developed probabilistic optimization code (PEZ) is interfaced with a well-validated & calibrated thermodynamic tool to analyse large sets of possible combinations and to provide best possible solution for a given design space. This concept is successfully applied for different compressor configurations by varying the stage numbers and process conditions. The results give an insight on the main sources and magnitude of variations on compressor performance, thus enabling to control the predictions in an efficient way.
This methodology will provide a novel and an efficient way to generate robust compressor performance, where it will be possible to take into account design and manufacturing uncertainty. The use of this methodology can thus drastically improve the performance predictability and risk associated with each compressor selection.