Aerodynamic Optimization of a Two-stage Centrifugal Compressor for a 1MW-class High-temperature Steam-generating Heat Pump using Reactive Response Surface Machine Learning

Abstract

This study presents the optimal design of a 1MW-class two-stage centrifugal compressor for high-temperature steam generation heat pump using an AI-assisted design framework based on reactive response surface machine learning (RRSML). The baseline configuration was obtained using an inverse design method, where the compressor flow was first determined by meanline and meridional design, and the three-dimensional blade geometry was subsequently generated by prescribing blade loading distributions and spanwise stacking conditions. High-fidelity CFD simulations were combined with RRSML-based Kriging surrogate model and adaptive sampling, enabling efficient exploration of the design space and rapid identification of optimal aerodynamic configurations. The optimized compressor achieved an overall efficiency of 85.3% while maintaining a surge margin of 38%, indicating both high performance and robust operational stability. These results confirmed that the AI-driven RRSML optimization framework provides an effective design methodology for designing high-performance centrifugal compressor for high-temperature steam generation systems.