Optimization of an ejector to mitigate cavitation phenomena with coupled CFD/BP neural network and particle swarm optimization algorithm

Ejector serves as a component to suck the coolant for taking the adverse heat away from the reactor core during the accident. Owing to the simple construction and high reliability, it is widely applied in the pipeline of the Pressurized Water Reactor (PWR). The potential harm is the hydraulic cavitation induced by the local negative pressure, which might lead to vibration beyond limit and even damage of pipeline. This paper aims to provide the improved designs of the ejector. To achieve this purpose, a parametric geometric model of ejector was established, the shape of the ejector was optimized by particle swarm optimization (PSO) algorithm, and the fluid physical quantities, including flow fluxes and vapor phase volumes, were extracted by computational fluid dynamics (CFD) simulation. The accuracy of CFD simulation was verified by laboratory experiments. Moreover, in order to expedite the optimization process, several back propagation (BP) neural networks were trained to predict the fluid quantities. The results show that the jet diameter has a significant influence on the flow state and the cavitation degree. The optimized structure of ejector leads to lower vapor phase area within the constraint of flow fluxes. This work is expected to provide a framework to reduce the vibration of pipeline motivated by cavitation in ejector.