Investigation on the Active Learning Optimization on Rotor Dynamics of SCO2 Turbine

Abstract

The supercritical carbon dioxide (SCO2) Brayton cycle has the characteristics of high power density and high thermal efficiency, which is an important development direction of the micro power plant. SCO2 turbine is the core component of the SCO2 Brayton cycle of which the dynamics have important influences on operational reliability of the entire system. With regard to the rotor of SCO2 turbine, the equations of motion is established by adopting finite element method. The complex eigenvalues of the rotor are solved in the state space, and the campbell diagram has been drawn to obtain the critical speeds. The steady state response of the rotor is obtained by the harmonic balance method, and the safety of the system is estimated based on API684. Results show that the rotor is not safe with its original geometric paramters. Aiming to improve the operational safety, an optimization method based on active learning is developed to maximize the separation margin. Results show that after the optimization, the separation margin has been greatly increased. Comparing with the genetic algorithm (GA) and the parttern search (PS) method, the iteration number of the active learning optimization method has been greatly reduced. The effectiveness of the developed optimization method is proved, and the study method and conclusions can serve as a reference to the optimization of SCO2 turbine rotors in the industry.