Design optimization of ejectors using ANN and GA

Abstract

AbstractEjectors are devices that are based on the principle of momentum transfer. A primary fluid passes through a nozzle that is usually of converging–diverging cross‐section so that the flow reaches supersonic velocity at the exit. Consequently, a low‐pressure region is created just outside the nozzle exit. This pressure gradient draws out the secondary fluid, into the ejector through the annular space—a phenomenon known as entrainment. This paper attempts to design and optimize an ejector with 1,1‐dichloro‐1‐fluoroethane as the working fluid. The governing equations that accurately predict the behavior of the working fluid, are solved using the finite volume method after the discretization of the flow domain, using ANSYS Fluent. A database is created over 1008 similar computational fluid dynamics simulations by recording the input parameter values and the corresponding output parameter values. It is then used to define a function that can precisely predict the output for an unknown set of input parameters. This is achieved through the implementation of artificial neural networks—a surrogate modeling technique. The accuracy of the model is determined from the coefficient of correlation. The objective function thus obtained is optimized with the help of a genetic algorithm (GA)—a nature‐inspired optimization technique. The optimal design of the ejector for a set of operating conditions is obtained as the output of the GA.