Surrogate model based multi-objective optimisation of supercritical CO2 ejectors

Abstract

A supercritical CO2 (sCO2) supersonic ejector improves the coefficient of performance (COP) in combined power and cooling systems by compressing a secondary stream through entrainment and mixing with a high-momentum primary stream. The performance of the ejector is crucial to system efficiency and is influenced by complex gas dynamic shock interactions and shear layers which is further complicated by the rapid variations in thermophysical of sCO2. In this regard, aerodynamic duct shaping plays a pivotal role in optimizing ejector efficiency. The present paper seeks to optimize a sCO2 ejector using a surrogate model derived from computational fluid dynamics data. The model relies on a comprehensive dataset generated using a simulation tool coupled with REFPROP database to account for variations in thermophysical properties of sCO2. Subsequently, a genetic aggregation technique is used to train and improve the model via supervised machine learning. The influence of critical design parameters such as radius of mixing section, nozzle exit point, and mixing duct length on the performance of the ejector is enabled by a sensitivity analysis study facilitated by design space exploration. Finally, a multi-objective evolutionary algorithm is incorporated in the surrogate model to optimize the ejector performance by maximizing entrainment ratio and compression ratio, while minimizing entropy generation. It is found that stagnation temperature ratio is a key influencing parameter responsible for enhancing mixing layer growth to improve the ejector performance. The optimized ejector shows an enhanced efficiency of ∼ 25 % compared to a non-optimized ejector.