A rapid and cost-effective optimization framework for supercritical CO2 centrifugal compressor blades using a hybrid prediction-CFD approach

The centrifugal compressor constitutes a core element of the supercritical carbon dioxide (SCO2) Brayton cycle, and its blade design is crucial for system efficiency and operational stability. Currently, traditional blade optimization methods using large-scale computational fluid dynamics (CFD) simulation have high costs and long single-point calculation times, limiting practical use. This research developed an efficient framework for blade optimization. It first creates a detailed aerodynamic analysis program, based on a loss model, to quickly predict blade aerodynamic performance. Next, we analyzed that the relative error of the program in predicting large-scale CFD samples follows a normal distribution. We mathematically proved that using small-scale CFD samples can effectively calibrate the program's prediction results. Applying this error calibration strategy to prediction results can quickly construct a high fidelity surrogate model training dataset. The framework cuts the time to build 100 sample datasets from 43 days using CFD to only 3 days. The dataset accelerates blade optimization, with average daily optimization rates for isentropic efficiency and total pressure ratio being 4.9 and 11.6 times higher, respectively, than those achieved with large-scale CFD. This research offers a practical, fast, and low-cost solution for optimizing SCO2 centrifugal compressor blades.